Frequency Control On An Island Power System

Frequency Control on an Island Power
System
with Evolving Plant Mix
by
Gillian R. Lalor
A thesis presented to
The National University of Ireland
in fullment of the
requirements of the degree of
Philosophiae Doctor
in the
School of Electrical, Electronic and Mechanical Engineering
University College Dublin
September 2005
Supervisor of Research: Professor M.J. OMalley
Nominating Professor: Professor A.M. de Paor
Abstract
Continual balancing of active power generated and consumed is vital for power system
security and stability, and to maintain frequency within an acceptable tolerance around
nominal system frequency. Due to the large size of individual generators with respect to
total system size, the loss of a generator in a small island system can cause a large power
imbalance, and consequently a signicant frequency excursion. Low system inertia
results in high rates of change of frequency when a power imbalance occurs. Therefore,
system frequency control on an isolated power system is particularly challenging.
As the generating mix on a power system evolves, moving away from traditional steam
generating units, the behaviour of the power system in response to a power imbalance
also changes. Both combined cycle gas turbine (CCGT) and wind turbine generators
have distinctive eects on system frequency control. As each technology comprises an
increasing proportion of generation on power systems worldwide, a clear understanding
of the eects of CCGT and wind turbine generator characteristics on system frequency
control is required in order to maintain secure and stable power systems.
A dynamic model of the Ireland electricity system is developed, tuned and validated
for the purpose of studying short-term frequency control on an island system. Each
frequency responsive generating unit on the Ireland system is modelled using low order
models tuned to extensive data from frequency events on the Ireland system. The
system load is modelled using a single measurement based dynamic load model, which
incorporates the frequency sensitivity and inertial contribution of the load during power
imbalances. Frequency control through under-frequency load shedding is also incorpo-
rated in the model. The system model was subsequently validated through comparison
with frequency events not previously used for tuning. The resultant system model has
the ability to predict the under-frequency behaviour of the Ireland power system for
up to 20 seconds following a loss of generation with a very good level of accuracy.
i
The active power generated by a base loaded CCGT is coupled to system frequency. A
model suitable for studying the short-term dynamic response of a combined cycle gas
turbine to a system frequency deviation is developed. The model is tuned and validated
with event data from combined cycle gas turbines on the Ireland electricity system. This
model is used in conjunction with the validated system model to study the impact of
increasing levels of CCGT generation on short-term frequency control of a small island
system during a loss of generation event. Results indicate that as the number and
proportion of base loaded combined cycle gas turbines increases, frequency control
may become more challenging. The magnitude of the system frequency excursion
increases non-linearly as the proportion of base loaded CCGTs increases. Therefore, if
the number of CCGTs increases, large frequency excursions will become more likely and
transmission system operators may need to review their frequency control strategies to
maintain current security standards and to avoid the shedding of customers.
Increased system inertia is intrinsically linked to the addition of synchronous genera-
tion to power systems. However, due to diering electromechanical characteristics, this
inherent link is not present in wind turbine generators. Dynamic models of two dier-
ent wind turbine technologies are integrated into the validated system model, which
is modied to represent the predicted 2010 Ireland electricity system. The eect on
system frequency during a loss of generation event is examined for varying wind pene-
trations on the system. The results indicate that regardless of wind turbine technology,
the displacement of conventional generation with wind will result in increased rates of
change of system frequency. The magnitude of the frequency excursion following a loss
of generation may also increase. Amendment of reserve policies or modication of wind
turbine inertial response characteristics may be necessary to facilitate increased levels
of wind generation, particularly for an isolated power system.
In addition to the short-term dynamic eects of wind generation on frequency control,
longer-term eects as a result of the wind generation characteristics of variability and
unpredictability need to be taken into account in order to maintain adequate levels
of system security in all time frames. However, while a clear understanding of the
technical aspects of frequency control is vital to ensure system security, they comprise
just one part of the whole frequency control issue. The technical aspects therefore need
to be put in perspective by examining them as part of the broader picture, which also
includes economic consequences. With rapidly increasing wind generation on many
power systems, the eect on all aspects of frequency control is being examined in
ii
detail, to assess and quantify the impact of wind integration. A review of a number
of previous wind integration studies is carried out and a preliminary methodology
is proposed for examining the eects of wind integration on all aspects of frequency
control over a number of time-frames. Some illustrative results are given for a sample
AC interconnected system.
iii
Acknowledgements
I would like to thank everybody whose help and support contributed to this thesis. In
particular, there are some without whom this thesis would not have been possible:
Professor Mark OMalley, whose guidance, help, expertise and encouragement through-
out the project was invaluable. Always making time for discussion, his supervision
throughout has been excellent and I am extremely grateful for everything over the last
four years. Thank you.
Dr. Damian Flynn and Julia Ritchie, of the Queens University of Belfast, who collab-
orated on the development of the Ireland electricity system model and with whom I
had many useful and informative discussions.
Dr. Lawrence Jones of Areva T&D, who gave me the opportunity to to spend 3 months
working with Areva T&D in Bellevue, Washington.
Professor Chen Ching Liu of the University of Washington, Seattle, without whom the
trip to Washington would not have been possible.
Colleagues in ESB National Grid, for many useful discussions and interactions, in
particular Jonathan OSullivan, Michael Power, Doireann Barry, Kate OConnor, Pat
McGrath and John Kennedy.
ESB Power Generation, in particular Michael OMahony, Alan Egan and Nicholas
Tarrant, for information and advice during the development of the CCGT model.
Tom Wilson of Viridian, for help and useful discussions during the development of the
CCGT model.
Dr. Alan Mullane, for his guidance and expertise in collaboration on the study into
iv
frequency control and wind turbine technology. Also for all the advice and questions
answered, about LaTex as well as wind, and proof reading this thesis.
Ronan Doherty, with whom I collaborated on the study into frequency control in com-
petitive market dispatch in addition to a number of dierent projects, for the many
useful and informative discussions throughout.
All occupants of Room 157 over the course of the last four years. In particular, Shane
Rourke for his help, advice and the invaluable discussions since I started and Hugh
Mullany for his advice and proof reading this thesis. Also Tim Hurley, Andy Keane,
Eleanor Denny, Garth Bryans and Ciara OConner for the constant moral support, tea
and coee breaks, and a enjoyable working atmosphere.
My family, Liz, Pamela, Richard, John, and in particular Mum and Dad. Thank you
for the constant support and encouragement, not just over the last four years, but in
everything I do.
All my friends, whose friendship I value greatly.
And nally James, for the endless encouragement, support, condence in me and, not
least, patience, which have been invaluable over the last number of years. Thank you
for everything.
v
Publications arising from this thesis
Journal Papers:
R. Doherty, G. Lalor and M. OMalley,Frequency Control in Competitive Electricity
Market Dispatch, IEEE Transactions on Power Systems, August 2005, Vol. 20, No.
3, pp. 1588-1596. (Appendix C)
G. Lalor, J. Ritchie, D. Flynn, and M. OMalley, The Impact of Combined Cycle Gas
Turbine Short Term Dynamics on Frequency Control, IEEE Transactions on Power
Systems, August 2005, Vol. 20, No. 3, pp. 1456-1464. (Appendix B)
G. Lalor, A. Mullane and M. OMalley, Frequency Control and Wind Turbine Tech-
nologies, IEEE Transactions on Power System. In press, 2005. (Appendix D)
Conference Papers:
G. Lalor and M. OMalley, Frequency Control on an Island Power System with In-
creasing Proportions of Combined Cycle Gas Turbines, presented at IEEE Powertech
Conference, Bologna, June 2003. (Appendix E)
G. Lalor, J. Ritchie, S. Rourke, D. Flynn, and M. OMalley, Dynamic Frequency Con-
trol with Increasing Wind Generation, presented at IEEE Power Engineering Society
General Meeting, Denver, Colorado, June 2004. (Appendix F)
vi
Table of Contents
1 Introduction 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Frequency Control of an Island Power System . . . . . . . . . . . . . . 4
1.2.1 Frequency Control . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.2 Island Power Systems . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2.3 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3 The Aims and the Scope of this Thesis . . . . . . . . . . . . . . . . . . 12
2 System Model 14
2.1 The Ireland Power System . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2 Modelling the Ireland Power System. . . . . . . . . . . . . . . . . . . . 16
2.2.1 Assumptions of the Model . . . . . . . . . . . . . . . . . . . . . 17
2.2.2 Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.2.3 Load . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.2.4 Connecting System . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.3 Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.4 Simulation Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
vii
2.5 Tuning the Ireland System Model . . . . . . . . . . . . . . . . . . . . . 35
2.5.1 Generating Units . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.5.2 Load Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.5.3 Connecting System . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.6 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.6.1 Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.6.2 Frequency Control in Competitive Electricity Market Dispatch . 46
2.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3 Frequency Control with Combined Cycle Gas Turbines 53
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.2 CCGT Background and Characteristics . . . . . . . . . . . . . . . . . . 54
3.2.1 Gas Turbine Component . . . . . . . . . . . . . . . . . . . . . . 55
3.2.2 The Heat Recovery Steam Generator and Steam Turbine Com-
ponents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
3.3 Literature review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
3.3.1 CCGT Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . 63
3.4 CCGTs on the Ireland System . . . . . . . . . . . . . . . . . . . . . . . 65
3.5 CCGT Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.5.1 CCGT Model Structure . . . . . . . . . . . . . . . . . . . . . . 66
3.5.2 Model Tuning and Validation . . . . . . . . . . . . . . . . . . . 70
3.6 The Impact of CCGT Dynamics on Frequency Control . . . . . . . . . 75
3.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
viii
4 Frequency control and Wind Turbine Technology 81
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
4.2 Wind Generation Technology . . . . . . . . . . . . . . . . . . . . . . . 82
4.3 Wind Turbine Generator Modelling . . . . . . . . . . . . . . . . . . . . 86
4.3.1 Fixed Speed Wind Turbine Model . . . . . . . . . . . . . . . . . 86
4.3.2 DFIG Wind Turbine Model . . . . . . . . . . . . . . . . . . . . 89
4.4 Wind Generation on the Ireland Electricity System . . . . . . . . . . . 92
4.4.1 Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.4.2 Simulating Procedure . . . . . . . . . . . . . . . . . . . . . . . . 93
4.5.1 Response of wind turbine technologies to system frequency devi-
ations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.5.2 System frequency control with increasing wind penetration . . . 95
4.5.3 Supplementary response from DFIG . . . . . . . . . . . . . . . . 100
4.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102
5 Supplementary Study: Wind Integration Studies and Frequency Con-
trol 105
5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
5.2 Review of Wind Integration Studies . . . . . . . . . . . . . . . . . . . . 107
5.3 Preliminary Wind Integration Frequency Control Study . . . . . . . . . 114
5.3.1 e-terra simulator . . . . . . . . . . . . . . . . . . . . . . . . . . 115
5.3.2 Wind Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
5.3.3 Determination of Wind Variability Costs . . . . . . . . . . . . . 118
5.3.4 Determination of Wind Unpredictability Costs . . . . . . . . . . 121
ix
5.4 Preliminary Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
5.4.1 Available Wind Data . . . . . . . . . . . . . . . . . . . . . . . . 122
5.4.2 Scope of the study . . . . . . . . . . . . . . . . . . . . . . . . . 122
5.4.3 Sample Test System . . . . . . . . . . . . . . . . . . . . . . . . 123
5.4.4 Scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
5.4.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
5.5 Discussion and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . 130
6 Conclusions 134
6.1 Synopsis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
6.2 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
6.3 Scope for future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
References 140
A Frequency Disturbance Event 152
B The Impact of Combined Cycle Gas Turbine Short Term Dynamics
on Frequency Control 155
C Frequency Control in Competitive Electricity Market Dispatch 166
D Frequency Control and Wind Turbine Technologies 176
E Frequency Control on an Island Power System with Increasing Pro-
portions of Combined Cycle Gas Turbines 186
F Dynamic Frequency Control with Increasing Wind Generation 194
x
List of Figures
1.1 Operating reserve time-scales . . . . . . . . . . . . . . . . . . . . . . . 6
1.2 Recorded system frequency on Ireland electricity system . . . . . . . . 9
2.1 Generation mix on the Ireland electricity system for 1995, 2005 and 2010 15
2.2 Steam unit model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.3 Open cycle gas turbine model . . . . . . . . . . . . . . . . . . . . . . . 24
2.4 Linear hydroelectric-turbine model . . . . . . . . . . . . . . . . . . . . 28
2.5 Ireland system model . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.6 Inertial response control loop . . . . . . . . . . . . . . . . . . . . . . . 37
2.7 Six frequency events on the Ireland system with corresponding power
output of a sample generator . . . . . . . . . . . . . . . . . . . . . . . . 38
2.8 Comparison between actual and simulated frequency response of a steam
unit to a low frequency event . . . . . . . . . . . . . . . . . . . . . . . 40
2.9 Turlough Hill generating unit response for two low frequency events . . 42
2.10 Actual and simulated system frequency for a 267 MW generation loss . 46
2.14 Simplied system frequency model . . . . . . . . . . . . . . . . . . . . 50
xi
2.15 Comparison of the generation response of the black box model with the
validated system model . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
3.1 Single-shaft CCGT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
3.2 Multi-shaft CCGT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.3 CCGT model structure . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
3.4 CCGT ambient temperature dependency . . . . . . . . . . . . . . . . . 71
3.5 CCGT ambient pressure dependency . . . . . . . . . . . . . . . . . . . 72
3.6 Change in power output of a typical near base loaded CCGT in response
to a frequency event on the system . . . . . . . . . . . . . . . . . . . . 73
3.7 Simulated power output of the GT component of a typical CCGT to a
frequency drop of 0.5 Hz . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.8 Winter peak scenario with 422 MW trip . . . . . . . . . . . . . . . . . 77
3.9 Summer night valley scenario with 400 MW trip . . . . . . . . . . . . . 78
3.10 Summer day valley scenario with 400 MW trip . . . . . . . . . . . . . . 79
3.11 Sensitivity of system frequency nadir to increasing proportions of CCGTs 80
4.1 Typical C
p
curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.2 Fixed speed wind turbine Generator . . . . . . . . . . . . . . . . . . . . 84
4.3 Doubly fed induction generator . . . . . . . . . . . . . . . . . . . . . . 85
4.4 DFIG model with FOC controller . . . . . . . . . . . . . . . . . . . . . 91
4.5 Comparison of xed speed WTG and DFIG WTG responses to the low
frequency event . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.6 Eect of increasing wind penetration on maximum rate of change of
frequency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
4.7 Simulated system frequency following the trip of largest infeed during
the SDV scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
xii
4.8 Frequency nadir and static reserve tripped following the loss of the
largest infeed for increasing wind penetration during the Summer Day
Valley scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.9 Frequency nadir and static reserve tripped following the loss of the
largest infeed for increasing wind penetration during the Summer Night
Valley scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.10 Supplementary control loop for DFIG WTG controller. . . . . . . . . . 102
4.11 Comparison of xed speed WTG and DFIG WTG responses to the low
frequency event, including supplementary control loop . . . . . . . . . . 103
4.12 Simulated system frequency following the trip of largest infeed during
the SDV scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
5.1 Test system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
5.2 Wind farm power output time series . . . . . . . . . . . . . . . . . . . 126
5.3 System frequency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
5.4 ACE: Control Area A . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
5.5 ACE: Control Area B . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
5.6 ACE: Control Area C . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
5.7 Generator power output: Control Area A . . . . . . . . . . . . . . . . . 131
5.8 Generator power output: Control Area B . . . . . . . . . . . . . . . . . 132
5.9 Generator power output: Control Area C . . . . . . . . . . . . . . . . . 133
A.1 Recorded system frequency on Ireland electricity system . . . . . . . . 152
xiii
List of Tables
2.1 Under-frequency setting for Turlough Hill operating modes . . . . . . . 28
4.1 Comparison of inertial response from various generators . . . . . . . . . 95
4.2 Maximum ROCOF following loss of largest infeed (422MW) for various
operating scenarios, wind turbine penetrations and wind turbine tech-
nology type. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
5.1 Test system generation capacity . . . . . . . . . . . . . . . . . . . . . . 123
5.2 Test system set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
5.3 Test case scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
xiv
Nomenclature
a = Frequency sensitivity of GT exhaust gas ow calculation factor
A = Area swept by wind-turbine rotor (m
2
)
b = Constant, such that a+b=1
B = Frequency bias setting (MW/0.1Hz)
C
p
= Performance coecient
C
D
= Boiler drum integral coecient (s)
C
SH
= Boiler superheater integral coecient (s)
f = System frequency (Hz)
F
A
= Actual control area frequency (Hz)
f
gen
= Under frequency relay setting for Turlough Hill gen mode (Hz)
F
HP
= Fraction of power output from high pressure ST stage
f
int
= Under frequency relay setting for interruptible customers (Hz)
F
IP
F
LP
f
min
= Under frequency relay setting for Turlough Hill min gen mode (Hz)
f
o
= Nominal system frequency (Hz)
f
pump
= Under frequency relay setting for Turlough Hill pump mode (Hz)
F
S
= Scheduled control area frequency (Hz)
f
spin
= Under frequency relay setting for Turlough Hill spin mode (Hz)
f
UFLS
= Under frequency load shedding relay setting (Hz)
G = Set of generators
G
w
= Gate position (per unit)
i = Current (A)
igv = Inlet guide vane angle (
)
I
j
= Inertia of generating unit j (kgm
2
)
INT = Logical operator for inertial control loop
xv
J = Polar moment of inertia of wind turbine and rotor (kgm
2
)
K = Friction drop coecient of orice between drum and superheater
K
1P
= Proportional gain for d axis current controller
K
2P
= Proportional gain for q axis current controller
K
1I
= Integral gain for d axis current controller
K
2I
= Integral gain for q axis current controller
K
BB1
= Black box model parameter 1
K
BB2
= Black box model parameter 2
K
i
= IGV controller constant
k
pf
= Steady state frequency sensitivity of the load
k
pv
= Active power and voltage load model parameter
k
qf
= Reactive power and frequency load model parameter
k
qv
= Reactive power and voltage load model parameter
K
scl
= Supplementary control loop constant
KE = Kinetic energy (MWs)
KE
i
= Kinetic energy of generator i (MWs)
KE
L
= Kinetic energy of system load (MWs)
KE
o
= Kinetic energy at nominal frequency f
o
(MWs)
L
igv
= Inlet guide vane position (per unit)
L
m
= Per phase mutual inductance (H)
L
r
= Per phase rotor inductance (H)
L
s
= Per phase stator inductance (H)
L
= L
2
m
L
r
L
s
(H)
m = Steam ow rate out of boiler drum (per unit)
m
s
= Steam ow rate into steam turbine (per unit)
m
w
= Steam ow rate into boiler drum (per unit)
N = System speed (per unit)
N
G
= Number of generators
N
ref
= Reference system speed (per unit)
NI
A
= Algebraic sum of the actual ows on all tie lines/interconnectors (MW)
NI
S
= Algebraic sum of the scheduled ows on all tie lines/interconnectors (MW)
p = Dierential operator
P = Active power (MW)
P
a
= Ambient pressure (mbar)
P
aero
= Accelerating aerodynamic power (MW)
xvi
P
D
= Boiler drum pressure (per unit)
P
elec
= Electrical power (MW)
P
f
= Number of machine poles
P
GEN
= Active power generated (MW)
P
k
= Amount of generation lost (MW)
P
load
= Active power required by the load (MW)
P
max
= Maximum rated generator power output (MW)
P
mech
= Mechanical power (MW)
P
min
= Minimum rated generator power output (MW)
P
o
= Steady state system demand (MW)
P
pu
= Active power (per unit)
P
T
= Boiler throttle pressure (per unit)
Q = Heat Energy (per unit)
Q
load
= Reactive power required by the load (MVAR)
R = Resistance ()
R
d
= Droop (%)
R
r
= Radius of rotor (m)
R
p
= Primary reserve available at 5 seconds (MW)
SP = Generator operating set-point (per unit)
T
1
= Load time constant (T)
T
a
= Ambient temperature (
C)
T
cd
= Compressor discharge time constant (s)
T
CH
= Steam transport and conversion time constant (s)
T
CO
= Steam turbine crossover and conversion time constant (s)
T
em
= Electromagnetic torque (N m)
T
emref
= Reference electromagnetic torque (N m)
T
gf
= Gas fuel system time constant (s)
T
i
= IGV controller integration rate (s)
T
igv
= IGV actuator time constant (s)
T
pf
= Ratio of load inertia to system frequency
T
pv
= Active power and voltage load model parameter
T
qf
= Reactive power and frequency load model parameter
T
pf
= Reactive power and voltage load model parameter
T
r
= Gas turbine rated exhaust gas temperature (
C)
T
ref
= Reference torque (N m)
xvii
T
RH
= Steam turbine reheater and conversion time constant (s)
T
s
= Droop governor time constant (s)
T
sc
= Supplementary control loop torque (N m)
T
t
= Temperature controller integration rate (s)
T
v
= Valve positioner time constant (s)
T
w
= Water time constant (s)
T
x
= Gas turbine exhaust gas temperature (
C)
T
xc
= Gas turbine corrected exhaust gas temperature (
C)
T
xm
= Gas turbine measured exhaust gas temperature (
C)
TD
CR
= Combustion reaction time delay (s)
TD
TE
= Exhaust gas transport delay (s)
u = wind speed (m/s)
V ,v = Voltage (V)
W
f
= Gas turbine fuel ow (per unit)
W
x
= Gas turbine exhaust gas ow (per unit)
= Blade pitch angle (rad)
P
int,j
= Inertial response of generating unit j (MW)
= Tip-speed ratio
= Flux linkage (Wb)
o
= Average slip of an average induction machine
= Supplementary control loop time constant (s)
i
= Inertial control loop time constant (s)
= Density of air (kg/m
3
)
= system speed (rad/s)
dq
= dq reference frame angular velocity (rad/s)
m
= Rotor mechanical angular velocity (rad/s)
s
= Shaft speed (rad/s)
t
= Wind-turbine rotor speed (rads
1
)
r
= Rotor electrical angular velocity (rad/s)
Subscripts
d, q = Direct, Quadrature axis component
I, P = Integral, Proportional
r, s = Rotor, Stator
xviii
Acronyms
AC Alternating Current
ACE Area Control Error
AGC Automatic Generation Control
BPA Bonneville Power Administration
DC Direct Current
CCGT Combined Cycle Gas Turbine
CER Commission for Energy Regulation
CHP Combined Heat and Power
CLP China Light & Power Co.
CO
2
Carbon Dioxide
DFIG Doubly Fed Induction Generator
EMS Energy Management System
ESB Electricity Supply Board
ESBNG ESB National Grid
FOC Field Orientated Controller
GT Gas Turbine
HEC Hong Kong Electric Co.
HRSG Heat Recovery Steam Generator
HVDC High Voltage Direct Current
IEC Israel Electric Corporation
IGV Inlet Guide Vane
NIE Northern Ireland Electricity
NO
x
Oxides of Nitrogen
OCGT Open Cycle Gas Turbine
POR Primary Operating Reserve
ROCOF Rate Of Change Of Frequency
SCADA Supervisory Control and Data Acquisition
SCIG Squirrel Cage Induction Generator
xix
SO System Operator
SONI System Operator of Northern Ireland
ST Steam Turbine
TH Turlough Hill
TNB Tenaga Nasional Berhad
UFLS Under Frequency Load Shedding
WTG Wind Turbine Generator
xx
Chapter 1
Introduction
1.1 Background
The function of a power system is to provide customers with an electricity supply of
acceptable reliability, where reliability signies the ability to supply adequate electric
service on a nearly continuous basis with few interruptions over an extended period of
time (IEEE/CIGRE, 2004). Therefore, to design and operate a power system within
adequate reliability margins such that overall costs are minimised is a key objective for
all system operators.
Power system security is an indication of the level of robustness of the power system
at any instant in time to a disturbance (Fink and Carlsen, 1978). When operating
in a secure state, a power system can withstand most severe disturbances without
interruption to customer supply. However, if operating in a state with reduced security
margins, a power system will be more susceptible to disturbance, resulting in a higher
likelihood of customer supply disruption. To maintain adequate reliability it is desirable
to maximise the time the power system is operating in a secure state, with frequency
and voltage levels within acceptable standards. In order for a power system to be
secure, the power system must be operating in a stable state. Stability indicates
the ability of the system to return to an equilibrium operating state subsequent to a
disturbance, and is dependent on both the type of disturbance and the initial power
system operating conditions (IEEE/CIGRE, 2004). Although the electricity industry
is undergoing regulatory and organisational changes, the basic concepts and rules for
1
Chapter 1. Introduction 2
reliable, secure and stable system operation remain unchanged.
Ancillary services can be broadly dened as the range of technical services required
by the system operators to maintain both secure and stable operation of the power
system. These include operating reserves for frequency control, voltage control and also
system restoration/black start capability. While the methods by which these services
are procured may vary and evolve with regulatory structure, the necessity of ancillary
services is unquestionable. This is highlighted by a number of recent contingencies
worldwide resulting in severe lapses in the security of power systems including blackouts
in the Eastern US and Canada, Italy and the UK (NERC, 2004; UCTE, 2003; NGC,
2003).
The control of system frequency is a vital aspect of secure and stable power system
operation. A continuous balance between active power generated and active power
consumed by the load and losses is required to maintain frequency constant at nominal
system frequency. Any imbalance in active power will result in a frequency devia-
tion. While precise instantaneous balancing of active power is not viable, frequency
control ensures that the system frequency remains within acceptable frequency limits.
Frequency control can be called upon for a variety of conditions ranging from a grad-
ual change in load levels over time to a sudden loss of generation or step increase in
demand.
A range of power system characteristics including system size, individual generator and
load frequency response characteristics and plant mix on the system inuence frequency
control. The size and speed of a frequency deviation depends on the magnitude of the
power imbalance and the power system size. Power system inertia is the resistance
of the individual rotating masses of the generator and load components synchronised
to the system to a change in system speed. The greater the inertia of the system,
the slower the rate of change of frequency in the event of a power imbalance of given
magnitude. Large interconnected power systems have high system inertia, due to the
large number of components synchronised to the system. In addition, the size of in-
dividual components, such as generators, tends to be small in comparison with total
system size. As such, large frequency excursions from nominal are uncommon, and the
rate of change of frequency is relatively slow due to high inertia. Small isolated power
systems, in contrast, have much lower system inertia. Combined with the fact that a
single generator can comprise a sizeable proportion of total generation, large power im-
balances relative to the system size are more frequent and frequency changes are faster.
Adequate frequency control on such a system is vital to prevent the excursion of system
frequency beyond limits where interruption to customer supply through load tripping
starts to occur. Therefore, maintaining system frequency at nominal frequency for a
small island power system with limited interconnection can be technically challenging.
Plant mix is continually evolving, for all power systems, large and small. Knowledge
of the impact that evolving plant mix will have on system frequency control is vital
to maintain a secure and stable power system with adequate reliability standards.
Traditionally large coal and oil fuelled thermal plant comprised the majority of the
generation mix on many power systems. However, due to economic and environmental
driving forces, increasing proportions of combined cycle gas turbines and open cycle
gas turbines are now being used to meet increasing demand and to replace older coal
and oil-red plants as they are retired.
Combined cycle gas turbines (CCGTs) oer higher eciency, greater exibility and
lower emissions than many conventional thermal generators, in addition to progressively
shorter installation times and reducing installation costs. As a result, CCGT generating
units comprise an ever increasing proportion of generation capacity for many electricity
systems. The eciency of combined cycle gas turbines is maximised when operating at
or near maximum or base load, and declines with decreased loading. The behaviour of
CCGT generators in response to frequency excursions diers from that of a conventional
steam turbine, and may have a detrimental eect on the system frequency response
when the CCGT is run at, or near, base load. This eect will be progressively more
apparent as CCGTs operating at or near base load comprise increasing proportions of
the generation.
In conjunction with the shift towards CCGT plant, many power systems worldwide are
also experiencing a rapid increase in wind generation. This trend is driven by a variety
of reasons including environmental concerns, targets for electricity production from
renewable energy resources, the desire for increased fuel diversity, constant advances
in technology and economic factors including declining costs. While the addition of
conventional synchronous generators to a power system will result in an inherent in-
crease in the system inertial response, this is not necessarily the case with wind turbine
generators. Therefore, if rapidly increasing levels of wind generation begin to displace
conventional synchronous generation, erosion of system inertial response may result.
This eect will result in increasing rates of change of frequency during power imbal-
ances, and the magnitude of frequency excursions may also rise. These eects will
inuence small isolated power systems, in particular, where system inertia levels are
inherently low.
1.2 Frequency Control of an Island Power System
1.2.1 Frequency Control
System frequency provides a instantaneous indication of system operating conditions,
as any imbalance between active power generated and consumed manifests itself as a
deviation from nominal system frequency. The magnitude of the frequency excursion
and the rate of change of frequency are dependent on a number of factors, including
the size of the power imbalance and the characteristics of the power system. While
small variations in system frequency will not result in a reduction is system reliability
or security, large frequency deviations can have a serious impact on power system
components and power quality is degraded. Damage to generators and transformers
can result from overheating due to increases in the volts/hertz ratio during times of
low frequency. In addition, generator damage due to mechanical vibrations can occur if
frequency deviations greater than 5% of nominal frequency occur (Kirby et al., 2002).
As a result, most power system components are equipped with protective relays, which
are triggered if system frequency reaches critical conditions. Therefore, control of
system frequency is vital for the secure, reliable operation of the power system.
The objective of frequency control is to maintain adequate balance between active
power consumed and generated on a power system such that frequency remains within
acceptable limits around nominal frequency. As the demand of a power system is con-
stantly changing, frequency control is continuously called upon to full this objective.
To a large extent, the changing system load is predictable and generators are committed
and dispatched based on the forecast load levels (Machowski et al., 1997). Therefore,
under normal operating conditions, the balancing of energy is achieved by adjusting
generator active power set-points. Signals to generators for such adjustments are ei-
ther issued by the system operator or automatically generated and issued by automatic
generation control (AGC).
In the event of an unpredicted increase in system load or an unexpected loss of genera-
tion or transmission line, an imbalance of active power will occur. Every power system
has stored kinetic energy by virtue of the masses of the generator and load compo-
nents rotating in synchronism, which is a function of both the system inertia and the
system frequency. In response to a power imbalance, stored kinetic energy is released
to redress the imbalance, resulting in an inherent reduction in system frequency. In
the event of active power generated exceeding demand, kinetic energy is absorbed and
an increase in system frequency results. However, while frequency control in the event
of high frequency events is essential and many issues discussed here are relevant, low
frequency events are the focus of this thesis.
In order to limit the frequency excursion from nominal system frequency, and to main-
tain a stable and secure system, action in addition to the inherent system inertial
response is required, i.e. frequency control. Frequency control may be broadly cat-
egorised into automatic and manual frequency control. The former responds auto-
matically to either a deviation from nominal system frequency or a rate of change of
frequency in excess of a predened threshold. Sources of automatic frequency control
are the natural reduction in system load with low frequency, the automatic increase in
generator active power output activated by the speed droop governor, low frequency
or rate of change of frequency triggered responses from pumped storage units and the
automatic shedding of load. Manual frequency control encompasses all instructions
issued by the system operator to generators (and load if applicable, i.e. in the event
of load participation) for changes from the reference set point of the generator (or to
current active power consumption in the case of load).
Additional active power capacity available (i.e. when compared to steady state oper-
ation prior to a frequency event) from generation units or through reduction in load
for the purpose of frequency control is known as operating reserve (ESBNG, 2005b).
Many dierent denitions for the categorisation of operating reserve exist. In this the-
sis, reserve is categorised into primary, secondary and tertiary operating reserve, as
dened by ESBNG (2005b), and illustrated in Fig. 1.1.
Primary operating reserve (POR) is the additional active power available from genera-
tors and through reduction of active power consumption of the load which is available
between 5 and 15 seconds subsequent to an event on the system. Secondary reserve is
dened to be the additional active power available and sustainable for the time period
Figure 1.1: Operating reserve time-scales (SEI, 2004)
from 15 to 90 seconds after the event. Tertiary reserve is the additional active power
available from 90 seconds to 20 minutes subsequent to the event. Finally replacement
reserve is the additional active power available from 20 minutes to 4 hours after the
event.
In the event of a power imbalance, POR automatically responds to arrest the falling
frequency and initiate recovery towards nominal frequency through the reduction of
the power imbalance. The predominant source of POR on the majority of systems is
the automatic droop governor response of generators operating below maximum rated
active power output to a deviation in speed. Other sources of POR from generators
can include an increase in active power when under-frequency relays or rate of change
of frequency (ROCOF) relays are triggered. One example is under-frequency relaying
triggering a rapid increase in active power generation from a pumped storage generating
unit.
System load also contributes to POR. In addition to the natural load reduction due to
low system frequency, system load can also provide static reserve in the form of either
interruptible customers or under-frequency load shedding (UFLS). Static reserve is
dened here as capacity available instantaneously when called upon, with negligible
dynamics.
Some customers (interruptible customers) are contracted to make their load available
for short term interruptions. Specic blocks of load are congured to be tripped by
under-frequency relays if frequency falls to a threshold level. UFLS, however, is the
tripping of uncontracted load at distribution system level and is called upon to pre-
vent system collapse only when other sources of POR fail to arrest falling frequency.
Discrete blocks of load are tripped until generation and load are once again in balance
(Machowski et al., 1997), and frequency decline is arrested.
Once the system frequency has been arrested and stabilised by the POR, it is the task of
secondary and tertiary reserve to restore the system frequency to nominal value. This is
achieved through a combination of automatic droop governor response while frequency
remains below nominal and through discrete instructions issued by the system operator
to generators for changes from the reference set point of the generator until power
is once again balanced, and frequency restored to nominal. Replacement reserve is
employed to replace operating reserve, restoring the system to a secure operating state.
Capacity to provide operating reserve is dispatched in conjunction with generation by
the system operator to ensure availability of adequate operating reserves in the event
of a power imbalance. For a secure system, adequate operating reserve is required
so that the power system can withstand most severe frequency disturbances without
interruption to customer supply. The majority of power system worldwide operate with
an N-1 security criterion (Bialek, 2003). This criterion states that the power system
should operate so instability or load shedding do not occur as a result of the most
severe single contingency. From a frequency control perspective, this entails having
sucient reserve to withstand the loss of the large power infeeds to the system.
1.2.2 Island Power Systems
Worldwide, power systems have a considerable range of characteristics including size,
both geographical and electrical, the extent of interconnection to other power systems
and generation mix. Many formerly isolated power systems with varying characteristics
have become part of larger synchronous power systems through the use of alternating
current (AC) interconnection. AC interconnection between power systems yields multi-
ple advantages, increased system inertia, trading of energy, sharing of spinning reserve
provision (operating reserve available from online generators) and mutual support dur-
ing contingencies to name just a few (Mak and Law, 1991). While direct current (DC)
interconnection allows energy exchange, power systems linked by DC interconnection
are not synchronous. Therefore, some advantages of AC interconnection such as in-
creased system inertia and the sharing of spinning reserves do not inherently occur
with DC interconnection. However, although frequency control is not inherent, DC
interconnections may be designed to provide spinning reserve.
In large interconnected power systems, the size of individual components such as gener-
ators tends to be small in comparison with the magnitude of the entire system. Power
imbalances due to the loss of a single component in such systems, when they occur, are
therefore generally small with respect to the total system size. In addition, the rate at
which the frequency changes tends to be low due to high system inertia. The construc-
tion of sizeable, more economically viable generating units is possible with minimal
risk to system security. In addition, the provision of operating reserve is shared over a
great number of generators, and may be shared between dierent systems within the
larger interconnected power system. Generally, geographical dispersion is a character-
istic of large interconnected power systems, and can contribute to a reduced capacity
requirement, as a result of load diversity. One example of the benets of geographical
dispersion is the staggered occurrence of peak demand when a power system spans
dierent time zones.
While large strongly interconnected systems comprise a large proportion of power sys-
tems worldwide, there are nonetheless a sizeable number of small isolated or poorly
interconnected systems, for example Israel, New Zealand, Crete, Cyprus and Ireland.
Small power systems that are either isolated or with only DC interconnection have low
system inertia. A power system with low system inertia is more sensitive to system
disturbances, due to less stored energy available to redress energy imbalances and to
slow the rate of change of frequency. In addition, for such power systems, system com-
ponents such as generators tend to be large in comparison with the total system size.
In particular at times of low load, a single generator can comprise a large proportion
of the total system generation. Therefore, in the event of a loss of generation, there is
a greater likelihood of a large frequency excursion as the power imbalance is large with
respect to total system size. On the Ireland electricity system, for example, frequency
deviations of 1% are not uncommon, while larger frequency excursions occur occasion-
ally, as illustrated by the recorded system frequency in Fig. 1.2. (This frequency event
is described in more detail in Appendix A.)
As a consequence of both low system inertia and potential large power imbalances
0 100 200 300 400 500 600 700 800
48.4
48.6
48.8
49
49.2
49.4
49.6
49.8
50
Time (s)
F
r
e
q
u
e
n
c
y

(
H
z
)
Figure 1.2: Recorded system frequency on Ireland electricity system
in addition to the relatively small range of generators to provide operating reserve,
frequency control on small, isolated or DC interconnected systems is particularly chal-
lenging. Distinctive operating and control strategies are necessary to maintain the
system within limits of reliability and security. The main dynamic operation problems
in small power systems relate to frequency control, in particular the behaviour of the
system in response to large disturbances (Kottick and Or, 1996). Frequency control
on such systems can in fact cause technical problems an order of magnitude greater
than those experienced on large interconnected systems (OSullivan et al., 1999). The
importance of frequency control on island systems during a contingency is evident in
the considerable volume of relevant literature.
1.2.3 Literature Review
The benets of AC interconnection between power systems are highlighted in Mak and
Law (1991), where the AC interconnection between The China Light & Power Co.
(CLP) and Hong Kong Electric Co. (HEC) are examined. The evolution of CLP from
an isolated system to one with AC interconnection to other systems was found to have
benecial eects on system performance during system contingencies and also resulted
in a more economical system operation. However, AC interconnection is not always an
option and therefore a clear understanding of the frequency control dynamics of island
power system is necessary to ensure optimal system security and reliability.
Two neural network models that predict the frequency nadir (minimum frequency)
and calculate the amount of UFLS during a loss of generation on the Israel Electric
Corporation (IEC) system are developed in Kottick and Or (1996). The IEC operates
an island system and at the time of the study the installed capacity was approximately
5050 MW, with a 550 MW unit as the largest infeed. As the transmission system is
strongly connected, frequency throughout the system is uniform. Therefore, transmis-
sion eects could be neglected and a single busbar model was employed. Both the
magnitude of the frequency excursion and the extent of UFLS are indicators of the
severity of the contingency, and comprise two components of the dynamic security as-
sessment for the IEC system. The models developed were demonstrated to perform
well in assessing the UFLS subsequent to a loss of generation. The potential eect on
frequency regulation (which is the automatic power balancing on a second to second
basis) of a 25 MWh capacity battery energy storage device on the IEC system was
investigated in Kottick et al. (1993), where a single busbar model was again used to
represent the power system. It was demonstrated that simulated frequency deviations
resulting from sudden demand variations were reduced considerably through the ad-
dition of the battery energy storage device, which was assumed capable of sustaining
a power output of 30 MW for 15 minutes. Due to a fast response time, the battery
energy storage device was found to be potentially useful for regulation and as rapid
operating reserve on an island system, where the rapid response time is critical due to
low system inertia.
The application of UFLS to the isolated power system of Cyprus is considered in Con-
cordia et al. (1995). At the time of the study, the Cyprus system (with a peak load of
500 MW) had a largest infeed of 60 MW, which was 12% of peak load and comprised
a signicantly greater proportion of load at times of low demand. While the general
principles of UFLS are independent of system size, the distinguishing characteristics
of isolated system must be taken into account when devising the UFLS plan. A well
devised UFLS schedule results in the system surviving situations that would have oth-
erwise resulted in blackouts. In Concordia et al. (1995), a criteria deemed appropriate
for UFLS on an isolated system was developed and applied to the Cyprus system. It
was also found that the eectiveness of load shedding increases with increasing system
load.
The eects of increasing proportions of renewable generation resources on the system
frequency control on the island of Crete have been the focus of several studies (Hatziar-
gyriou et al., 2000, 2002; Papazoglou and Gigandidou, 2003). In particular, Crete has
experienced a rapid growth in wind generation in recent years. These studies pre-
dominantly focus on the system under non-contingency operating conditions over the
economic dispatch and unit commitment time frames. The short-term dynamic eects
of wind turbine generators on the system frequency during a frequency event are not,
however, considered.
Another example of an island electricity system is the Ireland power system, which con-
sists of two synchronous power systems. Before interconnection between the Northern
Ireland Electricity (NIE) system and the Electricity Supply Board (ESB) system of the
Republic of Ireland, each system on the island of Ireland operated as a small isolated
system. With peak loads of approximately 1650 MW and 3300 MW respectively before
interconnection, each system had low system inertia and as a result emergency control,
i.e. frequency control in the event of a contingency, was critical.
The strategies of the NIE system for emergency control of frequency when operating
as an isolated system are outlined in Fox and McCartney (1988). Several obstacles
such as inaccurate unit response information and diculties with the coordination of
under-frequency relay settings with system dynamics are also discussed. Limited UFLS
was tolerated as a likely necessity on the NIE system at the time of the study in the
event of the loss of a major infeed. Further studies into the control and proper design
of UFLS arrangements are carried out in Fox et al. (1989) and Thompson and Fox
(1994). The use of rate of change of frequency as an activating signal for UFLS was
used in Fox et al. (1989), and found to provide more accurate load shedding than
the use of under frequency relays. System frequency was simulated using a single
busbar model. This approach was expanded in Thompson and Fox (1994), where each
UFLS relay uses system demand, spinning reserve, system inertia and the amount of
low priority load available for shedding elsewhere in conjunction with the local rate of
change of frequency to assess whether to operate. This approach resulted in a signicant
reduction in the amount of excessive UFLS when compared to the xed rate of change
of frequency scheme of Fox et al. (1989). The eect of ywheel energy injection on
emergency control of frequency on the NIE system has also been studied (Hampton
et al., 1991). Once again a single busbar system model to predict system frequency
following a unscheduled generation outage was used, and the model was validated by
comparison to actual power system measurements. It was found that the use of the
ywheel energy storage and retrieval scheme can contribute to considerable savings
through spinning reserve replacement if correctly designed and scheduled.
An emergency reserve model of the ESB system was developed and implemented to
study frequency control on an island power system in OSullivan and OMalley (1996),
OSullivan (1996) and OSullivan et al. (1999). The single busbar model was tuned
using actual frequency events on the ESB system to accurately account for the dy-
namic system characteristics following a loss of generation. The provision of frequency
control was shown to be a critical issue as electricity markets emerge for island systems
(OSullivan et al., 1999). As a result, the above model was subsequently incorporated
into a new methodology for the provision of reserve in a competitive market (OSullivan
and OMalley, 1999).
1.3 The Aims and the Scope of this Thesis
The objective of this thesis is to examine frequency control on an island system with
evolving plant mix. In particular, the inuence of the characteristics of CCGTs and
wind turbine generators on system frequency control will be examined, and the Ireland
electricity system is used as an illustration.
Simulation, using validated models, is a good rst step in understanding frequency
control in the context of evolving plant mix. A single busbar model of the Ireland
electricity system is developed, tuned and validated in Chapter 2. This model is suitable
for the study of frequency response behaviour of an island system for up to 20 seconds
after a power imbalance occurs. This system model is subsequently employed to tune a
black-box model, which is used as the basis for the derivation of a minimum frequency
control constraint (Appendix C).
A model suitable for studying the short-term dynamic response of a combined cycle gas
turbine to a system frequency deviation is developed, tuned and validated in Chapter
3. This model is then used in conjunction with the Ireland system model of Chapter 2
to study the impact of increasing levels of CCGT generation on frequency control of a
small island system (Appendix B).
Models for two dierent wind turbine technologies are presented in Chapter 4. To
examine the short-term dynamic response of an island power system to sudden power
imbalances with increasing proportions of wind generation, these models are integrated
into the Ireland system model of Chapter 2, which is modied to represent the proposed
2010 system model (Appendix D).
The impact of wind generation on both short-term and long-term frequency control
are assessed in Chapter 5. A review of a number of wind integration studies is carried
out. Consequently, a preliminary methodology for a wind integration frequency control
study using the Areva T&D e-terra simulator is proposed, which is applicable to both
island and interconnected power systems. This work was carried out during a three
month industry placement with Areva T&D in Bellevue, Washington.
Chapter 2
System Model
2.1 The Ireland Power System
The electricity system on the island of Ireland operates at 50 Hz, with a current peak
load of approximately 6100 MW (ESBNG, 2004b; SONI, 2003b). The Ireland electricity
system consists of two power systems: the NIE system, operated by System Operator
for Northern Ireland (SONI) and the ESB system, operated by ESB National Grid
(ESBNG). Prior to 1995, the ESB and NIE power systems operated in isolation, with
the limited connection between the two systems generally out of service and, as a
consequence, unreliable (OSullivan, 1996). In 1995, however, the two systems were
reconnected, and now comprise a single synchronous system, connected to each other
through a number of AC lines. The main connection between the NIE and ESB systems
consists of two 275 kV circuits, each with a capacity of 600 MW and of length 50 km.
There are also two additional 110 kV lines, with capacity of 120 MW, connecting
the systems at two separate locations along the interface between Northern Ireland
and the Republic of Ireland (ESBNG, 2004a). A single high voltage direct current
(HVDC) interconnection is in operation between Northern Ireland and Scotland with
a capacity of 500 MW (ESBNG, 2004b). However, this HVDC interconnection is not
currently congured to provide frequency response in the short time frame. With no
AC interconnection to other systems to increase the inertia of the system and share
reserve provision requirements, the Ireland electricity system is essentially an isolated
system.
Chapter 2. System Model 15
The generating capacity of the Ireland electricity system consists of a combination of
reheat and non reheat fossil fuelled steam turbine generators, open cycle gas turbines
(OCGTs), combined cycle gas turbines (CCGTs), hydroelectric generators, a single
pumped storage station and wind turbine generators. In addition, other resources such
as biomass generators, combined heat and power (CHP) and other renewables also
provide limited generating capacity. Generation mix is constantly evolving, with both
CCGTs and wind turbine generators in particular comprising increasing proportions of
generation on the system. A comparison of the generation mix on the Ireland system
in 2005 with the ESB and NIE systems in 1995 is illustrated in Fig. 2.1.
Figure 2.1: Generation mix on the Ireland electricity system for 1995, 2005 and 2010
((SONI, 2003b; ESBNG, 2004b))
The reduction in the proportion of steam units from 1995 to 2005, alongside the increase
in proportions of both CCGT and wind generation is clearly illustrated. The predicted
Ireland generation mix in 2010 is also included in Fig. 2.1, to illustrate the forecast
changing proportions of generation on the system.
The system operator (SO) of each system performs scheduling and dispatch indepen-
dently, while incorporating contracted ows on the interconnections between the two
systems. The provision of primary operating reserve, however, is shared between the
two systems.
In accordance with the system grid codes (ESBNG, 2005b; SONI, 2003a) frequency
regulation is provided by each generator on the system, by virtue of a droop governor,
with a compulsory droop setting of 4%. Operating reserve is divided into several
categories according to the timescale within which it is available in response to an event,
as described in Chapter 1. On the Ireland system, the primary operating reserve (POR)
requirement corresponds to 75% of the largest infeed onto the system. At present, the
largest infeed is 422 MW (the 500 MW HVDC interconnection to Scotland is operated
with a maximum limit of 400 MW for system security reasons i.e. to limit the size
of the largest infeed), thus making the primary reserve requirement 317 MW. The
availability of primary reserve to meet this requirement is divided such that the ESB
and NIE systems provide 67% (211 MW) and 33% (105 MW) respectively. Sources of
POR include spinning reserve from generating units online and static reserve, such as
interruptible load. Static reserve consists of blocks of reserve that are available almost
instantaneously when tripped by the system frequency falling below the predetermined
frequency setting of each block. Interruptible load is a form of static reserve, whereby
certain load on the power system has an agreement with the SO that some or all of
the load may be tripped during certain hours when the frequency falls below 49.3 Hz.
The proportion of the POR provided by spinning and static reserve sources varies with
time of day. The contribution of the pumped storage station to POR depends on the
operational mode in which it is running, as described later in Section 2.2.2.
2.2 Modelling the Ireland Power System.
The dynamic model of the Ireland system used in this thesis is based on two previous
models (OSullivan, 1996; Fox et al., 1989), with considerable enhancements introduced.
An emergency reserve model of the ESB electricity system circa 1995 was developed
by OSullivan (1996). The objective of this single busbar model was to accurately
predict the system frequency following a contingency, by simulating the primary reserve
response of the system. Installed generators are represented using low order models,
and include dynamics for prime mover, turbine and governor valve characteristics where
appropriate. A low order model, derived based on consideration of resistive and motor
loads, is used to represent the system load. A similar single busbar model of the isolated
NIE system circa 1989 was applied in Fox et al. (1989) for the evaluation of emergency
load shedding schemes.
The present Ireland electricity system has evolved and developed from the ESB and NIE
systems represented in OSullivan (1996) and Fox et al. (1989) respectively. A sizeable
growth in system load has occured and new generating plants have been introduced onto
the system and older plants decommissioned and removed. In particular, proportions
of both combined cycle gas turbines and wind generation have increased signicantly,
and are predicted to comprise increasingly large proportions of system generation mix
in the future, as illustrated in Fig. 2.1.
The dynamic model representative of the Ireland electricity system is developed based
on OSullivan (1996), and also Fox et al. (1989), augmented with more detailed and
additional models where necessary. Each generating unit on the Ireland system is
individually modelled, with the exception of wind, small hydro and other generators
subject to a de minimis level of 10 MW. The details of the individual unit models are
given in Section 2.2.2. The load model is presented in Section 2.2.3, and an overview
of the entire Ireland model, with details of the connecting system and is presented in
Section 2.2.4. The development and tuning of the Ireland electricity system model
was carried out in collaboration with Dr. Damian and Flynn and Julia Ritchie of the
Queens University of Belfast.
2.2.1 Assumptions of the Model
A fundamental assumption made in the development of the system model is that fre-
quency is uniform throughout the system. This assumption is made on the basis that
the electricity system of Ireland is tightly meshed and electrically short, with the rel-
ative impedances between nodes quite small. Therefore, during a major contingency
involving the loss of signicant generation, the system will remain in synchronism and
the frequency deviation will be very similar at all points on the system. This is borne
out by system studies carried out by the Transmission System Operators and by fre-
quency measurements during major events with frequency deviations of up to 0.8 Hz
during sudden generation decits. The system is designed and operated so that in
the event of the loss of the largest infeed there should be no consequential events (i.e.
the protection does not trip out any other devices, for example lines). Historical data
shows that the loss of a transmission line has never occurred during a loss of generation
event. In particular, during a major loss of generation there are noticeable changes in
power ow across the AC lines between the NIE and ESB systems, due to the shar-
ing of reserve (DETINI, 2003; ESBNG, 2005c). These rapidly changing power ows
have never caused any additional tripping of lines. Therefore, for frequency control
studies on the Ireland electricity system, a uniform system frequency is assumed and
a single busbar model has traditionally been employed (OSullivan, 1996; OSullivan
and OMalley, 1996; OSullivan et al., 1999) and is appropriate for this study. It is
also assumed that changes in voltage have negligible eect on real power balance on
the system. In the event of a loss of generation on the system, voltage deviations will
occur around the site of the contingency. However, these eects are only local and will
not manifest themselves globally (OSullivan, 1996).
The objective of the development of the Ireland electricity system model is to examine
the eect of evolving plant mix on frequency control during a frequency event. The
timescale of interest is the time immediately prior to and the rst 20 seconds of a
frequency disturbance event. For the purpose of short-term frequency control, as of
interest in the system model, dynamics outside the timescale of interest are neglected.
In addition, the system is assumed to be in steady state at nominal frequency prior to
any frequency event.
On the Ireland electricity system, if the frequency falls below 49.7 Hz (99.4% of nom-
inal), it is deemed to be a signicant frequency disturbance event (ESBNG, 2005b).
Below 47.5 Hz (95% of nominal), the system will lose synchronism as generating units
and load trip o the system. The system model is designed for relatively small fre-
quency deviations of less than 3% ( 1.5 Hz). Therefore, system frequency changes
can be assumed to be small with respect to nominal system frequency.
2.2.2 Generation
The majority of electricity generating units consist of a prime mover to produce me-
chanical energy and a generator to convert this mechanical energy into electrical energy
suitable for supply to the power system (Machowski et al., 1997). A variety of energy
resources may be used to impart energy to the prime mover, resulting in a number
of dierent prime movers types. These can be broadly categorised as steam turbines,
combustion turbines and turbines which are driven directly by the energy resource,
such as hydroelectric turbines, wind turbines and tidal energy turbines.
The energy resources associated with steam units are generally fossil fuels and nu-
clear ssion. The energy from combustion of the fuel (or heat energy resulting from
the nuclear ssion) is used to produce steam in a boiler, which drives the steam tur-
bine. Combustion turbines, alternatively, use the exhaust gases from the combustion
reaction to drive the prime mover. Hydroelectric turbines can have many dierent con-
gurations, but water is used to drive the prime mover in all cases. Similarly, energy
from wind and tides can be extracted in appropriately designed turbines to produce
mechanical energy to convert into electrical energy in the generator.
Steam Units
Fossil fuelled steam units have been in use for over 120 years and can have a number of
dierent congurations. However, the basic principle remains the same: the combustion
of the fuel in the furnace heats the water in the boiler to produce steam, which is used
to drive the steam turbine. Thus, the working uid for a steam unit is water. The
rotational energy of the steam turbine is then converted to electrical energy in the
synchronous generator.
The boiler can be either drum or once through conguration. In a drum boiler, water
enters the waterwalls from the reservoir of water in the drum. Heat energy from the
fuel combustion process is transferred through conduction, convection and radiation to
the water in the waterwalls. The resultant steam and water mixture then re-enters the
drum. In the drum, steam separates from water and travels to the steam turbine, via
the superheater, with steam owrate dependent on the pressure dierential between
drum and turbine. The pressure in the drum is dependent on the fuel ring rate, and
steam owrate can therefore be controlled by adjusting the ring rate. Superheated
steam temperature is controlled by means of spray water attemperation (Flynn, 2003).
Drum boilers may use either natural or forced circulation, but natural circulation is
the more common conguration. Although higher pressure, and thus eciency, can be
achieved through the use of pumps, it is generally only viable in very large plants.
The once through boiler conguration contains no internal reservoir and the water is
forced through a continuous pipe to the steam turbine by means of pumping. As a
result, steam owrate is determined by the boiler feed pump and steam temperature
is controlled by adjusting the fuel ring rate. Once through boilers are designed to
operate at supercritical temperatures, yielding higher eciency. However, although
increased eciency results in reduced operating costs, the increased installation costs
of such boilers generally make drum boilers more economically viable (Flynn, 2003).
Boiler dynamics are highly complex, and over long time frames, detailed models are
required to accurately capture dynamic behaviour. Complex models derived from phys-
ical principles have been developed (Chien et al., 1958; McDonald and Kwatny, 1970;
Kwan and Anderson, 1970; Flynn and OMalley, 1999) to capture the dynamics of the
boiler over dierent timescales. A low order nonlinear boiler model was developed in
IEEE (1973b). In this model, pressure and steam owrate are dened as functions of
the energy input to the boiler and the turbine control valve area.
DeMello (1991) examined and justied the use of simplied boiler models (IEEE,
1973b) for power system modelling. Various simplied models, which captured the
essential nonlinear characteristics of the boiler, were compared to a detailed model
based on mass, volume and energy balance equations and found to be adequate for use
in power system dynamic performance studies. The simplied model of IEEE (1973b)
and DeMello (1991) was adopted for the emergency reserve model of OSullivan (1996),
and is employed in the Ireland system model of this thesis to model the boiler compo-
nent of the steam units on the Ireland system.
The boiler model is illustrated in the schematic of the steam turbine model in Fig. 2.2,
which also includes the steam turbine model and the governor model.
There are time delays in a boiler associated with the fuel dynamics and the transfer of
heat energy to the water in the waterwalls when generating steam. The time constant
for the fuel dynamics varies from 20 to 40 seconds depending on the fuel type in use.
The waterwall lag has a time constant of approximately 5 seconds. Due to the relatively
long time constant for the fuel dynamics in comparison with the time scale of interest
here, the heat energy, Q, can therefore be assumed constant. However, as the heat
energy Q remains constant, the waterwall lag component, a rst order lag, may be
neglected. Therefore, steam generation m
w
is equal to the heat energy Q.
Pressure in the drum, P
D
, is proportional to the integral of the dierence between the
steam generation m
w
(i.e. steam owing into the drum) and the steam owrate out of
the drum, m. The throttle pressure at the entrance to the steam turbine valve, P
T
, is
propotional to the integral of the dierence between the steam owrate from the drum,
Figure 2.2: Steam unit model (See text for details)
m and the steam owrate exiting the valve into the steam turbine, m
s
. The non-linear
nature of the boiler process is due to the steam owrate from the drum to the throttle,
which is proportional to the square root of the pressure dierence between the two.
From the boiler, the steam enters the steam turbine, where the heat energy in the steam
is converted to rotational energy to turn the turbine. When a gas passes through
a nozzle, it expands converting heat energy into mechanical energy. Therefore, as
the steam expands through the stages of the steam turbine blades, temperature and
pressure drop as the energy is imparted to the rotating turbine.
A model which can represent both reheat and non-reheat turbine systems is presented
in IEEE (1973a). This model, also used in IEEE (1991) and OSullivan (1996), is
employed to model the steam turbine component of the steam units in this thesis, as
illustrated in Fig. 2.2.
The steam ow entering from the boiler, m
s
, directly determines the power generated
in the turbine, with appropriate time delays incorporated. In a reheat steam turbine,
three time delays are required. The time delay for transport of steam from the boiler
to the rst turbine stage and also the conversion of steam to rotational energy in the
rst or high pressure turbine stage is represented using a rst order lag with a time
constant T
CH
. The time delay due to the reheater and conversion of heat energy to
rotational energy in the intermediate stage is captured by the rst order lag with time
constant T
RH
. Finally, the transport delay to the nal or low pressure turbine stage as
well as the conversion of heat to rotational energy in this stage is captured by the a rst
order lag with time constant T
CO
. The fractions of the total power output generated
at each stage are represented by F
HP
, F
IP
and F
LP
. A non-reheat unit will result in
all power generation coming from the high pressure stage and therefore F
IP
and F
LP
,
the fractions of the total power generated in intermediate and low pressure stages of
the steam turbine, will be zero.
All dispatchable generating units on the Ireland system are tted with droop governors
and required by regulations (ESBNG, 2005b; SONI, 2003a) to have a droop setting of
4%. The governor model used for the steam turbine units is the simplied speed
governor of Elgerd (1982), and is illustrated in Fig. 2.2. Using the dierence between
actual and nominal frequency as input, the signal to the control valve is determined,
which is proportional to the inverse of the droop, R
d
. A time delay with time constant
T
s
due to the action of the hydraulics associated with the turbine valves is incorporated,
along with maximum and minimum limits (P
max
and P
min
) to maintain the valve area
within actual limits.
Open Cycle Gas Turbine Units
In the open cycle gas turbine, the working uid is air. Air at atmospheric conditions
enters the compressor, where energy is imparted to the air from the spinning compressor
stages. On exiting the compressor, the air enters the combustion chamber where fuel is
added and combustion occurs, resulting in the conversion of chemical energy in the fuel
into heat energy. The hot gases produced by the combustion process then enter the
gas turbine under high pressure and are expanded as they pass through the dierent
turbine stages, converting the heat energy of the gases into rotational energy of the
gas turbine. The gas turbine, compressor and generator rotate on a single shaft. The
mechanical energy produced by the turbine, less the mechanical energy required by the
compressor, is the net mechanical energy and this is converted to electrical energy in
the synchronous generator.
Numerous models of gas turbines have been developed, with varying degrees of com-
plexity and detail (Rowen, 1983b; Hung, 1991; Cohen et al., 1996; Kunitomi et al.,
2001; Pourbeik, 2002). The dynamic model developed by Rowen (1983b) incorporates
the main dynamics of a gas turbine for a wide range of operating speeds (95 - 107%)
and conditions, and is suitable for use in power system stability studies. As a result,
this model is used as the basis for many gas turbine and combined cycle gas turbine
modelling applications (Bagnasco et al., 1998; Kunitomi et al., 2001; Lalor et al., 2005).
The model developed in Rowen (1983b) includes the relationships between pertinent
gas turbine components and also the control systems for speed, temperature and ac-
celeration in addition to maximum and minimum fuel limits. Further simplications
of the model are also outlined.
At the time of development of the emergency reserve model (OSullivan, 1996), a num-
ber of OCGTs existed on the ESB system. However, due to an absence of data for any
gas turbines during frequency events, identication of the dynamic parameters for the
gas turbine model of Rowen (1983b) was not possible (OSullivan, 1996). As a result,
a simple ramp based model was implemented, using the approximate reserve charac-
teristics supplied by the system operator to provide limited identication (OSullivan,
1996).
For the Ireland system model developed here, OCGT generating units are explicitly
modelled, using a model adapted from that of Rowen (1983b). The OCGT model
structure is illustrated in Fig. 2.3.
As acceleration control generally only comes into play during start-up and shut-down,
the acceleration controller is neglected in this model, due to the assumption that the
model is initially generating in steady state. The output from the speed and temper-
ature controllers feeds into a minimum selector, where the lower of the two signals
determines the fuel ow. Under normal operating conditions, the fuel ow is under
the control of the speed controller (the governor). The speed controller consists of a
simple droop governor. The temperature control loop compares the measured temper-
ature of the exhaust gases, T
xm
, to the rated exhaust temperature, T
r
. If measured
exhaust temperature exceeds rated exhaust temperature, the temperature controller
signal falls below unity, and the temperature controller takes over the fuel ow control.
The exhaust temperature T
x
is calculated using the equation (2.1):
Figure 2.3: Open cycle gas turbine model (See text for details)
T
x
= T
r
700 (1 W
f
) + 550 (1 N) (2.1)
where W
f
is the gas turbine fuel ow (per unit) and N is the system speed (per unit).
Power output is the product of the torque and the system speed, where the torque
produced by the gas turbine is determined by the equation (2.2):
Torque = 1.3 (W
f
0.23) + 0.5 (1
N
N
ref
) (2.2)
where N
ref
is the reference system speed (per unit).
Combined Cycle Gas Turbine Units
Combined cycle gas turbines fuse the technologies of both steam and gas turbine units,
resulting in a unit capable higher eciency than either of the two individual compo-
nents. CCGTs are not included in the emergency reserve model of OSullivan (1996).
However, due to a signicant increase in CCGT generation on the ESB system, these
units comprise a large proportion of total generating capacity. The details of the
CCGT model structure used for the system model in this thesis, together with the
CCGT model tuning and validation methodology are given in Chapter 3, where the
impact of CCGT short term dynamics on frequency control is studied.
Hydroelectric Units
Hydroelectric generation uses the potential energy contained in water moving from a
higher to a lower height to turn a turbine. Thus the potential energy is converted into
kinetic or rotational energy of the turbine, which in turn is connected to an electric
generator where kinetic energy is converted to electrical energy. There are three prin-
cipal forms of hydro-turbines available and choice of turbine will depend on both the
potential energy or head of the water supply and also the volume of water available.
The three types are the Francis turbine, the Kaplin turbine and the Pelton wheel (Kun-
dur, 1994). Both Francis and Kaplin hydraulic turbines are reaction turbines, relying
on the weight of the water column to react against blades of the rotating element of
the turbine, the runner. The runner is fully immersed in water and is enclosed in a
pressure casing. The pressure dierences across the runner blades imposes lift forces,
which cause the runner to rotate. Francis turbines are the most widely used turbine
type and designed to operate under medium heads of water, while the Kaplan turbine
generally operates under lower heads of water with higher ow rates. Pelton wheels are
impulse type turbines where the runner is spinning in air and rotated by the impact of
a water jet on the blades. These are generally used only with high heads of water.
One distinctive property characteristic of hydroelectric generation is the initial drop in
power output that occurs when the opening of the intake gate is increased. This non-
minimum phase characteristic is caused by the inertia of the water and the momentary
reduction in head that occurs when the gate initially opens further, before the water
ow through the turbine increases again (Elgerd, 1982; OSullivan, 1996). Therefore
adequate representation of this phenomenon is required for accurate dynamic models
of hydroelectric generating units, in particular for frequency control.
The two hydro-turbine types used for generation on the Ireland system are Francis and
Kaplan hydraulic turbines (OSullivan, 1996; ESBNG, 2004). A number of dierent
models for hydraulic hydro-turbines and their speed controllers have been developed
and a number of models of varying complexity are presented in IEEE (1992). Dierent
models are suitable for power system studies of dierent types, and recommendations
for their use are made. As governor and turbine characteristics play a signicant
role in the response of the system frequency to contingencies in small isolated power
systems, speed governor action and turbine response must be included in any model
used. IEEE (1992) recommends that a non-linear model assuming a non-elastic water
column is used for the turbine in conjunction with an appropriate governor model
which captures the dynamic characteristics of the governor under transient conditions.
The hydroelectric generation on the Ireland system can be divided into two categories:
run-of-river and pumped storage.
Run-of-River Hydroelectric Generation
Run-of-river hydro generation is entirely dependent on the available river ow. As
such, generation exibility is limited. Traditionally, run-of-river hydroelectric genera-
tion units on the Ireland system operate with governors fully open and therefore provide
little or no reserve. Therefore, the contribution of run-of-river hydroelectric generation
to overall system response may be neglected during frequency events, barring the iner-
tial contribution of the generator and the prime mover. As a result, in the emergency
reserve model of OSullivan (1996), run-of-river hydro was neglected. In the system
model developed in this thesis, the contribution of run-of-river hydroelectric generation
to primary operating reserve is also neglected and only their inertial contribution is
included.
Pumped Storage
Pumped storage is a method of storing energy that can be quickly converted into
electrical energy when required. A pumped storage facility consists of two reservoirs at
dierent elevations, and water is pumped from the lower to the upper reservoir during
non-peak times. When additional generation is required on the system, such as at
times of peak demand or during a generation decit, water is allowed to run from the
upper to the lower reservoir through the hydro-turbines, producing mechanical energy
which is then converted to electrical energy in the generators.
The ESB system has one pumped storage plant, Turlough Hill, with a capacity of 292
MW, consisting of four 73 MW units. Typical operating eciency is approximately
70% (ESBNG, 2005b). These units, are signicant contributors of reserve to the system
during frequency events. Each unit has four distinct modes of operation when synchro-
nised to the system: gen, min gen, spin and pump. In addition, when oine, each
unit can be brought online and begin generating within a much shorter time period
than many other conventional generating resources, assuming sucient water levels in
the upper reservoir.
When operating in gen mode, the unit is generating electricity. The operating point of
the unit when in gen mode is between 30 MW and maximum rated output of 73 MW,
and the operational eciency increases with increasing operating point. Generally,
however, the unit will operate below 50 MW to allow the provision of POR. When
operating in min gen mode, the unit is generating electricity, but with an operating
point of approximately 5 MW. In this mode of operation, the unit is primarily providing
POR, and operational eciency falls o dramatically. For both gen and min gen
modes, when system frequency falls to a pre-dened frequency level, f
gen
and f
min
respectively, it triggers the opening of the governing valve on the unit completely, and
power output increases rapidly until the unit is generating at maximum rated output.
In spin mode, the turbine runner is spinning in air, and the water governing valve is
closed. Therefore, the turbine is actually consuming a small amount of energy to turn
the turbine (usually 1 - 2 MW). However, if system frequency falls to a pre-dened
level, f
spin
, the opening of the governing valve is triggered, water enters the turbine
and the unit starts to generate, ramping up to maximum power output. Once triggered
by f
min
, time to reach generation is just over 20 seconds. As this exceeds the timescale
of interest in this study, spin mode is neglected here.
In pump mode, the unit is pumping water from the lower reservoir to the upper reser-
voir. When pumping, each unit consumes 73 MW. If a frequency disturbance event
occurs, and the system frequency falls below the pre-dened frequency, f
pump
, the unit
is tripped o the system almost instantaneously, reducing system load by 73 MW. At
night, up to four 73 MW units may be operating in pumping mode, which trip when
the predetermined frequency levels are reached, to provide static reserve.
The current pre-dened frequency settings for each unit in Turlough Hill (TH) are
given in Table 2.1.
When a Turlough Hill unit is generating (i.e. in gen or min gen modes) and the
Table 2.1: Under-frequency setting for Turlough Hill operating modes
Unit TH1 TH2 TH3 TH4
f
gen
(Hz) 49.8 49.8 48.75 49.75
f
min
(Hz) 49.8 49.8 48.75 49.75
f
spin
(Hz) 49.7 - 49.65 -
f
pump
(Hz) 49 49.4 49.1 49.6
frequency falls below the threshold frequency, initiating ramping up to the maximium
rated output, the dynamic characteristics of the hydro-turbine need to be taken into
account. In OSullivan (1996), insucient data was available to allow identication of
the non-linear model recommended by IEEE (1992). Instead the classic linear model
of Ramey and Skooglund (1970), as illustrated in Fig. 2.4, was used. In the case of
the linear model, only a single parameter, the water time constant T
w
was required to
determine the power output from the gate position, G
w
.
Figure 2.4: Linear hydroelectric-turbine model (See text for details)
Although signicantly more data is available for the development of the system model
for this thesis, the resolution of the data was limited to 1 second. Therefore, adequate
capture of the initial dynamics of the hydro-turbine for accurate identication of the
non-linear hydro model of IEEE (1992) was not possible. Therefore, the linear model
of Fig. 2.4 was selected for use in the Ireland system model.
When operating in pump mode, the unit is modelled as a discrete block of static reserve.
If f
pump
is reached, the unit is tripped instantaneously from the system with negligible
dynamics, resulting in a step change in system load.
Wind Turbine Generators
Wind generation has increased signicantly over the last number of years since rst
installed on the system in 1992. Of all generation technologies on the Ireland power
system ESBNG and SONI, in common with system operators worldwide, have lit-
tle experience with wind generation due to its relatively recent introduction to the
system. This inexperience is compounded by the fact that wind is a variable energy re-
source, with limited forecastability, and also wind turbine technologies predominantly
use asynchronous generators. The term wind turbine generator covers a number of dif-
ferent technology types, the two predominant categories being either xed or variable
speed wind turbines. As proportions of wind generation increase rapidly on the system,
representation of wind generation within the system model is imperative. Details of
the models used for each technology type are given in Chapter 4, where the impacts of
dierent wind turbine generating technologies on frequency control are studied.
Interconnection
The HVDC interconnection between Northern Ireland and Scotland is not congured
to provide reserve in the timescale of interest, and therefore the power ow across
the interconnection may be assumed constant for the system model in this short-term
frequency control study.
2.2.3 Load
Each generation unit on the system, with the exception of those that do not respond
to changing frequency, is modelled as described in Section 2.2.2. Continuous balance
between the electrical output of generating units and the electrical load is required for
stable system operation, and the characteristics of the load play a signicant role in the
dynamics of the power system. Therefore, careful choice of load representation for any
model is imperative, as it can have a signicant impact on the analysis results (IEEE,
1993).
The electrical load consists of all components on the power system that consume power.
These include not only connected load devices, but also the eects of transformers,
transmission and distribution feeders, voltage regulators and reactive power compen-
sation devices (Kundur, 1994).
Load models can be divided into two categories: static and dynamic. A static load
model is dened as a model that expresses the active and reactive power at a given
instant in time as functions of the frequency and magnitude of bus voltage at the same
instant (IEEE, 1993). The voltage dependency of the power may be represented using
either an exponential model or polynomial model. The frequency dependency of the
power can be represented through the multiplication of the voltage dependency model
by a frequency factor. Static load models may be used for approximating dynamic load
components as well as for static load components (IEEE, 1993).
A dynamic load model models the active and reactive power at any instant in time
as functions of the frequency and magnitude of bus voltage at past (and usually also
present) instants in time (IEEE, 1993). In order to include the dynamic behaviour of
both induction and synchronous motors in the load model, the dynamic load model of
OSullivan (1996) is employed in the Ireland system model. This frequency dependent
load model is derived from Welfonder et al. (1989), and is described by 2.3 and 2.4.
P
Load
=
k
pf
+sT
pf
1 +sT
1
f +
k
pv
+sT
pv
1 +sT
1
V (2.3)
Q
Load
=
k
qf
+sT
qf
1 +sT
1
f +
k
qv
+sT
qv
1 +sT
1
V (2.4)
This load model is derived from the behaviour of the three main load types i.e. Resistive
or ohmic load, inductive and capacitive load and motor load. Changes in active power
P
Load
and reactive power Q
Load
of the load are calculated as functions of the changes
in frequency f and voltage V , where k and T are the load parameters, T
1
is the time
constant of the load, subscripts p and q refer to the real and reactive power respectively
and subscripts f and v refer to the frequency and voltage.
However, due to the assumption that voltage eects are negligible and as reactive power
has little eect on system frequency, the load model of 2.3 and 2.4 reduces to
P
Load
=
k
pf
+sT
pf
1 +sT
1
f (2.5)
where
k
pf
is the steady state frequency sensitivity of the load
T
pf
is the ratio of load inertia to system frequency
T
1
is the load time constant
The load time constant T
1
can be approximated by
T
1

=

o
f
o
T
pf
P
o
(2.6)
where
o
is the average slip of an average induction machine
f
o
is the steady state frequency
P
o
is the steady state system demand
There are two approaches used to obtain the load data required for the determination
of load model parameters. The rst, the component based approach, uses the aggrega-
tion of models of individual components of the load. The load is divided into classes,
e.g. residential, industrial etc., and then each class is further divided into load compo-
nents. Although this approach requires no actual measurements, it requires extensive
information on the load classes and the components of each class.
The second approach to gathering load data is the measurement based approach. In
this case, actual measurements of the load characteristics are taken to generate the
characteristics of the load under various conditions. The time of day, week or year can
result in large variations in load conditions. This approach was used to gather load
data for the tuning of the above load model in OSullivan (1996) and OSullivan and
OMalley (1996), and is applied for the load component of the Ireland system model.
Interruptible Customers and Under Frequency Load Shedding
Interruptible load is a form of static reserve. By prior agreement between the customer
and the system operator, specic loads on the power system are tripped by under-
frequency relays if frequency falls to a threshold level, f
int
. On the Ireland system, this
threshold frequency is set at 49.3 Hz.
Under-frequency load shedding (UFLS) is also a form of static reserve, which is called
upon only if all other sources of primary operating reserve fail to arrest declining
frequency. UFLS relays set at dierent threshold frequencies, f
UFLS
, trip discrete blocks
of load o the system. On the Ireland electricity system, UFLS commences at 48.9 Hz.
In the Ireland system model, both interruptible customers and UFLS are represented by
discrete blocks of load, which are tripped instantaneously when the system frequency
falls below the threshold frequency (f
int
or f
UFLS
) for each block, resulting in a step
change in load.
2.2.4 Connecting System
Frequency responsive models for both the generating units and the load have been
developed in Sections 2.2.2 and 2.2.3. Due to the assumption of uniform frequency at
all points on the system during a frequency event, and the assumption that voltage
eects may be neglected, the transmission system may be modelled as a single busbar.
This is a commonly adopted approach when modelling small isolated power systems
(Thompson and Fox, 1994; Kottick and Or, 1996; OSullivan and OMalley, 1999).
Therefore, in the Ireland system model, the issue of system balance is represented as a
single block, the connecting system, as discussed in OSullivan (1996).
When the system is operating in steady state, total generation P
GEN
balances total
system load P
LOAD
.
P
GEN
= P
LOAD
= P (2.7)
Assuming uniform frequency at all points on the system at nominal frequency, f
o
, the
kinetic energy, KE
j
, of each individual generator,j, on the system can be aggregated
to give a single kinetic energy term, KE. In the event of a power imbalance, total
generation no longer equals total system load, so
P = P
GEN
P
LOAD
(2.8)
where P is the net change in power of the system. This power imbalance is redressed
by a change in the kinetic energy, KE of the system, such that
dKE
dt
= P (2.9)
The kinetic energy of a rotating mass is proportional to the speed squared. Therefore,
KE = KE
o
(
f
f
o
)
2
(2.10)
where KE
o
is the kinetic energy of the system at nominal frequency f
o
. As a result,
equation (2.9) becomes
dKE
dt
= KE
o
d(
f
fo
)
2
dt
=
2KE
o
f
f
2
o
df
dt
= P (2.11)
As frequency changes are assumed small relative to nominal frequency, f f
o
. Thus,
2KE
o
f
o
df
dt
= (P
GEN
P
LOAD
) (2.12)
As a result, the change in system frequency may be calculated from
df =
f
o
2KE
o
(P
GEN
P
LOAD
)dt (2.13)
The connecting system, as described by equation (2.13), completes the feedback loop
for the system model. The change in system frequency is calculated as a function of the
power imbalance, and the generator inertia (load inertia is contained within the load
model). This frequency feeds to generator and load models, completing the feedback
loop. The entire system model is illustrated in Fig. 2.5.
Figure 2.5: Ireland system model
2.3 Data
A comprehensive up-to-date catalogue of all generation on the ESB system was com-
piled from data made available by the ESB system operator, ESBNG and the NIE
system operator, SONI. Included was generator type, fuel, capacity and congura-
tion in conjunction with unit inertias, maximum and minimum operating points and
documented reserve characteristics.
Data from a number of historical system frequency events (approximately fty), with
frequency deviations of up to 2%, that have occurred on the system over the last
ve years was also available from ESBNG for the tuning of model parameters and
validation purposes. Pre and post event system frequency and power output data from
all generating units on the ESB system was available, sampled at 1 Hz. The ambient
conditions at the time of each event were also obtained.
Similar historical data for a smaller number of frequency events was available from
SONI, many of which corresponded to the historical frequency events available from
ESBNG. In addition, system frequency traces with a resolution of 10 Hz was also
available for 8 historical frequency events for which generating unit response data was
available for both the ESB and NIE systems.
Due to the abundance of studies in power system component modelling available in
literature, as highlighted in Sections 2.2.2 and 2.2.3, parameter values for many of the
models employed in the system model are readily available in literature (IEEE, 1973a,
1991; Rowen, 1983b; Bagnasco et al., 1998; CIGRE, 2003). These parameter values
provide a reference point for the validation of parameter values of the tuned Ireland
system model components. Also, OSullivan (1996) and Fox et al. (1989) provide
parameter values for some of the generating units in the ESB and NIE systems circa
1995 and 1988 respectively.
2.4 Simulation Tools
The Matlab Simulink software package was used for the construction and implemen-
tation of the system model, due to its high degree of exibility. Matlab is a high-
performance technical computing environment. Simulink is a platform within the Mat-
lab environment for the design, building and simulation of dynamic systems, with an
interactive graphical environment. Models can be constructed using both predened
and customisable blocks. The resultant models can then be run using a range of nu-
merical integration algorithms or solvers. In addition to a number of xed and variable
step solvers, sti or discontinuous systems can also be simulated using several solvers
for such systems. A variable step continuous solver, where maximum and minimum
step sizes could be specied in addition to absolute tolerances, was initially used to
simulate the system model in this thesis, as all the dynamics within the models were
of similar speed. However, on integration of the wind turbine generator models, a vari-
able step size solver for sti systems was applied, due to the stiness of the resultant
combined model. The Matlab environment was used for data processing.
2.5 Tuning the Ireland System Model
The data described in Section 2.3 was employed to tune the individual generator models
and the load model.
2.5.1 Generating Units
Each generating units conguration, capacity, maximum and minimum operating
points and inertia was determined from the data available from both ESBNG and
SONI.
Initially, parameter values from OSullivan (1996) and Fox et al. (1989) were used for
all ESB generating units for which they were available, in addition to generator inertia
values from ESBNG and SONI. Parameter values from literature were initially used for
all generating units without parameter values in previous ESB and NIE system models.
All generating units on the Ireland system are required to operate with a droop of 4%
(ESBNG, 2005b; SONI, 2003a). It was, therefore, initially assumed that each generator
operated in accordance with this requirement, with the governor droop of 4%.
Each generator model was individually tuned as a stand alone unit using data from
the available system contingency events. The actual recorded system frequency was
input to each generator model, and the simulated active power response compared with
the recorded generator response for the event. As the inertial responses of the gener-
ating units are accounted for in the connecting system, the inertial response of each
individual unit is not accounted for in the individual models outlined in Section 2.2.2.
However, the actual recorded unit response data from system events incorporates both
the inertial and governor response of the generating unit. Therefore, for comparison
between simulated and actual unit responses, incorporation of the inertial response of
each generator was required for the individual tuning of the unit models.
The inertial response of a generator is described by equation 2.14
P
int,j
= I
j
d
dt
(2.14)
where P
int,j
is the inertial response of the generating unit j, I
j
is the inertia of the
generating unit j (kgm
2
) and is the system speed.
Therefore, the total response of each generating unit is calculated from equation (2.15).
P
j
= P
model,j
+ P
int,j
(2.15)
where P
model,j
is the power output from the appropriate model for that particular
generating unit. The additional control loop included in each model to simulate the
inertial response is illustrated in Fig. 2.6, where
i
is the time constant of the inertial
response control loop and INT is a logical operator. When INT has a value of 1, the
inertial response is included in the simulated power response, P
j
, from generating
unit j. When zero, the response from the generating does not include the inertial
response (i.e. P
int
= 0).
Figure 2.6: Inertial response control loop
It must be noted that for a number of generating units, not all available unit response
data was employed. In each case, a small number of the recorded responses diered
substantially to the majority of recorded responses from that unit to similar frequency
events. The discrepancies between the responses of each unit to dierent events was
attributed to unknown circumstances such as temporary component malfunction or
the temporary deration of a unit. As such, irregular generating unit responses were
neglected.
The power output data for a generating unit on the Ireland system during a number of
frequency disturbance events is illustrated in Fig. 2.7. The response of the unit in plot
(VI) B clearly diers substantially to the other ve responses of that generating unit
(plots (I) B - (V) B), taking into account the operating point prior to the on-set of the
events. The unit is operating at 93.5% of rated output prior to the event. However,
the unit does not increase power output in response to the frequency deviation, with
the exception of the inertial response.
Steam Units
Initial parameter values from OSullivan (1996) were used for the steam unit models.
In comparisons of simulated steam unit responses with actual recorded unit responses,
however, there were clear discrepancies between the two for a number of generating
units.
Figure 2.7: Six sample frequency traces for events on the Ireland electricity system,
with corresponding power output traces for a single generating unit. (I-VI) A: System
frequency (Hz). (I-VI) B Power Output (%). Note: All x-axes values are time (s).
Initially, the value of the governor droop was examined for each unit individually. A
number of frequency events were identied in which the system frequency remained
at a steady value below nominal frequency after the event. For those generating units
online during the events in question, it was possible to calculate approximate droop
values by equation 2.16.
R
d
=
N
P
pu
100 (2.16)
where R
d
is the droop (%), N is the change in system frequency (per unit), and P
pu
is the change in power output (per unit).
Calculations indicated that although a signicant number of steam units were found to
operate with a droop value of 4%, some droop values were found to range between 3 and
8%, and several units were found to consistently have negligible response to changing
system frequency. The droop value of each steam unit model, where applicable, was
altered to reect the calculated droop values.
Subsequent comparisons between simulated and actual steam unit responses showed
that although revising droop values improved the simulated responses, inconsistencies
between the actual and simulated responses remained for some units.
The parameter values in OSullivan (1996) were identied using an optimisation al-
gorithm based on the Nelder Mead simplex search method. As there are thirteen
parameter values in a steam unit with a reheat turbine, this order was found to be too
high to be handled eectively by the optimisation algorithm, given the available data.
Therefore, in OSullivan (1996), the parameter values of the turbine and the gover-
nor were set according to parameter values available in literature and based on ESB
experience. The parameter values of the boiler model were then identied using the
optimisation algorithm. Some parameter values identied in OSullivan (1996) were
found to outside the range of values in literature. Some discrepancy can be accounted
for by the fact that the inertial response of each individual unit was not modelled in
OSullivan (1996) and as a result not taken into account when obtaining parameter
values during optimisation.
Therefore, the boiler integral constants, C
D
and C
SH
, of those units for which good
agreement between actual and simulated was not present were manually tuned. Mod-
ication to the time constants of the turbine was also necessary for several units, due
to the model responding slower than the actual unit. This tuning led to further im-
provements in the model performance, resulting in good correlation between actual and
simulated steam turbine unit responses for the majority of events.
A comparison between actual and simulated frequency response of a steam unit is
illustrated in Fig. 2.8, where the importance of the incorporation of unit inertial re-
sponse for the tuning of each generating unit is also illustrated. If inertial response is
neglected, inaccurate parameter values could be obtained.
0 5 10 15 20 25
49.6
49.7
49.8
49.9
50
(I)
Time (s)
F
r
e
q
u
e
n
c
y

(
H
z
)
0 5 10 15 20 25
1
0
1
2
3
4
Time (s)
C
h
a
n
g
e

i
n

P
o
w
e
r

O
u
t
p
u
t

(
%
)
Actual
Simulated (No Inertia)
Simulated
(II)
Figure 2.8: Comparison between actual and simulated frequency response of a steam
unit to a low frequency event. (I) System frequency (II) Change in power output (%)
Open Cycle Gas Turbines
No parameter values from previous models of the ESB and NIE systems were available
for the OCGT units. Therefore, initial parameter values were adopted from literature
including Rowen (1983b); Ordys et al. (1994); Kunitomi et al. (2001), and the simulated
responses were compared with actual responses. As with the steam units, discrepancies
between simulated and actual responses were evident.
Once again, the rst step was to calculate the approximate steady state governor droop
value of each unit as above. Droop values were found to range between 4 and 8% for the
small number of OCGT on the Ireland system at present. The value of the rst order
time lag (representing the governor valve delay) from literature in the governor model
was also found to result in a simulated response which was slower than the actual unit
response, and this parameter value was reduced accordingly to speed up the response
of the unit to correspond to the response time.
Having tuned the governor parameter values, the response of the OCGT models using
parameter values from literature for the valve positioner, gas fuel system and compres-
sor discharge time lag were found to give a reasonable approximation of the actual
units responses. Adequate data for the detailed tuning of the temperature control loop
parameters was not available. Therefore, OCGTs are generally either operating near
minimum load capability and capable of providing primary reserve, or not generating
on the system, but ready to come online to provide replacement reserve when required.
Therefore, sucient event data for which the gas turbines are operating near maxi-
mum load (the conditions where the temperature controller will be activated) was not
available for tuning of the temperature controller. However, as the incorporation of
the temperature controller is vital to the behaviour of the OCGT response near peak
load conditions, this control loop is retained in the model and parameter values from
literature were used.
Pumped Storage
The use of the linear model of Ramey and Skooglund (1970) was proposed in Section
2.2.2 to model each Turlough Hill unit when operating in min mode . However, on
examination of all available data for Turlough Hill when operating in gen and min gen
modes, it was found that a linear ramp of power output against time after an event
provided a more accurate representation of the dynamic behaviour of the units during
a frequency event than the response provided by the linear model. Therefore, the linear
model was rejected and the response of each Turlough Hill unit when operating in gen
and min gen modes was represented with a simple ramping model.
Each Turlough Hill unit was tuned separately, to ensure any possible dierences in
operating characteristics are taken into account. The slope of the response of each unit
when operating in gen and min gen modes was determined by inspection of all available
data. Then, using an averaging process, the ramp rate of each unit was determined.
The increase in power output in response to two dierent frequency disturbance events
is illustrated in Fig. 2.9 for a generating unit in Turlough Hill.
0 10 20 30
0
5
10
15
20
25
30
35
40
45
(I)
Time (s)
C
h
a
n
g
e

i
n

P
o
w
e
r

O
u
t
p
u
t

(
M
W
)
Actual
Simulated
0 10 20 30
0
5
10
15
20
25
30
35
(II)
Time (s)
C
h
a
n
g
e

i
n

P
o
w
e
r

O
u
t
p
u
t

(
M
W
)
Actual
Simulated
Figure 2.9: Turlough Hill generating unit response when operating in gen mode for
two low frequency events, (I) and (II).
In addition to the ramp rates of the individual units, an additional level of complexity
is imposed by the existence of a maximum ramp rate for the entire Turlough Hill plant.
This maximum ramp rate is the result of a limit on the rate at which the ow of water
in the penstock from the upper lake can be increased. Therefore, if multiple units
are operating in min mode, the ramp rate of the plant may be less than the sum of
the possible individual ramp rates. To incorporate this phenomenon into the model, a
rate limiter is placed on the total power generated from the Turlough Hill plant. This
maximum rate determined from ESBNG experience, was veried through examination
of the total power produced by Turlough Hill during a frequency event.
As the spin mode time-frame is outside the POR range, Turlough Hill units operating
in spin mode were neglected for the purposes of the short-term frequency response
system model.
When in pump mode, if f
pump
is reached, the unit is tripped instantaneously from the
system, with negligible dynamics. Therefore, no tuning of parameter values for this
mode of operation was required.
Run of River Hydroelectric Units
On examination of historical data, run-of-river hydroelectric generation was found to
rarely contribute to the frequency response of the Ireland system. Therefore, the
frequency response of run-of-river hydroelectric generating units on the Ireland system
was neglected.
Combined Cycle Gas Turbines and Wind Turbine Generators
The tuning of the combined cycle gas turbine models is described in Chapter 3. At
present, wind turbine generators are not congured with governor response to changing
system frequency. Therefore, inertial response is the only response to changing system
frequency, and this is discussed further in Chapter 4.
2.5.2 Load Model
The load model parameter values of OSullivan and OMalley (1996) were employed
as initial parameter values. The proportion of dierent load components varies con-
siderably with time of day. Therefore the tuning of parameter values for a single load
model is problematic as parameter values will naturally vary with time of day.
From system operator experience (ESBNG, 2005b), the value of the steady state fre-
quency sensitivity of the load on the Ireland electricity system is approximately 2.5%
change in system load for a 1% change in frequency. This is consistent with the range
of values for the frequency sensitivity of dierent load components, as given in Kundur
(1994). However, the frequency sensitivity of the load is dependent on the time of day,
due to the variations in load mix. As the system model is being developed to predict
the frequency behaviour during a loss of generation at various times of day, the value
according to system operator experience of 2.5% is selected for k
pf
here. This frequency
sensitivity was also applied in the dynamic load model of (OSullivan and OMalley,
1996).
The total inertial response of the system is the aggregate of the generator inertial
responses and the inertial responses of the load. The initial rate of change of frequency
immediately following a loss of generation is determined solely by the system inertia.
Therefore, knowing the size of the event, the resultant frequency trace and the inertial
response of each generator, it is possible to use the connecting system of the system
model to compare simulated frequency slope with actual frequency slope. In addition
to comparing the slope of the two traces, the rate of change of frequency of the actual
and simulated frequency was plotted and compared. Load inertia was moved from the
load model to the connecting system for this process. Thus the total system inertia
which gave the closest match in initial slope and initial rate of change of frequency
when compared with the actual frequency trace was determined. The frequency traces
used were of 10 Hz resolution. Load inertia was found to vary with time of day and
week, but it was found that load inertia can be approximated as one third of the inertia
of system generation online (excluding wind generation on the system).
The parameter value T
1
can then be calculated using from equation (2.6), using T
pf
as
determined above, the steady state system demand P
o
, the steady state frequency, f
o
and
o
the average slip of an average induction motor.
2.5.3 Connecting System
The inertia of the system load is accounted for within the load model in the T
pf
and,
to a lesser extent, T
1
terms. In contrast, the inertia of each generating unit is not
included in the individual models, but instead as a single lumped kinetic energy term
within the connecting system, as explained in Section 2.2.4.
In order to validate the location of the generation inertia as a single aggregated value in
the connecting system, a series of simulations were undertaken. Each generating unit
model was designed such that the inertia could be included or excluded, as described
in Section 2.5.1. For each simulation, the inertia of each generator on the system was
accounted for either within the connecting system or within the individual generating
unit model. However, the inertia of each unit was only included once. The proportion
of inertia accounted for in the connecting system was varied. Therefore, while the total
system inertia remained constant, the location of the system inertia was varied.
It was found that the resultant simulated system frequency trace was not dependent
on the location of the inertia. However, with the exception of wind turbine models,
the inertia of each generator is located in the aggregate inertia term in the connecting
system, as described in Section 2.2.4.
2.6 Results and Discussion
The validation of the system model with frequency disturbance events on the Ireland
system is described in Section 2.6.1. This system model was developed primarily to
examine the eect of loss of generation events on system frequency under dierent
operating conditions. In this thesis, it is employed to examine the impact of increasing
proportions of CCGT and wind generation on short-term frequency control in Chapters
3 and 4.
One other application of the validated system model is outlined in Section 2.6.2, where
the system model is employed to derive a minimum frequency constraint. This work
was carried out in collaboration with Ronan Doherty (Doherty, 2005). The minimum
frequency constraint derived was incorporated in a co-optimised market dispatch. A
more detailed description of this work can be found in (Doherty et al., 2005).
2.6.1 Validation
Once the individual components of the system model were tuned, the system model was
validated using four events for which 10 Hz resolution frequency traces were available.
The four frequency events used for validation were not used in the tuning of the model,
ensuring blind validation. The system model was initially operating in steady state
and the disturbance was then applied. Simulated system frequency was compared to
actual system frequency. As illustrated in Fig. 2.10, Fig. 2.11, Fig. 2.12 and Fig. 2.13,
good correlation between actual and simulated frequency responses were observed.
0 5 10 15 20 25
49.3
49.4
49.5
49.6
49.7
49.8
49.9
50
Time (s)
F
r
e
q
u
e
n
c
y

(
H
z
)
Actual
Simulated
Figure 2.10: Actual and simulated system frequency for a 267 MW generation loss
(August)
2.6.2 Frequency Control in Competitive Electricity Market
Dispatch
The validated system model of the Ireland system is consequently used to derive a
minimum frequency constraint, which is one of two frequency-based constraints incor-
porated into a co-optimised linear programming (LP) market dispatch model (Doherty
et al., 2005; Doherty, 2005). Traditionally, primary operating reserve was dispatched in
a least cost manner to meet a simple heuristic reserve target. For the Ireland electricity
system, for example, this reserve target is 317 MW, regardless of system conditions.
However, with changing dynamic and inertial characteristics of power system due to
evolving plant mix, this approach to the provision of frequency control may no longer
be adequate to ensure security of the system. Therefore, two frequency constraints are
derived and included in a co-optimised market clearing algorithm so the system can be
dispatched in a least cost manner, while meeting frequency control criteria. The two
frequency constraints, a rate of change of frequency (ROCOF) constraint and a min-
0 5 10 15 20 25
49.5
49.55
49.6
49.65
49.7
49.75
49.8
49.85
49.9
49.95
50
Time (s)
F
r
e
q
u
e
n
c
y

(
H
z
)
Actual
Simulated
(January)
imum frequency constraint, are derived based on system variables that have a direct
impact on the frequency behaviour of the system. This methodology is applied to the
Ireland electricity system.
Minimum frequency reached in the event of a loss of generation is a function of the
frequency responses of both the generators and the load on the system in addition to
the magnitude of the power imbalance and the inertia of the system. The minimum
frequency reached in the event of a power imbalance on the Ireland system can be pre-
dicted using the fully validated system model, described above. The Ireland electricity
system is deemed secure if minimum frequency does not fall below 49.3 Hz and if rate
of change of frequency (ROCOF) does not exceed 0.25 Hz/s (ESBNG, 2005b).
For the accurate formulation of the minimum frequency constraint, the creation of
a very large database of events where either a minimum frequency of 49.3 Hz or a
maximum ROCOF of 0.25 Hz/s is reached is necessary. While the simulation time of a
single event using the validated system model is less than real time, the use of this model
0 5 10 15 20 25
49.2
49.3
49.4
49.5
49.6
49.7
49.8
49.9
50
Time (s)
F
r
e
q
u
e
n
c
y

(
H
z
)
Actual
Simulated
(June)
for the generation of a very large database of events would be very computationally
intensive. Therefore, the system model of Fig. 2.5 was reduced to a simplied model,
as illustrated in Fig. 2.14.
The combined generator reserve response in the simplied system frequency model is
determined using a black box model, which is tuned to the validated Ireland system
model. A number of frequency events were simulated using the validated system model,
where both the source and magnitude of primary operating reserve available from
generation, in addition to the system inertia and the magnitude of the power imbalance
were varied. In each case, a frequency nadir (minimum frequency) of 49.3 Hz was
reached. The magnitude of primary operating reserve available from each unit was
taken to be the maximum possible reserve response from that unit 5 seconds after the
on-set of the event. The black box model was subsequently tuned to the validated
system model, and is valid only for events that result in a minimum frequency of 49.3
Hz. A comparison between the responses simulated by the validated system model and
the black box model is illustrated in Fig. 2.15.
0 5 10 15 20 25
49.6
49.65
49.7
49.75
49.8
49.85
49.9
49.95
50
Time (s)
F
r
e
q
u
e
n
c
y

(
H
z
)
(April)
It was found that for frequency events reaching a minimum frequency of 49.3 Hz,
the black box model accurately approximated the validated system model over the
range of primary reserve available on the Ireland system. This simplied model was
subsequently used to generate a database of over 20,000 event scenarios where the
magnitude of the power imbalance, the system load and the reserve levels were varied
to encapsulate all likely N-1 events on the system. Upon examination of this database,
although the minimum frequency constraint is nonlinear in nature, it was however found
to remain convex. Thus the minimum frequency constraint could be approximated with
only very small errors using 5 four-dimensional linear functions (See Doherty et al.
(2005) for more details).
The ROCOF during a frequency event is a function of both the size of the power
imbalance and the inertia of the power system components synchronised to the system.
Using a maximum allowable ROCOF of 0.25 Hz/s, the maximum ROCOF constraint
Figure 2.14: Simplied system frequency model
is formulated as follows (Doherty et al., 2005)
N
G
i=1,i =k
KE
i
P
k
KE
L
, k G (2.17)
where
N
G
is the number of generators,
KE
i
is the kinetic energy of generator i (MWs)
P
k
is the size of the generator lost to cause the power imbalance (MW)
is a scalar with value 100 (s)
KE
L
is the total kinetic energy of the load (MWs)
G is the set of generators.
The two frequency based constraints described above are subsequently incorporated
into an LP market dispatch algorithm, which results in a least cost solution to meeting
frequency control criteria. The consideration of system kinetic energy and the size of
the largest possible contingency in addition to primary reserve results in a more secure
system at least cost. In contrast, frequency control dispatched using simple heuristic
methods is not always adequate to meet the frequency criteria for a secure system.
0 5 10 15 20
0
100
200
300
400
500
(B)
Time (s)
C
h
a
n
g
e

i
n

P
o
w
e
r

G
e
n
e
r
a
t
e
d

(
M
W
)
System Model
Black Box Model
0 5 10 15 20
0.8
0.6
0.4
0.2
0
(A)
Time (s)
F
r
e
q
u
e
n
c
y

D
e
v
i
a
t
i
o
n

(
H
z
)
Figure 2.15: Comparison of the generation response of the black box model with the
validated system model. (A) Simulated system frequency (Hz). (B) Change in power
generated (MW)
2.7 Conclusions
A short-term dynamic system model of the Ireland electricity system suitable for simu-
lating system frequency for up to 20 s during a frequency disturbance event is developed
and tuned. Individual generator models were tuned using historical data. The results
of tuning to recorded system data indicated that generator governor droop values on
the system range between 3% and 8%. The current system operator requirement is
4% (SONI, 2003a; ESBNG, 2005b). A number of generating units do not respond in
accordance with this requirement. Therefore, actual droop characteristics need to be
considered when scheduling reserve to meet the reserve targets. Not taking the actual
droop characteristics into account could lead to insucient POR on the system to
maintain the required system security level.
The system load was represented using a single measurement based dynamic load
model, and tuned to historical data, taking operator experience into account for certain
parameters. This load model showed good accuracy to represent the load over a wide
range of system conditions. However, the variation of system load mix with changing
system conditions was not accounted for, due to insucient data to tune a load model
for each condition. It is possible that the tuning of the load model to dierent system
conditions would increase the accuracy of the load model frequency response. However,
a signicant increase in system event load data would be required.
The simulated system frequency of the resultant model showed good agreement with
system frequency from low frequency events on the Ireland system. Therefore, it can be
concluded that the derived system model can be used to predict the system frequency
during a loss of generation scenario on the Ireland system with reasonable accuracy.
The system model is appropriate for examining the consequences of varying generation
mix, reserve mix and reserve targets on system frequency during loss of generation
events (Doherty et al., 2005).
Chapter 3
Frequency Control with Combined
Cycle Gas Turbines
3.1 Introduction
With ever increasing proportions of combined cycle gas turbine (CCGT) plant in the
majority of electricity systems worldwide, the dynamic characteristics of CCGTs have
become an issue of considerable interest over the last ten years. While the eciencies
of steam turbine generators are in the region of 40%, and of open cycle gas turbines
approximately 35%, CCGTs have a superior eciency in excess of 50% (Ordys et al.,
1994). In addition to superior eciency, the greater exibility and lower emissions
of CCGT units in comparison with many conventional thermal generators, combined
with progressively shorter installation times and reducing installation costs are the
basis for the increasing trend in CCGT generation. Knowledge and understanding of
the dynamic behaviour of CCGT generating plant is crucial to the maintenance of both
system reliability and security in electricity systems. This is particularly true as the
move towards competitive electricity markets means system operators now have little
or no control over the type and location of new plant investment.
The response of a CCGT during frequency excursions from nominal frequency diers
from the response of conventional steam turbines. An understanding of this response
characteristic to frequency events is essential. For small systems, the response of the
CCGT during a frequency deviation could potentially have a relatively large inuence
Chapter 3. Frequency Control with Combined Cycle Gas Turbines 54
on the severity of the event. While frequency deviations on larger systems are generally
less sizeable, knowledge of the behaviour of the CCGTs during islanding contingencies,
where CCGTs are possibly exposed to large generation/demand imbalances, is also
vital. In addition, as the proportion of CCGTs on larger systems increases, their
inuence will become more signicant.
An example of the importance of understanding the dynamic behaviour of CCGTs of
particular relevance is the August 1996 blackout of Peninsular Malaysia that occurred
as the result of a serious generation loss. The electricity system in the Peninsular
and Sahab regions of Malaysia, operated by the Tenaga Nasional Berhad (TNB) power
utility, has a peak load approaching 14,000 MW, with approximately 29 % of the gener-
ation consisting of CCGT and gas turbine units (Asean Energy, 2005). The behaviour
of gas turbines and CCGT units in response to the frequency disturbance contributed
signicantly to the severity of the event (Asean Energy, 2005). As a consequence of
the incident, several modications to improve the response of both gas turbine (GT)
and CCGT controllers to large frequency excursions were incorporated.
The objective of this work is to develop a CCGT model suitable for studying the
dependency of CCGTs on system frequency, which can then be tuned to represent
individual CCGT units on a power system. Analysis of CCGT dynamic response
during a power imbalance event is then carried out with a timescale of interest of up to
20 s following a system disturbance, and the consequent impact on system frequency
control is evaluated.
3.2 CCGT Background and Characteristics
Combined cycle gas turbines integrate the technologies of both the gas turbine and
the steam turbine. The exhaust gases from the gas turbine are fed into the heat
recovery steam generator (HRSG), which produces a supply of steam to drive the
steam turbine. The result is a generating unit that has a higher combined eciency
than either of the two individual components. The MW rating of the gas turbine is
generally approximately twice that of the steam turbine.
Combined cycle gas turbines may be divided into two general congurations: single-
shaft and multi-shaft. For the single-shaft conguration, the gas turbine and steam
turbine are on a single shaft and drive a single generator, as illustrated in Fig. 3.1.
Figure 3.1: Single-shaft CCGT (CIGRE, 2003)
Locating the generator between the gas and steam turbines on the shaft and employing
a clutch between the steam turbine and the generator allows for the disconnection of
the steam set if required and this technology is prevalent in the majority of recent
single shaft gas turbines. It must be noted, however, that the eciency of the CCGT
declines rapidly when the steam turbine component is disconnected. The advantages
of the single-shaft CCGT conguration over the multi-shaft conguration are lower
initial capital costs and a smaller area requirement.
The multi-shaft conguration has separate generators and thus separate shafts for both
gas turbine and steam turbine components. This conguration allows multiple gas
turbines, each with individual generator, and possibly individual HRSGs, to supply a
single steam turbine. The multi-shaft CCGT has the advantage of having multiple gas
turbine components, allowing for increased exibility and the possibility of construction
in stages.
3.2.1 Gas Turbine Component
The gas turbine component of the combined cycle plant is very similar to that of an
open cycle gas turbine (OCGT) and consists of a compressor, fuel injection system
and combustor, and turbine. Air, initially at atmospheric conditions, enters the com-
pressor. Variable inlet guide vanes (IGVs), tted at the entrance to the compressor,
can be adjusted to control the incoming airow. The air is then compressed to the
appropriate pressure for combustion and passed into the combustion chamber. Once in
Figure 3.2: Multi-shaft CCGT (CIGRE, 2003)
the combustion chamber, fuel is mixed with the compressed air, and combustion takes
place, thus raising the temperature of the air. The resultant hot exhaust gases are
expelled from the combustion chamber and expanded through the gas turbine where
they impart energy to the blades of the turbine, causing the turbine to rotate. The net
mechanical power output is the total mechanical power produced in the gas turbine
less the power required to drive the compressor. This net mechanical power is then
converted to electrical power in the generator.
The primary dierence between the OCGT and the gas turbine component of the
CCGT is the exhaust temperature control. While both OCGT and CCGT technolo-
gies have a maximum allowable temperature imposed by the turbine blade materials,
any variation in the temperature of the exhaust gases entering the HRSG will aect
the eciency of the steam production and, as a result, the eciency of the steam
turbine. Therefore, in order to achieve optimal eciency in CCGTs, the exhaust gas
temperature should be maintained at the maximum allowable level, even when oper-
ating at part load conditions. Conversely, in an OCGT, the exhaust temperature does
not approach maximum allowable temperature unless operating near rated gas turbine
power output.
The exhaust temperature of the gas turbine component of the CCGT is maintained at
this optimal exhaust temperature level by controlling the ows of both air and fuel.
The variable IGVs at the compressor inlet regulate the incoming airow. During gas
turbine start-up, the IGVs are positioned to ensure a smooth run up of the compressor
(avoiding stall zones) until the full-speed no-load condition is reached. In order to
maintain constant outlet temperature it is necessary to adjust the airow as the fuel
ow changes. Therefore, once at full-speed no-load, the IGVs increase from their
minimum on-load position in proportion with the admission of fuel to the combustor
in order to maintain the target exhaust gas temperature as the unit is ramped up to
the required set-point. When operating at partial loads, the IGVs remain partially
closed, to maintain the exhaust temperature at the target level. However, as the gas
turbine operating point approaches base load (100% of rated output) the IGVs are in
the fully open position, and airow cannot be increased any further.
Combined cycle gas turbines are, in general, fast responding units. When a frequency
excursion occurs on the system, the speed controller responds within a very short time
to falling system frequency and results in more fuel being injected into the combustor.
However if fuel ow increases too rapidly, the IGVs may be unable to maintain the
correct air-fuel ratio to keep the temperature of the exhaust gases below the rated
temperature. The temperature controller will therefore come into play, reducing fuel
levels to maintain the correct air-fuel ratio, and then retarding the rate of increase of
the fuel ow to a level in keeping with the rate at which the IGVs are opening.
Some CCGTs are equipped with fast acting actuators on the IGVs, signicantly im-
proving the responsiveness to frequency events (Bohrenkmper et al., 2004). On these
units, system frequency is monitored and in the event of a large frequency deviation,
an anticipated increase in outlet temperature, due to future changes in fuel ow, will
be minimised through feed-forward action suitably increasing airow.
In order to better understand the dynamic behaviour of CCGT units to frequency dis-
turbances, it is important to recognise the eects that changing system frequency will
introduce. The CCGT, like all conventional synchronous generators on the system,
will provide an inertial response proportional to both the magnitude and rate of the
frequency change. This response is vital in helping to maintain system security by alle-
viating the rate at which the frequency is falling. In addition, as the system frequency
falls, the compressor slows down, since it is synchronised to the system. A reduction
in compressor speed leads to a drop in the pressure ratio across the compressor, and
thus the airow into the combustion chamber is reduced. Consequently the pressure
ratio across the gas turbine is reduced, causing a decline in the power output.
If the unit is operating at partial load, the IGVs are not fully open. As such, it is
possible to open these IGVs further, thus osetting the reduction in airow across the
compressor. The speed at which the IGVs can respond determines how quickly this
can take eect. However, when the CCGT is operating at base (100 %) load, the
IGVs are now in the fully open position, and cannot be further adjusted to increase
the airow through the compressor. The reduction in airow from the compressor thus
causes an increase in the fuel to air ratio, which directly determines the turbine inlet
temperature, and consequentially, the exhaust temperature. Therefore, any rise in the
fuel to air ratio results in an increase in the exhaust temperature. Simultaneously, the
fuel input will tend to rise due to the droop characteristic of the speed. Therefore, as
the exhaust temperature rises, the temperature control system quickly overrides other
inputs and reduces the fuel ow to restore the correct fuel to air ratio. As such, the
power output is reduced due to the reduction in fuel ow. These phenomena can also
be signicant when a CCGT is operating slightly below base load, in order to provide
reserve for the system operator during a contingency. While the frequency remains
below nominal, the maximum output available from the CCGT will be less than its
rated maximum.
Most gas turbines used in CCGTs are heavy-duty gas turbines, consisting of a compres-
sor and turbine on a single shaft. However, aero-derivative gas turbines are sometimes
used in applications with lower power requirements (CIGRE, 2003), and have a dier-
ent conguration to that of the heavy-duty gas turbines. The compressor is on the same
shaft as the high-pressure stage of the gas turbine, which provides the power to drive
the compressor. The intermediate and low-pressure stages of the gas turbine are on a
separate shaft and the mechanical power from these is converted to electrical power in
the generator. Therefore, the compressor and high-pressure stage of the gas turbine
are eectively mechanically decoupled from the intermediate and low-pressure stages,
and the generator. As a result, the compressor of the aero-derivative gas turbine is in-
dependent of system frequency and frequency excursions below nominal frequency will
not result in reduced compressor speed, and a subsequent reduction in power output.
In addition, the range of airow control is greater than for a single-shaft gas turbine.
However, due to this decoupling of the compressor and high-pressure stage of the gas
turbine from the remaining stages of the gas turbine and the generator, the inertial
contribution of the aero-derivative gas turbine to the power system will be signicantly
lower than for a standard heavy-duty gas turbine. As heavy-duty gas turbines are the
predominant technology for CCGTs, and air-derivative gas turbines dont experience
a reduction in power with falling frequency, the focus of this study will be on the
heavy-duty type congured gas turbines.
The working uid of the gas turbine is air. Therefore, the power output of the gas
turbine is also highly dependent on the ambient atmospheric conditions of the air:
pressure and temperature (Pourbeik, 2002). The density of air is directly proportional
to the ambient air pressure. Therefore, when ambient pressure varies from rated am-
bient pressure, the airow through the compressor changes accordingly, resulting in
a change in fuel ow to maintain the rated air-fuel ratio. A change in power output
directly proportional to the change in ambient air pressure results.
Variations in the atmospheric air temperature from the ambient temperature for which
the gas turbine is rated can result in signicant changes in the maximum power out-
put achievable (Kunitomi et al., 2001; Pourbeik, 2002; CIGRE, 2003). As ambient
temperature increases, the temperature dierential across the compressor is reduced,
requiring less work to attain the rated compressor exit temperature. Consequently, the
pressure ratio across the compressor and, therefore, across the gas turbine is reduced,
leading to a reduction in the maximum power output. Simultaneously, reduced airow
will result in an increase in the air-fuel ratio, necessitating a reduction in fuel ow to
maintain rated exhaust temperature, thus further reducing power output. Therefore,
the maximum power output of the gas turbine is a complex nonlinear function which
decreases as ambient temperature increases.
Combined cycle gas turbines are primarily designed to operate at base load i.e. the
condition for maximum eciency (Hung, 2001). As the operating point is reduced
from base load, the eciency of the gas turbine is similarly reduced, leading to higher
per unit fuel costs. Both the compressor and turbine components are designed to
operate at specic rated conditions, and any deviation from these optimal conditions
will result in a decrease in their respective eciencies. Furthermore, operation of the
compressor depends on several parameters such as ambient conditions, air mass ow
and shaft speed. As the operating point of the compressor moves away from its rated
operating point, and air mass ow is reduced, the compressor surge margin (the margin
of safety between the normal compressor operating point and the stability limit) is also
reduced. Since the speed of rotation of the compressor also aects the surge margin, if
the unit is operating at partial loads and a reduction in the system frequency occurs, the
possibility of encountering problems such as compressor surge increases (Transpower,
2001).
One particular attraction of CCGT generation is the reduced emissions levels in com-
parison to conventional generation. Carbon dioxide (CO
2
) emissions per unit of energy
from CCGTs are signicantly less than from conventional thermal units burning oil
or coal. Both CO
2
and oxides of nitrogen (NO
x
) levels at rated operated conditions
are minimised. However, as the operating point moves away from optimal, the level
of NO
x
emissions per unit of power produced increases, while below approximately
60-70% of rated load, these emissions increase dramatically (Doherty et al., 2004).
Therefore, operation at or near base load is desirable to minimise emissions in addition
to maximising eciency.
3.2.2 The Heat Recovery Steam Generator and Steam Tur-
bine Components
The exhaust gases from the gas turbine enter the heat recovery steam generator, where
the energy contained within the exhaust gases is converted into steam through the use
of heat exchangers. As mentioned in Section 3.2.1, the HRSG is designed to operate
optimally at the rated exhaust gas temperature from the gas turbine, and any decline in
this temperature signicantly aects the eciency. Although the majority of CCGTs
employ HRSG with no supplementary ring (CIGRE, 2003), the burning of additional
fuel to provide extra steam may be required for some applications. The HRSG can
therefore be designed to incorporate supplementary ring in such cases.
From the HRSG the steam then enters the steam turbine component and is expanded
to produce mechanical energy, which is converted to electrical energy in the generator.
The steam turbine can consist of multiple stages. In some cases, additional heat energy
can be imparted to the steam during a reheat cycle, when the steam exiting one steam
turbine stage is routed back through the HRSG before entering the next stage.
The steam turbine component of the CCGT can be operated in two modes: sliding
pressure mode and xed pressure mode. In sliding pressure mode, the power output
from the steam turbine depends directly on the exhaust gas ow from the gas turbine
and the resultant ow of steam from the HRSG. No throttling of steam occurs at the
entrance to the steam turbine and as a consequence, the steam turbine has minimal
stored energy available. Unit response depends entirely on that of the gas turbine
component, with a signicant delay due to the HRSG before changes in the gas turbine
fuel ow result in a notable change in the power output of the steam turbine (Bagnasco
et al., 1998; Kunitomi et al., 2003). (A typical boiler time constant in the HRSG is
of the order of 300 s (Undrill and Garmendia, 2001; CIGRE, 2003).) Therefore, the
dynamics of the steam turbine component are essentially negligible for the rst 20 to
30 seconds subsequent to a change in power output of the gas turbine component.
Fixed pressure mode involves the throttling of steam at the entrance to the steam
turbine to maintain constant pressure. Therefore, a certain amount of stored energy
is available to allow limited frequency response of the steam turbine through governor
action. However, the operation of the steam turbine in xed pressure mode near
full load conditions results in a loss in overall cycle eciency. Therefore, as sliding
pressure mode is the predominant operating mode in use for CCGT plant (Undrill and
Garmendia, 2001; CIGRE, 2003), where frequency response is readily available from
the gas turbine component, this mode of operation is adopted in this study.
3.3 Literature review
As the proportion of combined cycle gas turbines worldwide increases rapidly, CCGTs
have become the focus of a signicant amount of attention. As a result an extensive
amount of literature has become available, ranging from overviews of the basic concepts
of combined cycle gas turbines to intricate details of the individual and combined
processes, characteristics and behaviour. Both the gas turbine and the steam turbine
components of the CCGT have been used individually in the electricity industry for
many years and are widely understood and well documented (IEEE, 1991; DeMello,
1991; Cohen et al., 1996). However, when the two technologies are combined with a
heat recovery steam turbine as the interface, a further level of complexity arises. A
useful high-level synopsis of the individual components of combined cycle power plants
and the possible congurations of the plant are given in Ordys et al. (1994) and CIGRE
(2003). In conjunction with details of the conguration of CCGT plant, CIGRE (2003)
also gives useful insight into the growth in the proportion of CCGTs on several power
systems worldwide and presents the main issues that dierent system operators view
as important with regards to this trend. One major concern highlighted was the need
for reliable and validated models appropriate for power system simulation studies.
The characteristics of CCGT plant that must be understood for power system studies,
including frequency regulating response and frequency dip response, are reviewed in
Undrill and Garmendia (2001), and simulated examples are presented. The steam
turbine, operating in sliding pressure mode, was determined to be a slow source of
reserve, with an overall time constant of approximately 300 s.
In Rowen (1983a), the control exibility and performance, in conjunction with the
economics of operation, of both single and two-shaft gas turbines are examined for
heat recovery applications. In order to prevent a rapid decline in eciency of the
heat recovery steam generator (HRSG) due to the exhaust temperature of the gas
turbine declining at partial loading, several methods were proposed. For single shaft gas
turbines, the use of variable inlet guide vanes to regulate the airow and thus maintain
the exhaust temperature near maximum was examined. For two-shaft gas turbines, a
combination of airow control through variable inlet guide vanes and compressor speed
control was proposed.
The dierence in performance between full and partially loaded CCGTs is examined in
detail in Kim (2004). It was found that the higher the eciency of the gas turbine at
base load, the less degradation in eciency that occurred during part load operation,
resulting in higher overall CCGT eciencies. This work also reinforces previous con-
clusions (Rowen, 1983a) that maintaining high gas turbine exhaust temperature results
in signicantly greater eciency in the steam turbine.
The dependency of the power output from a combined cycle gas turbine on system
frequency, and also the eects of ambient conditions are two very important features
that require careful consideration. The theory behind low frequency behaviour of gas
turbines is described in Pourbeik (2002), and the relationships between gas turbine
power output, system frequency and ambient conditions are examined in detail.
3.3.1 CCGT Modelling
A great deal of work has been carried out on the development of accurate dynamic
models of gas turbines, and it is this work, which generally forms the foundation for
most combined cycle gas turbine models. The basis of many gas turbine and combined
cycle gas turbine studies is the model of the gas turbine developed by Rowen (1983b,
1992). Initially, a mathematical model (Rowen, 1983b) of a gas turbine was developed
for use in power system stability studies. The control system included speed, temper-
ature and acceleration controllers, as well as limits for fuel ow. Parameter values for
GE heavy duty single shaft gas turbine models were included, and the possibility of
further simplications depending the the focus of study were also outlined.
The initial gas turbine model (Rowen, 1983b) was later expanded (Rowen, 1992) to
include the eects of modulating inlet guide vanes, along with the eects of ambient
temperature and the dependency of airow through the compressor on the system
speed. Although a reasonably simplied model, it nevertheless was considered capable
of capturing the dynamic characteristics of the gas turbine in a reasonably accurate
manner and therefore suitable for power system dynamic studies.
Bagnasco et al. (1998) incorporated the gas turbine model (Rowen, 1983b, 1992) into
a model for a combined cycle power plant, through the addition of a simplied model
of a realistic HRSG and steam turbine. A regulator model for the gas turbine com-
ponent was also developed. Subsequently, this CCGT model was used to examine the
behaviour and performance of combined cycle power plant during islanding conditions.
A CCGT model, including the addition of a PID controller for the gas turbine com-
ponent, developed with reference to Rowen (1983b, 1992) and (Bagnasco et al., 1998)
was presented in Zhang and So (2000). The response of the CCGT model to system
disturbances over a time scale of 10 seconds was examined, and the PID controller
was found to improve the dynamic performance of the system. However, the eects of
ambient operating conditions were neglected.
More complex combined cycle plant models have been developed using detailed physi-
cal relationships to represent each component individually (Ordys et al., 1994; IEEE,
1994). In IEEE (1994) the modelling requirements, and subsequently a model based
on physical principles, is presented for a sample CCGT with the purpose of providing
a starting point for future work. The model includes a single-shaft constant speed
gas turbine conguration with variable inlet guide vanes, and the speed, temperature,
acceleration and airow control loops that play a role in the dynamic performance of
the plant.
Kunitomi et al. (2001) present a gas turbine model that incorporates the frequency
dependency of the gas turbine while also recognising the impact of ambient conditions
on the performance of the unit. This detailed model is based on Rowen (1983b, 1992),
but the turbine thermodynamic calculations are replaced with new equations derived
from physical principles, while frequency dependency is estimated from the ambient
temperature dependency. Therefore, gas turbine variables need to be recorded over ei-
ther a large range of shaft speeds or over a signicant range of ambient temperatures in
order to nd parameter values for the frequency and ambient temperature dependency.
This gas turbine model was expanded to a CCGT model (Kunitomi et al., 2003) by
the incorporation of simple mathematical models of the HRSG and steam turbine and
pertinent controls. Model performance for various dynamic event scenarios was exam-
ined, and the behaviour of the gas turbine during low frequency events, as described in
Pourbeik (2002), was demonstrated. In addition, the invariance of the steam turbine
power output for up to 100 s during a low frequency event was illustrated. The model
developed Kunitomi et al. (2001, 2003) is suitable for use in long-term dynamic sim-
ulations, although one drawback with the more extensive models is the level of detail
and test data required for tuning and validation.
A comprehensive study has recently been published by CIGRE, based on work carried
out by a large committee from various industrial and academic backgrounds (CIGRE,
2003). An extensive examination of combined cycle gas turbines is presented, with
analysis of the various characteristics that set these units apart from more traditional
generating plant. A dynamic model for a CCGT is developed, based on the broad
experiences of the committee, and previously published work in this area (Rowen,
1983b; IEEE, 1994; Undrill and Garmendia, 2001; Hannett and Feltes, 2001; Hajagos
and Berube, 2001; Kunitomi et al., 2001, 2003), for use in power system simulation
studies. The frequency dependency of the gas turbine maximum power output is
represented in this model using either a look-up table or a piece-wise linear function.
While this method of representation is not based on physical principles, it provides
a simple and adequate representation for power system studies. However, extensive
manufacturer data is required to determine the dependency of maximum power output
on system frequency over a range of ambient temperatures.
The importance of having proper parameters to accurately capture the behaviour of
the CCGT, in addition to the appropriate model structures, is highlighted in a study
by Hannett and Feltes (2001). The type of CCGT plant testing required to determine
the parameter values, along with a list of signals to be measured are outlined and
discussed.
3.4 CCGTs on the Ireland System
At present, CCGT plant comprise almost 27% of the total installed generating capacity
on the island of Ireland, and this gure should continue to rise further, with over 32%
penetration expected by the end of 2005 (ESBNG, 2003; CER, 2004a). At times of
low demand, a single CCGT can comprise more than 15% of the total generation.
Therefore, the impact of the dynamic behaviour of CCGTs in response to frequency
events on the system needs to be assessed carefully. Furthermore, CCGT ratings tend to
be large, and therefore, with the addition of CCGTs to the system, the size of the largest
infeed, and, as a consequence, the size of the largest possible contingency has increased.
Severe frequency events are rare on the Ireland system with only two event in the past
ten years resulting in the unanticipated shedding of normal customers. However, as
large CCGTs replace a number of smaller conventional plant, major frequency events
may become more likely and the impact on the system needs to be studied. In particular
the impact on system integrity and the possibility of interruption to customers needs
to be assessed.
3.5 CCGT Model
As the timescale of interest in this study is up to 20 s following a system disturbance,
and all CCGTs on the Ireland system operated in sliding pressure mode, the response of
the steam turbine can therefore be considered negligible (Bagnasco et al., 1998; Undrill
and Garmendia, 2001; Kunitomi et al., 2003).
3.5.1 CCGT Model Structure
The CCGT model used in this study has been adapted from that developed by Rowen
(Rowen, 1983b, 1992) by making reference to more recent CCGT models (IEEE, 1994;
Bagnasco et al., 1998; CIGRE, 2003). An outline of the structure is shown in Fig. 3.3.
All parameters are given in per unit, with the exception of temperature and pressure,
which are measured in

C and mbar respectively.
Figure 3.3: CCGT model structure (See text for details)
The required inputs to the model are the steady-state set-point of the unit, SP, ambient
temperature, T
a
, and ambient pressure, P
a
. The eect of ambient temperature on
the rating of the gas turbine is incorporated as a correction factor, developed from
historical data, which is applied to the unit set-point input. This diers from previous
models, where ambient temperature was either neglected or incorporated inside the
model structure, for example within the exhaust temperature calculations (Rowen,
1983b, 1992). The eect of ambient pressure on the gas turbine output, which is not
addressed in some models, was also incorporated in a similar way.
There are three key control loops on a gas turbine: speed control, temperature control
and acceleration control. However, acceleration control generally only comes into play
during start-up and shut-down conditions. As the model is assumed to be initially op-
erating at steady state conditions, for simplication, acceleration control has therefore
been neglected.
Under normal operating conditions, speed control, by means of governor action, reg-
ulates the fuel supply. Governor models can vary signicantly. In this case, a simple
droop governor is implemented, consistent with the governors used on generating units
on the Irish system. The output of a droop governor is proportional to the error
between actual and nominal speed.
The outputs from both the speed and temperature controllers are fed into a minimum
selector block, such that the lesser of the two signals determines the variable fuel ow
into the gas turbine, within allowable limits. There is a no-load or xed fuel requirement
in a gas turbine in order to power the compressor, which is generally about 23 % of
maximum. The controllers regulate the variable fuel ow between this minimum no-
load point and maximum fuel ow.
In previous models (Rowen, 1992, 1983b; Bagnasco et al., 1998), it was assumed that
the fuel pump was dependent on the system frequency, and therefore fuel ow into the
gas turbine was a function of the system speed, as often the case with liquid fuel supply.
For this study, however, the fuel supply is assumed to be gaseous and independent of
frequency. Therefore, fuel ow depends only on the control signal. The delay associated
with the fuel supply when a change is fuel ow occurs are accounted for in the valve
positioner and the gas fuel system blocks in the model. The delay that occurs due to
the combustion process is represented in the combustion reaction time delay block, and
the delay associated with the discharge from the compressor is represented by a rst
order lag in the compressor discharge block, prior to the gas turbine torque calculation.
Since the gas turbine is essentially a linear device (Rowen, 1983b), the fuel ow in the
gas turbine directly determines the output torque. However if the system frequency
falls, the pressure ratio across the gas turbine will be reduced. At constant fuel ow,
the torque produced by the gas turbine depends on this pressure ratio. Therefore, the
torque produced by the gas turbine is a function of both the fuel ow (per unit), W
f
,
and the system speed (per unit), N.
Torque = 1.3 (W
f
0.23) + 0.5 (1 N) (3.1)
The total per unit fuel ow, minus the no-load fuel requirement, is scaled such that at
nominal system frequency, the torque is directly proportional to the variable fuel ow.
The torque is modulated by system frequency when a deviation from nominal system
frequency occurs. The power output of the gas turbine is subsequently determined
from the product of the torque and the system speed.
The transport delay associated with the transport of the exhaust gases from the com-
bustor through the gas turbine is incorporated prior to the turbine exhaust temperature
calculation. The calculated turbine exhaust temperature, T
x
is a function the rated
turbine temperature T
r
, fuel ow W
f
, the system speed N and the airow through the
compressor, as determined by the inlet guide vane angle, igv.
T
x
= {[T
r
453 (N
2
+ 4.21 N + 4.42) 0.82 (1 W
f
)]
+ 722 (1 N) + 1.94 (igv
max
igv)}
(3.2)
This calculation takes the inlet guide vane position into account, as applicable on a
standard gas turbine. However, the eect on the exhaust temperature of modulating
the airow by means of the inlet guide vanes to maximise the part load exhaust tem-
perature, as in the case of the gas turbine component of the CCGT, is not taken into
account. Therefore, the calculated exhaust temperature, T
x
is divided by the per unit
turbine exhaust ow (per unit), W
x
, to attain the exhaust temperature corrected to
take IGV modulation into account, T
xc
.
In reality, however, the turbine exhaust temperature is not calculated but measured,
T
xm
. This measurement process is not instantaneous. Both the radiation shield and the
thermocouple have time delays associated with them, as incorporated in the model, and
these will aect the rate at which changes in the exhaust temperature will be detected.
Airow in the gas turbine model is regulated using an IGV controller. The measured
exhaust temperature, T
xm
is compared with the rated exhaust temperature T
r
such
that if there is a dierence, the IGVs modulate the airow in order to bring the ex-
haust temperature back to the rated value, at a rate determined by the IGV actuator.
However, airow is dependent on the speed of the compressor as well as IGV position.
The airow expected by virtue of the IGV position is therefore modulated using the
actual speed, yielding the calculated airow. It is assumed in this model that the
volume of fuel ow is negligible with respect to airow, so that the volume of exhaust
gases may be assumed equal to the volume of air owing through the compressor, and
the pressure ratios are the same across both the compressor and the gas turbine (i.e.
no air is extracted for cooling). Therefore, the turbine exhaust ow (per unit), W
x
is
calculated as follows, where L
igv
is the IGV position (per unit).
W
x
= N (L
igv
)
0.257
(3.3)
The model is also equipped with an over-ring capability, which allows the temperature
limits to be increased for a short period of time during a frequency transient.
In the model illustrated in Fig 3.3, the system speed input to the turbine exhaust tem-
perature calculation (3.2) and the turbine exhaust ow calculation (3.3) incorporates a
scaling mechanism to regulate the frequency sensitivity of the unit. This was developed
to take into account the variations in frequency sensitivity conditional on technology
and age of CCGT units, which were evident on examination of frequency responses of
the dierent CCGT units. The system speed, N, is multiplied by a scalar, a, and then
a constant, b, is added, such that a+b=1.
N
c
= a N +b (3.4)
Therefore, N in equations (3.3) and (3.2) is now replaced by N
c
.
T
x
= {[T
r
453 (N
2
c
+ 4.21 N
c
+ 4.42) 0.82 (1 W
f
)]
+ 722 (1 N
c
) + 1.94 (igv
max
igv)}
(3.5)
W
x
= N
c
(L
igv
)
0.257
(3.6)
3.5.2 Model Tuning and Validation
The combined cycle gas turbine model developed in Section 3.5.1 was tuned to represent
each individual CCGT unit on the Ireland system, using available data from over fty
system frequency events that have occurred on the system over the last ve years. Post
and pre event system frequency and power output data from every unit was available,
with a sampling rate of up to 1 Hz. Technical data for each individual generating
station was available, and the ambient conditions at the time of each event were also
obtained.
In order to complete each CCGT model it was necessary to determine the eect of ambi-
ent conditions on unit output. This relationship was determined using historical hourly
data, comprising ambient conditions, IGV position and generated power. However, this
data was not available for all CCGT generating units on the system. Therefore, a per
unit approximation of the eect of ambient conditions was made based on the avail-
able historical data and applied to each gas turbine component for which information
was not available. The sensitivity of a typical gas turbine component to ambient tem-
perature is illustrated in Fig. 3.4, with power output falling as ambient temperature
increases. Although the relationship between maximum power output and ambient
temperature is generally agreed to be highly complex and non-linear (Pourbeik, 2002;
CIGRE, 2003), the temperature-power output curve obtained in this study was found
to be approximately linear within the temperature range of the available data.
In Fig. 3.5, the sensitivity of the a sample gas turbine to ambient pressure is illustrated.
The pressure-power output relationship was found to be linear, as expected (Pourbeik,
2002), with power output increasing with ambient pressure.
These sensitivities, established using the hourly historical data, were found to agree
reasonably well with available correction curves from manufacturers of the sample unit
(Siemens, 2001). The calculated sensitivities to both ambient temperature and pressure
were marginally more than those given in manufacturer data. This could be attributed
to wear and tear, and fouling of the vanes in the compressor and turbine.
Having established the above relationships, CCGT model tuning and validation was
performed by driving the model with the observed frequency trace for a particular event.
The ambient temperature and pressure correction factors were introduced as inputs to
5 0 5 10 15
10
5
0
5
Temperature deviation from rated temperature (Celsius)
C
h
a
n
g
e

i
n

G
a
s

T
u
r
b
i
n
e

P
o
w
e
r

O
u
t
p
u
t

(
%
)
Figure 3.4: The change in maximum rated power output of a typical CCGT gas
turbine component for temperature deviations from rated ambient temperature.
the model. The simulated response, incorporating both the ambient temperature and
pressure corrections, was then compared to the actual response of the unit. Three
years of frequency event data was used in most cases to tune the CCGT units, with
the remaining two years being used to validate the tuned models.
The droop characteristic of each unit was calculated from steady state frequency, as
described in Section 2.5.1. Most droop characteristics of the CCGT generating units
on the system were found to be close to the required droop of 4%, with the range of
values found varying between 4% and 5%.
The speed controller (governor model) and the frequency dependency of each model
was subsequently tuned to measured data. Parameter values in the exhaust tempera-
ture loop, the fuel ow loop and the IGV controller were taken from literature Rowen
(1983b), IEEE (1994) and CIGRE (2003). However, the IGV controller parameters for
two CCGT models were altered from literature values, as these units were equipped
with fast-acting IGVs. Therefore, these parameter values were based on the manufac-
turer time constants.
Fig. 3.6 illustrates such a comparison between the actual power output of a CCGT and
50 25 0 25 50
6
4
2
0
2
4
6
Pressure deviation from rated pressure (mbar)
C
h
a
n
g
e

i
n

G
a
s

T
u
r
b
i
n
e

P
o
w
e
r

O
u
t
p
u
t

(
%
)
Figure 3.5: The change in maximum rated power output of a typical gas turbine for
pressure deviations from rated ambient pressure.
the corresponding simulation model. It can be clearly seen that when frequency eects
are included in the CCGT model that a much better representation of the actual power
output is achieved, despite the limited resolution of the actual data.
The frequency dependency of CCGTs on the system was found to vary both with
age and technology. The system speed input to the turbine calculation of exhaust
temperature T
x
and the turbine exhaust ow W
x
has been modied as illustrated in
Fig. 3.3 to incorporate a scaling mechanism to regulate the frequency sensitivity of
the unit. Improvements in CCGT technology are on-going, and this is evident in
the improved responses of newer units. However, many CCGTs have already been
installed in electricity systems worldwide, with varying degrees of technology and,
unless modied, their responsiveness will depend on the technology at the time of
installation.
A small number of CCGTs on the Ireland system incorporate fast acting inlet guide
vanes, and these have been represented in the model by considerably reducing the time
constant on the IGV actuator. This was seen to capture the characteristics of such fast
acting IGVs to a level of accuracy that the inclusion of an additional control loop to
Figure 3.6: Change in power output (including the inertial response) of a typical
near base loaded CCGT in response to a frequency event on the system. (A) System
frequency, (B) (a) Response of actual CCGT unit, (b) Simulated response of CCGT
model, and (c) Simulated response of CCGT model with frequency dependency of
model removed.
represent them was not considered necessary. The signicant role of fast acting IGVs
can be better understood by examining Fig. 3.7, which considers the response of a
base loaded CCGT to a 0.5 Hz frequency drop compared to the response of the same
CCGT when operating at 95 % of base load. When operating at base load, the power
output from the CCGT declines with the system frequency, and remains below rated
output as long as the frequency is less than nominal. When operating at partial loads,
however, the CCGT response to changing frequency is rapid, and the inuence of fast-
acting IGVs is clearly illustrated. In both cases, the power output initially increases
in response to falling frequency. With conventional IGVs, however, as the fuel ow
increases, the exhaust temperature will increase causing the temperature controller to
reduce the fuel ow and hence the power output of the gas turbine. The temperature
controller then only allows the fuel ow to increase as the airow increases, so the
response of the unit is limited by the operational speed of the IGVs. With fast-acting
IGVs, the airow increases much more quickly, and so the response of the CCGT is
no longer constrained in the same way. The maximum response is instead curtailed by
the reduction in airow due to reduced compressor speed.
0 5 10 15 20 25 30
49.4
49.6
49.8
50
F
r
e
q
u
e
n
c
y

(
H
z
)
0 5 10 15 20 25 30
98.5
99
99.5
100
P
o
w
e
r

O
u
t
p
u
t

(
%
)
0 5 10 15 20 25 30
96
98
100
Time (s)
P
o
w
e
r

O
u
t
p
u
t

(
%
)
(a)
(b)
(i)
(ii)
(iii)
Figure 3.7: Simulated power output of the GT component of a typical CCGT to a
frequency drop of 0.5 Hz (inertial response neglected). (i) Frequency trace (ii) the
CCGT operating at base load and (iii) the CCGT operating at 95 % capacity with (a)
conventional IGVs and (b) fast-acting IGVs
3.6 The Impact of CCGT Dynamics on Frequency
Control
Three dierent operating scenarios are examined. These scenarios are carefully chosen
to represent the most extreme situations that could occur on the Ireland electricity
system.
Winter Peak (WP)
This is the peak load occurring on the system, and occurs during a winter business day.
In order to meet this peak load, a substantial amount of generating plant on the system
is online, and system inertia is near maximum. Therefore, from an inertial response
perspective, this case represents a best case scenario.
Summer Night Valley (SNV)
This is the minimum load on the Ireland system, and occurs during a summer night.
System inertia is very low at this point, and from an inertial response perspective this
case represents a worst case scenario.
Summer Day Valley (SDV)
This represents the minimum daytime load, which occurs during a summer non-business
day. While the system inertia is not as low as in the SNV case, the frequency response
may be signicantly dierent as the SNV case has substantial static reserve response.
For each of the three scenarios described above, two cases are examined. The system is
dispatched based on a merit order to meet demand and reserve requirements, with all
dispatched CCGTs operating at base load. In the rst case, the frequency dependency
of all dispatched CCGTs is removed. Consequently, the power output from the base
loaded CCGTs is constant, with the exception of the inertial response. For the second
case system dispatch is identical, with unchanged system inertia, the only dierence
being the frequency dependency in the CCGT models. In each case, the same gener-
ating units, and interruptible load service if appropriate, provide the primary reserve
requirement.
For each scenario, the most serious single disturbance to the system is examined, which
involves tripping the largest infeed to the system. During the winter peak, this is likely
to be 422 MW, while during the summer night valley and summer day valley scenarios,
this gure is reduced to 400 MW, accounting for the reduction in power output of this
largest infeed, a CCGT, due to a higher ambient temperature. System operators on the
island of Ireland do not consider multiple independent events occurring concurrently.
Following an event system operators restore the system back to a secure operating
state as quickly as possible. Therefore N-2 contingencies are not considered here and
the tripping of the largest infeed is considered the worst case scenario. The Ireland
electricity system has never experienced the loss of transmission capacity during a low
frequency event, therefore cascading events caused by tripping of transmission lines
during the loss of the largest infeed are not considered.
For each of the three considered scenarios, the eect on system frequency following the
loss of the largest generating infeed to the system was modelled. For the winter peak
scenario, illustrated in Fig. 3.8, the load is 6065 MW and CCGT generation comprises
33.5% of the total.
During the winter peak, typically between 17.00 and 19.00 hours, there is no interrupt-
ible load service on the system, and the primary reserve requirement must be entirely
met by generator spinning reserves.
It can be seen that the frequency nadir falls a further 15% when all CCGT generation
are operating at base load and the frequency dependency of their response is modelled,
demonstrating that a small drop in the power output, due to frequency, of a single
CCGT in isolation becomes more signicant as the number of base loaded CCGTs
increases.
The load on the system for the summer night valley scenario is set at 2170 MW, with
49.6% of generation provided by CCGTs. There is no interruptible load service on
the system and three 73 MW pumped storage units are operating in pumping mode,
thereby providing almost instantaneous reserve when triggered by falling frequency.
The remaining primary reserve is met by spinning reserves from online (mainly thermal)
generators. The loss of a 400 MW generator during the summer night valley is shown
in Fig. 3.9.
0 5 10 15 20 25 30
48.8
49
49.2
49.4
49.6
49.8
50
Time (s)
F
r
e
q
u
e
n
c
y

(
H
z
)
No frequency dependency
Frequency dependency included
Figure 3.8: Winter peak scenario with 422 MW trip, when base loaded CCGTs com-
prise 33.5% of generation
In this case, when system inertia is very low, it can be seen that the system frequency
falls very fast, with an initial rate of change of frequency of 0.495 Hz/s, and frequency
drops below 49 Hz. The frequency nadir falls by an additional 5.4% when the CCGTs
are base loaded and the frequency dependency of their response is modelled, if load
tripping is not incorporated. This additional fall in frequency is critical when the
system is already severely stressed. The frequency falls below the upper threshold
for under-frequency load shedding of 48.9 Hz, leading to 8% of load being shed, as
illustrated in Fig. 3.9. This summer night valley scenario with low system load and
inertia indicates that with additional CCGTs on the system events of this nature will
be more likely and the transmission system operators on the island of Ireland should
review their frequency control strategies in the future so as to avoid the shedding of
customers.
However, the dierence between the two frequency nadirs is not as large as for the
winter peak load scenario. This illustrates that although the percentage generation
provided by base loaded CCGTs has grown from 33.5% to 49.6% penetration, the
0 5 10 15 20 25 30
48.8
49
49.2
49.4
49.6
49.8
50
50.2
Time (s)
F
r
e
q
u
e
n
c
y

(
H
z
)
(a)
(b)
(c)
Figure 3.9: Summer night valley scenario with 400 MW trip, when base loaded CCGTs
comprise 49.6% of generation. (a) all CCGTs have frequency dependency removed, (b)
all CCGTs have frequency dependency included, with no load tripping on the system
and (c) all CCGTs have frequency dependency included, with load tripping of 8% of
system demand at 48.9 Hz.
magnitude of generation provided by these units is less, and therefore the eect on
system frequency is smaller.
Finally, of the 3150 MW of generation on the Ireland system for the summer day
valley scenario, 34.2% consists of CCGT generation. During this scenario, 60 MW of
the primary reserve on the system is in the form of interruptible load service, which
is activated if the frequency falls to 49.3 Hz. The results of the summer day valley
scenario simulation are presented in Fig. 3.10. The frequency nadir falls 6.3 % further
when the CCGT generating units are base loaded and the frequency dependency of
their response is modelled.
In Fig. 3.11, the eect on the frequency nadir of increasing proportions of base loaded
CCGT generating units, with the frequency dependency of their response modelled, is
illustrated for a typical winter day load, with 60MW of interruptible load service and
0 5 10 15 20 25 30
48.8
49
49.2
49.4
49.6
49.8
50
Time (s)
F
r
e
q
u
e
n
c
y

(
H
z
)
No frequency dependency
Frequency dependency included
Figure 3.10: Summer day valley scenario with 400 MW trip, when base loaded CCGTs
comprise 34.2 % of generation.
the remaining 256 MW of spinning reserve requirement met by thermal generators. It
is shown to be non-linear, with the eect becoming more pronounced as the magnitude
of base-loaded CCGTs increases. This exponential fall in frequency as more CCGTs
are added is due to frequency falling further, which causes unit outputs to fall to a
greater extent, in turn causing the frequency to fall further.
The technical problems of operating a CCGT at maximum output may be overcome by
a number of methods. Over-ring is one option, but is currently not allowed for units
on the Ireland system. De-rating the CCGT is also possible, but the overall eciency
of the unit will be reduced. Other options include increasing the primary operating
reserve requirement, to oset the negative power response of base loaded CCGTs, or
reducing the size of the largest infeed to the system, thus limiting the largest possible
single contingency. However, a signicant cost can be associated with each solution,
and detailed cost/benet analysis will be required to quantify and compare these costs,
which is not undertaken here.
25 30 35 40 45 50
10
15
20
25
Base Loaded CCGTs as a percentage of total generation (%)
P
e
r
c
e
n
t
a
g
e

I
n
c
r
e
a
s
e

i
n

f
r
e
q
u
e
n
c
y

n
a
d
i
r

(
%
)
(a)
(b)
Figure 3.11: Sensitivity of system frequency nadir to increasing proportions of CCGTs
(a) The percentage increase in the frequency nadir as the proportion of base loaded
generation increases and (b) the percentage increase if the increase was linear.
3.8 Conclusions
A CCGT model incorporating frequency dependency, ambient conditions dependency
and fast acting inlet guide vanes is developed and tuned. This CCGT model is incor-
porated into the validated system model to study the eect of base-loaded CCGTs on
system frequency during a frequency disturbance event. Results indicate that in the
event of a power imbalance, the decline in active power from base loaded CCGTs results
in a greater frequency excursion from nominal frequency. It was found that this eect
increases non-linearly as the number of base loaded CCGTs on a system increases.
Therefore, if the proportion of base-loaded CCGTs continues to rise, it will result in a
detrimental eect on minimum system frequency during a frequency disturbance event,
increasing the likelihood of under frequency load shedding. As such, the system oper-
ators may need to review system reserve strategies with increasing proportions of base
loaded CCGT to prevent a decline in system security.
Chapter 4
Frequency control and Wind
Turbine Technology
4.1 Introduction
The electricity industry worldwide is turning increasingly to renewable sources of en-
ergy to generate electricity. Environmental concerns about fossil fuelled conventional
generators, the desire to increase the diversity and security of fuel supply, and increas-
ing fossil fuel costs are all motivating factors behind this upward trend. Global targets
for the reduction of carbon dioxide (CO
2
) and other greenhouse gases have been intro-
duced (UN, 1997) and separate targets for minimum quantities of electricity generated
from renewable energy sources have also been established in many parts of the world
(EU Directive:77, 2001). Wind is the fastest growing and most widely utilised of the
emerging renewable energy technologies in electricity systems at present, with a total
of approximately 40 GW installed worldwide at the beginning of 2004 (EWEA, 2003).
However, the majority of wind resources are still untapped, and the potential for en-
ergy from wind generation is vast. Constant advances in technology and the relatively
low capital costs when compared with other forms of renewable energy are signicant
contributing factors to the ongoing rapid growth in the proportion of electricity being
generated from wind.
The ease with which wind generation is integrated into existing electricity systems de-
pends on a number of factors, both technical and regulatory (EWEA, 2003). Technical
Chapter 4. Frequency control and Wind Turbine Technology 82
aspects that need to be considered include the amount and location of wind generation,
wind turbine technology and the size and characteristics of the electricity system. Reg-
ulatory issues such as compliance with existing system operator rules and regulations
also arise. Of the technical aspects that need to be considered, system inertia plays an
extremely important role as it determines the sensitivity of system frequency to supply
demand imbalances. The lower the system inertia, the faster the frequency will change
if a variation in load or generation occurs. Large interconnected electricity systems
generally have sizeable system inertia and large frequency deviations from nominal are
rare. However, frequency deviations on small, isolated electricity systems, when they
occur, tend to be more sizeable. When connecting wind turbines to small isolated
power systems their contribution to system inertia must be considered. The addition
of synchronous generation to a power system intrinsically increases the system inertial
response. This intrinsic increase does not necessarily occur with the addition of wind
turbine generators due to their diering electromechanical characteristics. Therefore,
the displacement of conventional synchronous generation with wind generation may
result in an erosion of system inertial response resulting in increased rates of change of
frequency (ROCOF) and larger frequency excursions. In small isolated systems these
phenomena are particularly challenging due to low system inertia. Where wind tur-
bines are found to severely impact on system inertial response, system operators need
to consider altering their frequency control strategies to avoid large rates of change of
frequency and/or large frequency excursions.
Ireland is experiencing a rapid increase in wind generation, and has recently completed
the development of transmission and distribution grid codes for wind (CER, 2004b,c).
The main issues that are addressed, similar to other wind grid codes developed, are
the specic requirements for frequency, voltage and fault ride through behaviour (CER,
2004b; ELTRA, 2000).
4.2 Wind Generation Technology
Wind turbine generators convert the kinetic energy in the wind into electrical energy
for supply to the power system. The aerodynamic power captured by the turbine rotor
P
aero
is a function of air density, , swept area of wind turbine, A, the power coecient,
C
p
and wind speed, u and is given by equation 4.1
P
aero
=
1
2
AC
p
u
3
(4.1)
The power coecient C
p
is dependent on the tip speed ratio which is given by
equation 4.2, where
t
is turbine rotor speed and R
r
is the rotor radius.
=

t
R
r
u
(4.2)
A typical relationship between C
p
and is illustrated in Fig. 4.1 (Mullane, 2004).
Clearly, maximum C
p
occurs for a single value of tip speed ratio. In order to maximise
energy capture, the tip speed ratio needs to remain at optimal tip speed ratio. There-
fore, if turbine rotor speed
t
can change with wind speed, optimal tip speed ratio can
be maintained, which will result in optimal energy capture.
Figure 4.1: Typical C
p
curve
Wind turbine generators (WTGs) generally fall into one of two fundamental categories
derived from the energy convertor systems employed to convert the captured aerody-
namic power of the wind turbine to electrical power. These two categories are xed
speed and variable speed.
A squirrel-cage induction generator (SCIG) is employed as the energy converter system
for a xed speed generator. The wind turbine is connected through a gear box to the
generator, which in turn is connected directly to the power system. Although known
as xed or constant speed WTGs, the induction machine slip actually varies slightly
with power generated. However, these slip variations are only of the order of 1 - 2%
(Mullane, 2004). During a frequency excursion, the relationship between the system
frequency and the electromagnetic torque of any induction machine will determine
the inertial response. As there is a strong coupling between the squirrel cage induction
generator stator and the power system, and due to the low nominal slip, any deviations
in system speed will result in a change in rotational speed. This linking of rotor speed
with system speed gives rise to an inertial response from the xed speed WTG when
the system frequency falls.
Induction machines generally consume reactive power, as the rotor requires reactive
current for magnitisation. Therefore, capacitor banks for power factor correction
are required and these are generally rated at approximately 30% of WTG capacity
(Holdsworth et al., 2003). The conguration of a xed speed wind turbine is illus-
trated in Fig. 4.2.
Figure 4.2: Fixed speed wind turbine Generator
The xed speed WTG conguration generally has greater simplicity and reliability than
WTGs with variable speed capability, and has the advantages of being low-cost and
robust. However, as speed variations are extremely limited, optimal energy capture
only occurs at a single wind speed and xed speed WTGs oer limited energy capture
capabilities when wind speeds vary from this rated wind speed. In addition, signicant
wear and tear on the wind turbine occurs due to the large torque variations transmitted
in the drive train during gusts of wind, and over-design of the drive train is commonly
used to oset this eect. One enhancement of the xed speed wind turbine is the
addition of a distinct second slower operating speed. At low wind speeds, the slower
operating speed is chosen to improve the rotor eciency and reduce noise (Manwell
et al., 2002).
Variable speed WTGs oer a number of benets including potential operation at maxi-
mum eciency over a wide range of wind speeds, resulting in increased energy capture,
and also a reduction in wear and tear on the wind turbine (Manwell et al., 2002).
The speed variation range of a variable speed WTG is dependent on the technology
employed. Speed variations of up to 10% can be achieved using a wound rotor induction
generator with an electrically controlled resistance in series with the rotor resistance.
While this narrow band variable speed conguration has limited speed variability, it
has some of the advantages of variable speed WTGs such as the absorption of wind
gusts without the costly addition of a power electronics converter.
The doubly fed induction generator (DFIG) is a limited range variable speed WTG,
which employs a power electronics converter in addition to a wound rotor induction
generator to achieve variable speed operation. The generator rotor is connected to
the power system through a back-to-back AC/DC/AC converter, while the stator is
connected directly to the power system, as illustrated in Fig. 4.3.
Figure 4.3: Doubly fed induction generator
The net power output from the DFIG is the sum of the power outputs from both the
stator and the rotor (Pourbeik et al., 2003; Ekanayake et al., 2003). The use of the
back to back AC/DC/AC converter enables variable speed operation by decoupling the
power system electrical frequency and the rotor mechanical frequency, and the extent of
the speed variation is determined by the power electronics converter rating (Mullane,
2004). During a system frequency excursion, any inertial response provided by the
DFIG depends on the relationship between the electromagnetic torque of the machine
and system frequency. This relationship, in turn, depends upon the type of controllers
used in the converter, and the parameters of the controllers (Mullane and OMalley,
2005). The benets of DFIG technology include increased energy capture, and limited
reactive power, although the addition of the power electronics converter and the slip
rings for the wound rotor increase installation and maintenance costs.
An alternative conguration to those presented above is one with a power electronics
converter attached to the generator stator. This eectively decouples the generator
frequency from the power system frequency, and both synchronous and squirrel cage
induction generators can run at variable speed. However, as the stator is isolated from
and, consequently, unaected by any changes in the frequency of the power system,
the power output from the WTG does not change and no inertial response is obtained
during a frequency event. The power electronics converter between the generator and
the power system results in reduced drive train torque and full control of the power
factor in addition to optimal aerodynamic energy capture. However, these advantages
are counteracted by the high expense of the power electronic converters, which needs
to be rated at wind turbine rating.
4.3 Wind Turbine Generator Modelling
4.3.1 Fixed Speed Wind Turbine Model
Although not commonly used for electricity generation until recently, the operation and
behaviour of induction machines in power system studies is well documented due to
induction machines comprising a large proportion of many power system loads. There-
fore, a number of induction machine models of varying degrees of complexity have been
developed and evaluated. The most commonly used model is a fth order dq model
(Kundur, 1994), in which the machine equations are written in terms of the phase
variables and then simplied through the appropriate transformation into components
along rotating axes, where the reference frame is rotating at synchronous speed. In
this model, saturation, hysteresis and eddy currents are neglected and ux waves are
assumed to have a purely sinusoidal distribution. When tested with perturbations
to supply frequency, shaft torque and voltage magnitude, this fth order model was
found to accurately predict rotor speed, electrodynamic torque, both active and reac-
tive power and also stator current responses (Thiringer and Luomi, 2001). However,
while lower order models oer the advantages of simplicity, accuracy was found to be
reduced with reduction in the order of the model. Although simplied models were
found suitable for the analysis of certain disturbances (Thiringer and Luomi, 2001),
the fth order model is the model most commonly used for the analysis of induction
generator based wind turbines in power system simulation studies (Holdsworth et al.,
2003; Slootweg et al., 2003; Mullane and OMalley, 2005). The fth order induction
machine model described in Kundur (1994) is used in this study.
The basic induction machine equations in the dq reference frame are outlined in equa-
tions (4.3), (4.4),(4.5) and (4.6). The stator and rotor voltages in terms of the d and
q components are:
v
ds
= R
s
i
ds

qs
+p
ds
v
qs
= R
s
i
qs
+
ds
+p
qs
v
dr
= R
r
i
dr
(
r
)
qr
+p
dr
v
qr
= R
r
i
qr
+ (
r
)
dr
+p
qr
(4.3)
The stator and rotor ux linkages in terms of the d and q components are:
ds
= L
s
i
ds
+L
m
i
dr
qs
= L
s
i
qs
+L
m
i
qr
dr
= L
r
i
dr
+L
m
i
ds
qr
= L
r
i
qr
+L
m
i
qs
(4.4)
where
v
ds
, v
qs
are the stator voltages in terms of the d and q components
v
dr
, v
qr
are the rotor voltages in terms of the d and q components
i
ds
, i
qs
are the stator currents in terms of the d and q components
i
dr
, i
qr
are the rotor currents in terms of the d and q components
R
s
, R
r
are the stator and rotor phase resistances
ds
,
qs
are the stator ux linkages in terms of the d and q components
dr
,
qr
are the rotor ux linkages in terms of the d and q components
L
s
, L
r
are the per phase stator and rotor inductances
L
m
is the per phase mutual inductance
s
is the stator eld angular velocity (i.e. synchronous speed)
r
is the rotor electrical angular velocity
is the dq reference frame angular velocity
s is slip, where
r
= (1-s)
s
p is the dierential operator
The electromagnetic torque, T
em
, in terms of the rotor is given by
T
em
=
3P
f
4
(
qr
i
dr

dr
i
qr
) (4.5)
where
P
f
is the number of machine poles
The mechanical torque produced by the wind turbine rotor, T
mech
acts to accelerate
the rotor, while the electromagnetic torque T
em
acts as to decelerate the rotor. The
relationship between T
em
and T
mech
is given in equation 4.6.
T
mech
T
em
= J
d
m
dt
(4.6)
where
J is the polar moment of inertia of the rotor and wind turbine
m
is the rotor mechanical angular velocity, where
m
=
2
P
f
r
The basic induction machine equations presented in equations (4.3), (4.4),(4.5) and
(4.6) are suitable for the modelling of dierent congurations of the induction machine.
In the case of a xed speed wind turbine generator, a squirrel cage induction generator
is employed. In the squirrel cage conguration the rotor windings are short circuited
and rotor voltages are zero. Thus, v
dr
= v
qr
= 0.
The parameter values for the xed speed wind turbine were taken from Thiringer
and Luomi (2001) and Mullane (2004) and an inertial constant of 3.5 s is assumed
(Ekanayake et al., 2003).
4.3.2 DFIG Wind Turbine Model
The DFIG wind turbine employs a wound rotor induction generator in conjunction with
a power electronics converter with the appropriate controller system to attain variable
speed operation. Therefore, in order to understand the behaviour of the DFIG during
power system simulations, appropriate modelling of the converter and controllers is
required in addition to the generator. As in the case of the xed speed WTG model,
the fth order dq induction machine model (Kundur, 1994) is employed. However, for
the wound rotor conguration of the DFIG, the rotor windings are not short circuited
and therefore v
dr
and v
qr
are non-zero.
Speed control is attained through the selection of an appropriate controller. One preva-
lent form of controller conguration for AC induction machines is eld orientated con-
trol, initially developed by Blaschke (1972). In eld orientated control, selection of an
appropriate reference frame allows electromagnetic torque to be controlled independent
of the ux. When applied to induction motors with squirrel cage congured rotors, the
eld orientated controller attains electromagnetic torque control through control of the
stator current (Blaschke, 1972; Hur et al., 2001), which is divided into separate elec-
tromagnetic torque and ux components through transformation into the appropriate
reference frame .
Control of electromagnetic torque allows for the control of the speed, the basis of
the variable speed nature of the DFIG wind turbine. However, the power electronics
converter of the DFIG is connected to the rotor windings through slip rings, unlike
in the squirrel cage conguration. Therefore, for use in conjunction with wound rotor
induction generator of the DFIG, independent control of the electromagnetic torque
and stator reactive power is achieved through control of the rotor current (Tapia et al.,
2003; Mullane and OMalley, 2005). The eld orientated controller (FOC) model of
Mullane and OMalley (2005) is implemented for the controller component of the DFIG
model in this thesis.
The electromagnetic torque of equation 4.5 can be rewritten in terms of the stator
uxes and the rotor currents to give equation 4.7
T
em
=
3P
f
4
L
m
L
s
(
qs
i
dr

ds
i
qr
) (4.7)
If a reference frame is chosen such that
ds
= 0 (Mullane and OMalley, 2005), then
the electromagnetic torque becomes a function of a single variable, the rotor current
i
dr
, as shown by equation 4.8.
T
em
=
3P
f
4
L
m
L
s
(
qs
i
dr
) (4.8)
Therefore, in order to control electromagnetic torque, control of i
dr
is necessary. Rewrit-
ing the induction machine voltage equations 4.3 in terms of rotor currents and stator
uxes, and setting
ds
to zero by the choice of reference frame yields
v
ds
=
RsLm
Ls
i
dr

qs
v
qs
=
RsLm
Ls
i
qr

p +
Rs
Ls
qs
v
dr
=
R
r

L
Ls
p
i
dr
+ (
r
)
L
Ls
i
qr
(
r
)
Lm
Ls
qs
v
qr
= (
r
)
L
Ls
i
dr
+
R
r

L
Ls
p
i
qr

Lm
Ls
p
qs
(4.9)
In order to attain the reference frame where
ds
= 0, the angular velocity of the
reference frame is calculated from equations (4.9) to be
=
R
s
L
m
qs
L
s
i
dr

v
ds
qs
(4.10)
The control of i
dr
may also be simplied considerably through the careful design of the
d and q axes voltages, such that
v
dr
= v
dr
(
r
)
L
m
L
s
qs
+ (
r
)
L
L
s
i
qr
(4.11)
and
v
qr
= v
qr
+
L
m
L
s
p
qs
(
r
)
L
L
s
i
dr
(4.12)
where v
dr
and v
qr
are auxiliary signals in the controller reference frame and are the
outputs from the d-axis and q-axis proportional integral current controllers,
v
dr
= (i
dr
i
dr
)
K
1P
+
K
1I
p
(4.13)
v
qr
=
qr
i
qr
K
2P
+
K
2I
p
(4.14)
where K
1P
, K
1I
, K
2P
and K
2I
are the proportional and integral gains for the d and q
axes current controllers respectively, and i
dr
and i
qr
are the reference d and q axes rotor
currents. A more detailed description of equations (4.11-4.14) is provided in Thiringer
and Luomi (2001).
Figure 4.4: DFIG model with FOC controller
Therefore, the complete model of the DFIG wind turbine consists of the standard fth
order induction machine dq model (equations (4.3)-(4.6)) and the FOC, modelled using
equations (4.11)-(4.14) as illustrated in Fig. 4.4. This model, combined with the system
model described in Chapter 2, can be used to examine the inuence that increasing
proportions of DFIG wind turbines will have on system frequency control.
Wind model input assumptions vary from constant torque to constant power (Ekanayake
and Jenkins, 2004; Tapia et al., 2003). The frequently made assumption of constant
torque means any changes in shaft speed will result in a change in captured aerody-
namic power. The constant power assumption is also sometimes applied. The captured
aerodynamic power is given by equation (4.1), where the power coecient, C
p
is a func-
tion of shaft speed, w
s
, wind speed, u, and blade pitch angle, . The motivation behind
the constant power assumption is that C
p
remains constant for any changes in shaft
speed,
s
. This is achieved by varying the blade pitch, . The eect on the output
of the WTG models is examined for the both the assumption of constant power and
constant torque.
4.4 Wind Generation on the Ireland Electricity Sys-
tem
At present, the installed wind capacity on the Ireland system of approximately 500
MW (ESBNG, 2005a; SONI, 2005) is rapidly increasing and is expected to reach 1,000
MW within the next three years. In order to meet the targets of 13.2% of all electricity
in the Republic of Ireland and Northern Irelands proportion of the UK target of 10%
of all electricity to come from renewable sources by 2010 (EU Directive:77, 2001),
a minimum of 1,100 MW of wind generation is required on the Ireland system by
2010. This signies a much more rapid increase in wind generation than predicted
system load over the next ve years, resulting in a large increase in the proportion of
wind generation on the Ireland system. However, if the rate of installation of wind
generation continues to follow current trends on the Ireland system, the capacity of
wind generation in 2010 could easily exceed that required to meet the EU targets.
4.4.1 Scenarios
The predicted Ireland electricity system of 2010 is examined in this study to investigate
the impact of increasing wind capacity on frequency control. The validated system
model of Chapter 2 has been extended to take into account new generating plant
coming onto the system (ESBNG, 2003, 2004a; SONI, 2003b). The system load is
increased in line with predicted values. The sources of generation in the predicted 2010
electricity system are assumed to be similar to the current system, although a number
of older thermal units will be decommissioned, and all new generation connected to
the system, with the exception of wind generation, will be either combined cycle gas
turbines or open cycle gas turbines. Additional generation on the system is assumed to
operate with a droop characteristics of 4% and with the same characteristics of current
generation of the same technology. Transmission eects are neglected in this study, as
it is assumed that the transmission system will be adequately enhanced for additional
conventional generating plant and wind generation connecting to the system. It is also
assumed that by 2010 an additional HVDC interconnector with a 500 MW capacity will
link the Ireland system to Wales, and that both this interconnector and the current
HVDC interconnector with Scotland will be capable of providing limited frequency
response (SEI, 2004) of up to 50 MW. When available, it is modelled as static reserve,
which is triggered when the frequency reaches 49.5 Hz. For 2010, it is assumed that
the largest infeed and the POR remain at 422 MW and 317 MW respectively.
The system is examined for increasing installed wind capacities under three dierent
scenarios: winter peak, summer night valley and summer day, as outlined in Section
3.6. However, in this case, the predicted WP, SNV and SD scenarios for the 2010
Ireland electricity system are used instead of current scenarios, which were used in
Section 3.6.
A range of wind penetration levels from 0 to 2000 MW was examined for each sce-
nario. Due to insucient wind generation data in existence for the Ireland system,
analysis of the likelihood of dierent wind generation levels coinciding with the three
scenarios above was not possible. Therefore, no assumption about the coincidence of
any particular wind penetration with dierent load levels is made.
4.4.2 Simulating Procedure
A merit order dispatch is employed in the system model, determined using historical
dispatch data and forecast dispatches for future years (ESBNG, 2004a). Initially, the
system model is in steady state, with generation and load balanced, and frequency
constant at 50 Hz. At time t=1 s, the largest infeed to the system model, a 422
MW generator, is tripped, resulting in a power imbalance. System frequency and the
resultant response of each generator are simulated for 20 s subsequent to the trip-
ping. For each scenario outlined in Section 4.4.1, the impact of this disturbance on the
system frequency for increasing wind penetration and for dierent wind turbine tech-
nologies (xed speed and DFIG WTGs) is examined. It is assumed that the capacity
of wind generation on the system displaces an equivalent amount of conventional gen-
eration, and that this wind generation is not constrained downwards. The conventional
generation providing POR within each scenario remains the same regardless of wind
penetration.
4.5.1 Response of wind turbine technologies to system fre-
quency deviations
The simulated system frequency trace resulting from the loss of the largest infeed during
the WP scenario is illustrated in Fig. 4.5 (I). The frequency falls to 49.58 Hz, reaching
the nadir or minimum frequency approximately 3.5 s after the start of the event.
0 5 10 15
49.6
49.7
49.8
49.9
50
F
r
e
q
u
e
n
c
y

(
H
z
)
(I)
0 5 10 15
1
0
1
2
3
4
(II)
Time (s)
C
h
a
n
g
e

i
n

P
o
w
e
r
(
%

o
f

C
a
p
a
c
i
t
y
)
(a)
(b)
(c)
(d)
Figure 4.5: (I) Simulated system frequency resulting from the loss of the largest infeed
during WP. (II) Comparison of xed speed WTG and DFIG WTG responses to the
low frequency event in (I): Change in power output (%) of (a) a synchronous machine
with inertial constant of H=4.2 s, (b) the xed speed WTG, assuming constant power,
(c) the xed speed WTG, assuming constant torque and (d) the DFIG, for constant
power and constant torque.
The inertial responses of a conventional synchronous generator, with locked governor
to illustrate the inertial response only (for Section 4.5.1 only), an induction machine
Table 4.1: Comparison of inertial response from various generators
Generator Type Max. Change Time to
in Power (%) Max. (s)
Conventional Synchronous 4.03 0.80
Fixed Speed WTG, constant power 1.35 2.50
Fixed Speed WTG, constant torque 1.20 2.20
DFIG WTG, constant power 0.01 2.50
DFIG WTG, constant torque 0.01 2.20
based xed speed wind turbine generator and a DFIG based wind turbine generator to
this frequency event are shown in Fig. 4.5(II) and summarised in Table 4.1. The iner-
tial constant of the synchronous generator is taken to be 4.2 s, in comparison with 3.5 s
(Ekanayake et al., 2003) for the xed speed WTG inertial constant, since conventional
synchronous generators generally have higher inertial constants than WTGs. It can
be seen that the conventional synchronous generator produces the maximum inertial
response in the shortest time. The response of the induction machine based xed speed
wind turbine is slower and lower, due to the reduced coupling of induction generator
rotational speed to system speed compared with the synchronous generator and the
smaller inertial constant. The assumption of either constant torque or constant power
has a small impact on the maximum inertial response, with models assuming constant
power producing a larger response. The DFIG WTGs show negligible inertial response,
for both constant power and constant torque, as the DFIG controllers eectively de-
couple the generator rotational speed from the system speed (Mullane and OMalley,
2005). From a system frequency control perspective, the constant torque assumption is
the slightly more conservative approach, and was therefore assumed in the subsequent
results (Sections 4.5.2 and 4.5.3).
4.5.2 System frequency control with increasing wind penetra-
tion
The impact of increasing proportions of wind generation on the system frequency dur-
ing a variety of low frequency events was examined. Proportions of DFIG vs. xed
Table 4.2: Maximum ROCOF following loss of largest infeed (422MW) for various
operating scenarios, wind turbine penetrations and wind turbine technology type.
WTG Proportions Wind Penetration (MW)
DFIG Fixed 0 500 1000 1500 2000
WP
0% 100% 0.29 0.30 0.32 0.34 0.36
25% 75% 0.29 0.30 0.32 0.34 0.36
50% 50% 0.29 0.30 0.32 0.34 0.36
75% 25% 0.29 0.30 0.32 0.34 0.36
100% 0% 0.29 0.30 0.32 0.34 0.36
SNV
0% 100% 0.48 0.54 0.62 0.72 0.87
25% 75% 0.48 0.54 0.62 0.72 0.87
50% 50% 0.48 0.54 0.62 0.72 0.87
75% 25% 0.48 0.54 0.62 0.72 0.87
100% 0% 0.48 0.54 0.62 0.72 0.87
SDV
0% 100% 0.44 0.48 0.54 0.61 0.69
25% 75% 0.44 0.48 0.54 0.61 0.69
50% 50% 0.44 0.48 0.54 0.61 0.69
75% 25% 0.44 0.48 0.54 0.61 0.69
100% 0% 0.44 0.48 0.54 0.61 0.69
speed wind turbines ranging from 0 to 100% were examined to represent numerous
possibilities of installed wind capacity in 2010.
The loss of the largest infeed (422 MW) during the 2010 winter peak, summer night
valley and summer day valley were simulated. It should be noted that the probability
of a generation trip coinciding with such extreme conditions is very unlikely, but pos-
sible, and the general operating point of the system lies between these extremes. The
important characteristics of the observed system frequency responses, namely maxi-
mum ROCOF and frequency nadir (minimum frequency), are summarised in Table
4.2, Fig. 4.6, Fig. 4.7, Fig. 4.8 and Fig. 4.9.
As increasing amounts of installed wind generators displace conventional generation, re-
gardless of whether DFIG WTGs with negligible inertial response or xed speed WTGs
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Wind Penetration (MW)
M
a
x
i
m
u
m

r
a
t
e

o
f

c
h
a
n
g
e

o
f

f
r
e
q
u
e
n
c
y

(
H
z
/
s
)
(a)
(b)
(c)
Figure 4.6: Eect of increasing wind penetration (100% DFIG) on maximum rate of
change of frequency following the loss of the largest infeed (422 MW) during the (a)
WP scenario, (b) SNV scenario and (c) SDV scenario.
with slow inertial response, the net eect is a reduction of system inertial response.
This results in increased rates of change of frequency during frequency transients. It can
be seen from Table 4.2 that maximum ROCOF is sensitive to the amount of installed
wind. Table 4.2 also illustrates that the type of wind turbine technology installed has
little inuence on the maximum ROCOF. Fig. 4.6, demonstrating the 100% DFIG case,
further illustrates the increase in maximum ROCOF with increasing wind penetration.
It should be noted that identical responses to those shown in Fig. 4.6 will be observed
for any of the proportions of xed speed to DFIG wind turbines listed in Table 4.2.
Considering that the ROCOF protection setting for generators connected to the Ireland
electricity system is 0.5 Hz/s, it can be seen from Fig. 4.6 that these protection settings
would not be exceeded for the winter peak scenario. This is due to the large amount
of conventional generation online. For the summer night and summer day valley sce-
narios, protection settings would be exceeded at penetrations of wind of approximately
140MW and 700MW respectively. Above these penetrations, there is a possibility that
additional generation will trip, exacerbating the initial power imbalance which caused
the low frequency event.
It should be noted here that the maximum ROCOF rates presented in Table 4.2 and
Fig. 4.6 represent maximum simulated rates based on a ltered dierentiation of the
frequency signal. The detailed operation of ROCOF relays installed on the system
would have to be known in order to determine if generation would trip for the sce-
narios examined (Beddoes et al., 2005). In particular the time period over which the
rate of change of frequency calculation is made would have to be examined in detail.
Regardless of the operation of ROCOF relays the general trend of increased rates of
change of frequency will be observed as wind displaces conventional generation.
While Table 4.2 illustrates that maximum ROCOF is independent of wind turbine tech-
nology, the frequency nadir is found to depend on the type of wind turbine technology.
An example of the eect of WTG technology type on frequency nadir during the SDV
scenario is depicted in Fig. 4.7, showing system frequency responses for no wind and
each of 2000 MW of xed speed and DFIG wind generation. While the ROCOF is
initially similar for both WTG types, the frequency nadirs reached dier substantially.
The xed speed WTG case results in a frequency nadir reaching the same frequency
level as for the no wind case, while the frequency nadir in the case of the DFIG WTGs
is considerably lower.
While Fig. 4.7 shows the frequency responses for a single penetration level of wind, the
eect of increasing penetrations of both xed speed and DFIG WTGs on frequency
nadir following the loss of the largest infeed (422 MW) is shown in Fig. 4.8 and Fig. 4.9.
Both SDV (Fig. 4.8) and SNV (Fig. 4.9) scenarios are shown in order to examine the
inuence of increasing wind penetrations on static reserve usage. As outlined in Section
4.4.1, the mix between static and dynamic reserve varies for SDV and SNV.
During the SDV scenario, following the loss of the largest infeed, a constant amount
of static reserve (50 MW, tripping at 49.5 MW) is tripped in all cases, so Fig. 4.8
primarily illustrates the diering eect on frequency nadir of the two WTG technology
types. When xed speed WTGs comprise the wind generation, the frequency nadir
rises by a small amount as wind penetration increases, due to the distinctive slower
inertial response of the xed speed WTG. However, if DFIG WTGs comprise the wind
generation, the frequency nadir falls further as wind penetration increases, due to
the negligible response of the DFIG WTGs to changing system frequency. Thus, given
constant capacity and dynamic/static mix of POR, the frequency nadir is detrimentally
0 5 10 15
49.3
49.4
49.5
49.6
49.7
49.8
49.9
50
Time (s)
F
r
e
q
u
e
n
c
y

(
H
z
)
(a)
(b)
(c)
Figure 4.7: Simulated system frequency following the trip of largest infeed (422 MW)
during the SDV scenario with (a) No wind generation, (b) 2000 MW xed speed WTGs
and (c) 2000 MW DFIG WTGs.
aected if DFIG WTGs displace conventional generators, and is largely unaected if
xed speed WTGs displace conventional generation.
Both the inuence of increasing wind penetrations on static reserve and the impact of
static reserve on frequency nadir can be seen in Fig. 4.9. When xed speed WTGs
comprise the wind turbine generation, 123 MW of static reserve is tripped in all cases,
and the frequency nadir does not fall below the 0 MW wind penetration case. However
for DFIG WTGs, the tripping of an additional 73 MW of static reserve is required to
prevent the frequency nadir from falling signicantly below the 0 MW wind penetration
case. As such, it may be concluded that increasing DFIG WTG penetration requires
increasing availability of static reserve to maintain the frequency nadir above a given
threshold.
0 200 400 600 800 1000 1200 1400 1600 1800 2000
49.3
49.32
49.34
49.36
49.38
49.4
F
r
e
q
u
e
n
c
y

N
a
d
i
r

(
H
z
)
(a)
(b)
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
50
100
150
200
250
S
t
a
t
i
c

R
e
s
e
r
v
e

T
r
i
p
p
e
d

(
M
W
)
(c)
Figure 4.8: Frequency nadir (Hz) and static reserve tripped (MW) following the loss
of the largest infeed (422 MW) for increasing wind penetration during the Summer Day
Valley (SDV) scenario, where (a) frequency nadir for 100% xed speed WTG case, (b)
frequency nadir for 100% DFIG WTG case and (c) static reserve tripped (for both
cases).
4.5.3 Supplementary response from DFIG
The magnitude of the inertial response of a DFIG depends on the extent to which
the rotational speed changes in response to changing system frequency. As the DFIG
is designed to provide accurate control of rotational speed, the coupling of rotational
speed to system frequency and the resulting inertial response is largely removed (Mul-
lane and OMalley, 2005). However through the addition of a simple supplementary
loop to the controller of the DFIG, it is possible to congure the DFIG to change the
electromagnetic torque in proportion to rate of change of frequency (Ekanayake and
Jenkins, 2004).
As a direct dierentiation of measured system frequency is undesirable due to suscep-
tibility to noise, the supplementary control loop torque, T
sc
, is determined using the
0 200 400 600 800 1000 1200 1400 1600 1800 2000
49.35
49.37
49.39
49.41
49.43
49.45
F
r
e
q
u
e
n
c
y

N
a
d
i
r

(
H
z
)
(a)
(b)
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
50
100
150
200
250
(c)
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
50
100
150
200
250
S
t
a
t
i
c

R
e
s
e
r
v
e

T
r
i
p
p
e
d

(
M
W
)
(d)
Figure 4.9: Frequency nadir (Hz) and static reserve tripped (MW) following the loss of
the largest infeed (422 MW) for increasing wind penetration during the Summer Night
Valley (SNV) scenario, where (a) frequency nadir for 100% xed speed WTG case, (b)
frequency nadir for 100% DFIG WTG case, (c) static reserve tripped for 100% xed
speed WTG case and (d) static reserve tripped for 100% DFIG WTG case.
following equation (4.15) as illustrated in Fig. 4.10.
T
sc
+pT
sc
= 2K
scl
p (4.15)
The supplementary control loop torque T
sc
is added to the reference torque of the DFIG
WTG, T
ref
, to provide the reference electromagnetic torque, T
emref
. This signal, in the
same way as a natural inertial response, will impart energy to the system as frequency
drops, and absorb energy from the system as frequency increases, thus increasing the
response of the DFIG above that which would be normally observed. The magnitude
of the response will depend on the controller parameters (supplementary control loop
constant K
scl
and time constant ). The achievable inertial response of the DFIG
WTG is limited by operational constraints such as current limits. In this study it is
assumed the these limits are not exceeded. It must be noted that in practise, however,
some design modications may be required to prevent deviation outside operational
constraints.
Figure 4.10: Supplementary control loop for DFIG WTG controller.
A comparison of the inertial response during a frequency event from a synchronous
generator, a xed speed WTG and a DFIG WTG with the supplementary control loop
included is illustrated in Fig. 4.11(II). Instead of a negligible response to system fre-
quency, as shown in Fig. 4.5(II), the addition of the supplementary controller results in
a DFIG response similar to the inertial response of the conventional generator. The size
of the response is dependent on the value of parameter K
scl
within the supplementary
control loop, which was chosen to be 3.5 for this study.
The system frequency following the loss of the largest infeed during the SDV scenario
is illustrated in Fig. 4.12. This gure is similar to Fig. 4.7 but includes 2000 MW of
DFIG WTGs with supplementary control. The addition of the supplementary control
loop to the DFIG WTG results in an improvement in system frequency response, both
in terms of the rate of decline of the frequency and the maximum frequency excursion
that occurs, as indicated in Fig. 4.12. However, these eects depend on both the value
of the constant K
scl
and the time constant within the controller.
While these results indicate that the addition of the supplementary control loop to the
DFIG WTG can potentially be benecial to system frequency response, the ease of
integration into the current technology is not assessed here.
4.6 Conclusions
Wind turbine generator models of two dierent technology types, xed speed and
DFIG WTGs, are incorporated into the validated system model to examine the eects
of increasing wind penetration on short-term system frequency control. The validated
system model is adapted to represent the predicted 2010 Ireland electricity system, and
wind penetration levels of up to 2000 MW are examined.
0 5 10 15
49.6
49.7
49.8
49.9
50
(I)
F
r
e
q
u
e
n
c
y

(
H
z
)
0 5 10 15
1
0
1
2
3
4
(II)
Time (s)
C
h
a
n
g
e

i
n

P
o
w
e
r
(
%

o
f

C
a
p
a
c
i
t
y
)
(a)
(b)
(c)
(d)
(e)
Figure 4.11: (I) Simulated system frequency resulting from the loss of the largest
infeed during WP. (II) Comparison of xed speed WTG and DFIG WTG responses
to the low frequency event in (I): Change in power output (%) of (a) a synchronous
machine with inertial constant of H=4.2 s, (b) the xed speed WTG, assuming constant
power, (c) the xed speed WTG, assuming constant torque and (d) the DFIG with
supplementary control loop included (K
scl
= 3.5), assuming constant power and (e)
the DFIG with supplementary control loop included (K
scl
= 3.5), assuming constant
torque.
The results indicate that the rate of change of frequency during a loss of generation
event will increase as increasing proportions of wind generation displace conventional
generation. This increase in ROCOF was found to be independent of wind turbine
technology. Therefore, as wind penetration increases in the future, displacing conven-
tional generation, the system may experience an erosion of system inertia, particularly
during periods of low load, with greater rates of change of frequency. If ROCOF in-
creases suciently, the activation of protection relays may result in loss of generating
units and other components on the system. As a consequence, the tripping of system
load may be necessary.
The minimum frequency reached during a loss of generation event for increasing wind
penetrations was found to be dependent on wind turbine technology. The minimum
0 5 10 15
49.3
49.4
49.5
49.6
49.7
49.8
49.9
50
Time (s)
F
r
e
q
u
e
n
c
y

(
H
z
)
(a)
(b)
(c)
(d)
Figure 4.12: Simulated system frequency following the trip of largest infeed (422 MW)
during the SDV scenario with (a) No wind generation, (b) 2000 MW xed speed WTGs,
(c) 2000 MW DFIG WTGs and (d) 2000 MW DFIG WTGs, with supplementary
control loop added
frequency reached with increasing proportions of xed speed WTGs was found not to
fall below the level reached in the no-wind case. However, increasing DFIG generation
results in progressively greater frequency deviations from nominal and may therefore
necessitate additional availability of static reserve to maintain system frequency above
a given threshold. An increase in POR can oset the increase in frequency deviation
from nominal in the case of DFIG generation, to prevent possible load tripping. How-
ever, POR can not substitute for reduced system inertia to slow the rate of change of
frequency (Lalor and OMalley, 2003).
One possible solution is to congure the DFIG WTGs to respond in proportion to
changing system frequency through the addition of a supplementary control loop. The
results demonstrate that the addition of this control loop to DFIG WTGs is benecial
for system frequency control. However, the technical limitations such as current limits
and the ease of integration are not assessed here.
Chapter 5
Supplementary Study: Wind
Integration Studies and Frequency
Control
5.1 Introduction
The technical aspects of short-term frequency control on an island power system with
evolving plant mix have been introduced and examined in Chapters 2, 3 and 4. How-
ever, frequency control encompasses a range of time-frames from seconds up to hours
and days and in order to maintain a secure and reliable power system, a clear under-
standing of the technical aspects of frequency control in each time-frame is required.
AC interconnection to neighbouring power systems oers many advantages for fre-
quency control, as highlighted in Section 1.2.3. However, maintaining scheduled power
ows on the interconnections as well as internal power balancing needs to be taken
into account for frequency control. In addition, the technical aspects comprise just one
part of the whole frequency control issue and therefore need to be put in perspective
by examining them as part of the broader picture, which encompasses the economic
side in addition to the technical details.
Consistent categorising and denitions for system frequency control is dicult, as time-
frames overlap and vary from system to system. Here, system frequency control has
been broadly categorised into three dierent time-frames: system frequency regulation,
Chapter 5. Wind Integration Studies and Frequency Control 106
load following and unit commitment.
System frequency regulation is the balancing of minor power uctuations in the second
to minute time-scale, to maintain system frequency at nominal. Regulation is achieved
automatically, either by a combination of AGC and governor action (Kundur, 1994),
or in the absence of AGC, by governor action alone (ESBNG, 2005b). AGC is the
automatic control of the set-point of generating units. The objectives of AGC are to
maintain system frequency at nominal frequency, maintain power ows on intercon-
nections between power systems or control areas as scheduled and allocate the power
regulation between generating units to minimise costs and maintain system security
(Machowski et al., 1997).
AGC operates to minimise the system area control error (ACE), which is a function
of the deviation from scheduled system frequency and the deviation from scheduled
active power ow on the interconnections to neighbouring systems. Neglecting meter
error corrections, which should generally be zero or negligible, the ACE is calculated
from equation 5.1 (NERC, 2005).
ACE = (NI
A
NI
S
) 10B(F
A
F
S
) (5.1)
where
NI
A
is the algebraic sum of the actual ows on all tie lines or interconnectors (MW)
NI
S
is the algebraic sum of the scheduled ows on all tie lines or interconnectors (MW)
B is the frequency bias setting (MW/0.1Hz)
F
A
is the actual frequency (Hz)
F
S
is the scheduled frequency (Hz)
Load following frequency control is the adjustment of online generating unit set-points
to follow trends in the system load and variable generation and thus maintain active
power balance while minimising cost. Load following occurs over a longer time frame
than regulation, ranging from ve or ten minutes to hours. Finally unit commitment
occurs from hours to days ahead of real time, and involves the scheduling of generating
units to be on or oine.
Wind generation has a number of characteristics that dierentiate it from conventional
synchronous generating units connected to a power system, such as their distinct contri-
bution to overall system inertial response, as discussed in Chapter 4. The power output
from a wind turbine generator (WTG) is variable, with this variability depending not
only on the operating status of the generator but also on the wind conditions, which
in turn depend on weather patterns and local terrain. In addition, the predictability
of WTG power output is generally considerably lower than that of a conventional gen-
erator and as a consequence, controllability is also limited. The variable and uncertain
nature of wind power generation need to be taken into account in order to maintain
adequate frequency control resources in all time-frames. The eects of the short-term
wind turbine dynamics on frequency control in the event of a sudden low frequency
event was examined in Chapter 4. With rapidly increasing wind generation on many
systems, however, the eect on all aspects of frequency control is being examined in
detail, to assess and quantify the impact of wind integration.
A review of a number of previous wind integration studies is carried out in Section 5.2.
Subsequently a preliminary methodology is proposed for examining the eects of wind
integration on all aspects of frequency control over a number of time-frames. Some
illustrative results are given for a sample interconnected system.
5.2 Review of Wind Integration Studies
The predominant focus of many initial studies of wind generation was the impact of
wind turbine generators on the transmission/distribution system at the point of inter-
connection and also their technical operational capabilities. This early focus, driven
by the diering characteristics of wind turbine generator to conventional synchronous
generators, was to maintain system security as wind generation came onto the system.
As a result, grid codes taking account of the characteristics and technical impacts of
wind turbine generators were developed by a number of system operators (ELTRA,
2000; Jauch et al., 2004; CER, 2004b,c).
However, as proportions of wind generation continue to increase rapidly worldwide,
system operators and power producers alike require a clear indication of the inuence
of wind generation on system operations and other generation on the system, in order
to determine optimal operational strategies and minimise costs. As a consequence, the
focus of many recent wind integration studies has turned to the economic impacts of
wind generation. Integration of a variable generating resource such as wind genera-
tion, which is typically uncontrollable, impacts the system operation in all time frames
(Brooks et al., 2002). Therefore, the time scales of these studies range from the ex-
amination of short-term behaviour of wind generation to the establishment of capacity
values for wind generation for use in the long-term operation and planning of power
systems. The eects of the characteristics of wind generation on operating reserve,
regulation, load following and unit commitment are studied, and any resultant change
in system operational costs are quantied where possible (Hirst, 2002a,b; Brooks et al.,
2003; Milligan, 2003; PaciCorp, 2003; SEI, 2004; Zavadil et al., 2004; GE, 2004, 2005).
As increasing amounts of wind generation are integrated into power systems, the general
consensus is that increasing wind will inuence system operating costs. The extent of
this inuence will vary with wind penetration, system generation characteristics such
as system size and generation mix, system load, the method of operation of the system
and also time scale. In the short term, many system characteristics are xed, such
as the system size, generation mix and load characteristics. However, operational
processes may be modied, given sucient need or incentive. Studies such as Hirst
(2002b) and Brooks et al. (2003) look at the inuence of existing wind generation on
system operations and try to quantify the ancillary service costs attributable to wind.
In the long term, however, some of the characteristics that are outlined as xed in
the short term become more exible, for example the technology and location of new
generating resources and the development of exible load. Several studies (Milligan,
2003; PaciCorp, 2003; Hirst, 2002a; SEI, 2004; Zavadil et al., 2004; GE, 2004, 2005)
look at future predicted or possible wind penetration levels on a number of power
systems, and analyse the potential inuence of wind generation on costs.
Hirst (2002b) assesses the interaction of wind generation with the hour ahead energy
market, the real time intra-hour imbalance market and the regulation market of the
PJM system, which has a peak load of approximately 52,000 MW. Actual 1-minute
average output wind data for a 103.5 MW wind farm for two 1-week periods is employed,
although it is important to note that the wind farm is not located within the PJM
system, so the assessment of correlation between load and wind is not possible. For
each week, the costs and payments are assessed for two dierent operating strategies.
In the rst, the wind forecast (determined using persistence modelling) is input into
the hour ahead imbalance market, and any dierences between actual and forecast
are settled in the imbalance market. For the second operating strategy, the actual
wind energy is input straight into the imbalance market. The revenue achieved by
the wind farm ranged between 34.70 and 31.30 $/MWh, depending on the week in
question and operational strategy. Slightly higher prices were achieved when wind
participated in the hour ahead market. However, the price received by the wind farm
was calculated based on historical market prices, which did not take into account the
eect of wind on market prices, even when the wind farm is scaled to 50 times actual
capacity. The resultant costs to the wind farm, although dependent on that particular
market structure and prices, and given the limited time span of the wind data and
thus the study time frame, do give a reasonable order of magnitude approximation of
the costs. One major development made, as highlighted in Parsons et al. (2003), is
that Hirst (2002b) is the rst to recognise that each variation in wind power output
does not need to be matched exactly by an equal and opposite variation in another
resource. Instead, it is the overall system balance that is important, and the addition
of a variable resource of wind merely adds to the overall system imbalance.
Brooks et al. (2003) examined the eect on operating costs of 280 MW wind on a
system with a peak load of 8000 MW (Xcel Energy North) over time scales ranging
from regulation to unit commitment. The objectives of this study were to identify and
quantify the incremental operational costs in real time operations due to the existing
280 MW of wind generation and to determine the eect on the short-term scheduling
and commitment operating costs due to wind generation forecast uncertainty. A Monte
Carlo approach was adopted for the study, where many wind power time series were
used for each time scale of interest. Therefore, a distribution of results that are statis-
tically representative of wind impacts is established. These time series were generated
using historical 1 second, 5 minute and 1 hour resolution wind data. They are based
on a probabilistic model that quanties the probability distribution of wind generation
for a specic time period given the historical data. The wind generation uctuation
statistics in the regulation time frame were extracted by taking the dierence between
the original historical data and the 30-minute moving average. Overall, costs due to
forecast errors in the unit commitment time-frame were found to be dominant, rang-
ing from 0.39 - 1.44 $/MWh of wind power. The costs of non-optimal commitment
due to inaccurate forecasts were determined from the dierence between the imperfect
and perfect forecast runs. It was also found that this cost impact decreases as fore-
cast accuracy increases. For 280 MW of wind penetration, the reserve requirement for
load following was not increased signicantly. Thus the reserve component of the load
following cost was assumed zero. The load following energy component, which is the
cost of deploying existing load following reserve on the system, was found to be 0.41
$/MWh. In the regulation time-frame, there was negligible change in ACE statistics,
implying that at this wind penetration, short term uctuations of wind are small in
comparison with load uctuations, and there is little impact on control performance.
Regulating burden was found to increase by about 4 %, but no additional regulating
reserve was deemed necessary.
As well as examining actual wind generating penetrations, future predicted or possi-
ble wind penetration levels are often studied to attempt to assess and quantify the
various impacts of increasing wind penetration before actual connection to the power
system. Milligan (2003) quanties the additional load following requirement caused
by increasing wind penetration with a range of geographical dispersions on the Iowa
power system. Analysis of up to 1600 MW of wind generation plant (22.7% of peak
load), based on 1 year of actual wind speed data, is carried out. The study recognises
the importance of geographical dispersion and wind dispersion is varied from all wind
generation in a single location to a wide geographical spread of the wind farms. The
study initially examines the load following requirements arising from load and wind
variability, ignoring forecast errors in both, and quanties the contribution of each to
load following for dierent geographical dispersions. Wind contribution to load fol-
lowing is calculated as a percentage of installed wind capacity and is calculated to be
approximately 3.6% for 1600 MW wind with little dispersion. Increased geographi-
cal dispersion reduces this load following requirement to 2.5%. Wind forecast errors,
determined using a simple persistence modelling method, and load forecast errors are
introduced and wind forecast accuracy is varied to see what eect improvement will
have on load following requirements. As expected, load following requirements at-
tributable to wind decreases as forecast accuracy improves. No attempt to state which
generators should provide additional load following or to quantify the costs associated
with it are made in this study. Instead, the focus of the study is entirely on physical
requirements imposed on system by additional wind generation. Although specic to
Iowa system, this methodology could be expanded to other systems.
The impact of increasing wind generation on the PaciCorp system was examined in
PaciCorp (2003). The incremental reserve requirement and ensuing costs were quan-
tied using one year of hourly wind and load data. The imbalance costs, which are
believed to be the primary cost impacts of wind on the system, were also quantied.
For the incremental reserve requirement quantication, the wind was treated as neg-
ative load, thus recognizing that the balancing of the entire system is adequate for
frequency control. It was found that as wind penetration increases, the increment re-
serve requirement increases faster at higher penetration rates, and this concurs with
the results of Milligan (2003) and SEI (2004). For the imbalance cost calculation, wind
was modeled as hourly blocks with constant output and used as input to an hourly
dispatch model. Wind penetration was varied and compared to the base case of no
wind. While a simple representation not taking forecast accuracy into account is used
(100% forecast accuracy assumed), a preliminary approximation of the imbalance costs
of wind generation is presented. This approximation is believed to be a good rst at-
tempt at determining at least the order of magnitude of the cost of wind integration
(Smith et al., 2004).
Hirst (2002a) assessed the additional operating costs of running the Bonneville Power
Administration (BPA) system with 1000 MW of wind over the three distinct timescales
of day ahead (unit commitment), load following and regulation. Actual 1 minute
resolution wind data for 4 months from 164 MW of wind farms was available and this
was conservatively scaled up to represent 1000 MW of wind. This scaling was deemed
conservative, as no account was taken of diversity benets. Load and available reserve
capacity data of 1-minute, 5-minute and 1-hour resolution was also available for the
control area under analysis. As no wind forecast was available, two were developed. The
rst determined the forecast error based on the dierence between the actual hourly
wind data and the daily average, and the second was based on the monthly average.
For the day-ahead analysis, load and wind forecast errors were found to be highly
uncorrelated, so the eects of wind on the power system are generally small. Total
forecast error (load and wind forecast error) did not increase signicantly from load
forecast error for either of the wind forecast methods. Load following requirements of
wind and load were also largely uncorrelated, so incremental load following requirement
of wind was found to be only small, and minute to minute regulation costs even smaller.
Overall, total costs were found to be dominated by the forecast error costs and the costs
decreased with the time scale. The total cost to integrate 1000 MW of wind generation
was found to be less than $5/MWh of wind generation. While the study provided
a useful conclusion, accuracy of results are dependent on the accuracy of the wind
forecast developed.
Zavadil et al. (2004) looks at 1,500 MW wind spread over large geographical area, on a
system with peak load of 10,000 MW with the objective of conducting a comprehensive,
quantitative assessment of integration costs and reliability impacts on the Xcel Energy
Minnesota control area of 2010. Wind speed data was consisted of three years of 10-
minute data simulated using advanced wind speed simulation methods comprised the
wind speed data. This was then converted to wind generation data by applying power
curves for existing and prospective commercial wind turbines at each of the grid points
in the wind speed simulator, and validated using actual data from an existing wind
farm within the control area. Therefore, geographical dispersion eects were captured.
The study concluded that the cost of integrating 1500 MW wind was no higher than
$4.60/MWh of wind generation. These costs, concurring with previous studies (Hirst,
2002a), were dominated by the day-ahead costs due to the signicant variability of
wind generation and wind generation forecast in this time frame. Negligible cost was
associated with the load following as wind variability was negligible in comparison with
the load changes. However, an impact on control performance was found. Ten-minute
changes of larger magnitude occur more frequently with wind generation, so although
sucient reserve was available to cover the changes, ramping rates were not always
sucient. As such, an increase of 1 - 2 MW/min in ramping would be necessary to
maintain current performance control levels if 1500 MW of wind was integrated into
the control area.
A study, reported in two phases, GE (2004) and GE (2005), was undertaken for the NY
ISO on the eects of integrating wind power on transmission system planning, reliabil-
ity and operations. The study is comprehensive, attempting to encompass all aspects
of wind penetration relevant to system operators in a single study. Initially, system
operators experiences worldwide were drawn upon to identify the critical issues with
respect to the integration in wind power. In GE (2004), the maximum wind generation
that could potentially be interconnected to the New York State Bulk Power System
(NYSBPS) given existing transmission infrastructure and viable wind farm sites was
determined to be approximately 3300 MW, or 10% of peak load for projected 2008
system. It was found that this penetration of wind would not aect system reliability,
measured using a loss of load expectation (LOLE) criterion, given the intended future
power system for 2008. However, if a reduction of existing or planned thermal gener-
ation on the system were to occur due to increased wind generation, reliability may
degrade.
The inuence of wind generation on regulation, load following and the scheduling and
dispatch of generation was studied in GE (2005). In addition, reliability and genera-
tion capacity as well as stability performance following major grid disturbances were
examined. For all time scales it was found that the existing processes and resources in
the NYSBPS could accommodate an incremental increase in wind generation. Existing
day-ahead and hour-ahead energy markets were deemed to be suciently exible to
accommodate wind without any major changes, although wind variability was found to
add further uncertainty to that already present due to the load variability, particularly
day-ahead. The importance of accurate forecasting was underlined and the advantages
of a centralised forecast highlighted. Load following, calculated on a 5-minute time
frame, resulted in a small increase in the load following requirements. An additional
36 MW of regulation (calculated every 6 s) was found necessary to maintain the same
level of performance with the addition of 3300 MW wind to the NYSBPS. No change
to spinning reserve requirements was deemed necessary, as suciently large changes
in wind generation to warrant such a change were not found in the three years of
data analysed. Stability was found to improve with the addition of wind generators,
assuming state of the art technology. The study concludes NYSBPS can reliably ac-
commodate at least 10% penetration with just minor changes to existing planning,
operation and reliability practices, and would result in a reduction in NOx and SOx
emissions and an average reduction in the spot price of electricity in New York State
based on available data.
In SEI (2004), a study carried out to quantify any additional operating reserve require-
ments on the Ireland electricity system due to increasing wind generation, future wind
penetrations of up to 1950 MW were examined. This level of wind generation would
represent 20% of total generating capacity on the predicted 2010 Ireland electricity sys-
tem, and at times of high wind speed and low load, a signicant portion of generation
could consist of wind turbine generators. The sensitivity of operating reserve require-
ments to various wind levels and also system operation modes was examined for time
frames ranging from POR to replacement reserve, examining a number of sample days.
The study concluded that although additional operating reserves would be required
with increasing wind penetration, they would not be substantial if wind generation
and wind forecast error were taken into account when dispatching system generation.
Wind power variability was found to be considerably less than load variability, even
with 1950 MW of wind generation on the system. Therefore, load variability domi-
nated the total variability. For short time horizons, wind forecast accuracy is high and
therefore, little or no additional reserve is required. As the time horizon increases, the
wind forecast become less accurate, and in longer time frames, reserve targets may be
breached occasionally. The cost of additional reserve was estimated to be less than 0.50
euro/MWh in 2010 with 1950 MW wind generation. The eect of correlation between
wind farms was taken into consideration in this study, and it was found that for the
Ireland system the benets of diversity were limited due to geographical size.
Wind studies can not easily be generalized to dierent systems, because results are
dependent on system size and characteristic, generation mix, wind and load character-
istics and fuel costs. However, a framework can be developed, to provide a guideline
for the implementation of future wind studies on various systems, and the pertinent
parameters of interest extracted.
5.3 Preliminary Wind Integration Frequency Con-
trol Study
The particular characteristics of wind that are highlighted in many wind integration
studies, a number of which are considered in Section 5.2, are variability and unpre-
dictability. While these characteristics are not exclusive to wind, the degree of vari-
ability and unpredictability of wind generation tends to be higher than other conven-
tional generation resources. An understanding of the inuence of such characteristics
on the power system to which the wind generation is connected, and in the case of
interconnection, the neighbouring power systems, is useful for all participants.
A preliminary methodology to examine the eects of both variability and unpredictabil-
ity of wind generation on system frequency control is described. However, in contrast to
many previous studies where the inuence of wind generation as a whole is examined in
each time scale, the proposed methodology attempts to separate the costs attributable
to wind into costs due to variability and those due to unpredictability. The objective
of separating the costs of unpredictability is to establish the value of wind forecasting
and determine a limit, beyond which the cost of improving wind forecasting may not
be of economical benet.
Application of the preliminary methodology to a sample interconnected power system
using the AREVA T&D e-terra simulator is proposed. The method of assessment of
the eects of wind variability and unpredictability on the operations and costs of sys-
tem operators and power producers using the e-terra simulator is outlined and some
sample illustrative results are presented. The suitability of the e-terra simulator for
such a study is assessed and, where appropriate, suggestions for possible improvements
are made. Transmission eects are neglected for the study, as the impact of wind gen-
eration on frequency control is the primary objective. However, the e-terra simulator
includes a full transmission and distribution system model, and therefore, a study im-
plemented using this simulation platform could be expanded with ease to incorporate
the transmission and distribution systems.
5.3.1 e-terra simulator
The e-terra simulator is a dispatcher training simulator developed by Areva T&D, which
provides a realistic simulation of both power system and control centre environment.
It is a component of the e-terra platform, an energy management system (EMS), but
has the exibility to be operated independently of the other components of the system
if necessary. While primarily used for training of dispatchers for various scenarios, it
is also a useful predictive and analysis tool for running real time simulations. The
three core elements that comprise the e-terra simulator are (a) the EMS model, which
replicates the actual EMS functions, (b) the power system dynamic simulator, which is
a model of the power system, and (c) the instructional subsystem, where the simulation
of dierent scenarios may be set-up and controlled.
EMS Model
The EMS model for the e-terra simulator contains the same components as an actual
EMS. These include SCADA (Supervisory Control and Data Acquisition), RTGEN and
ALARM. SCADA in the e-terra simulator is identical to SCADA in a real EMS except
the data exchange occurs with the dynamic model of the power system instead of the
actual power system. The primary function of the RTGEN program is AGC. RTGEN
regulates the generation to control system frequency and ow on the interconnectors,
schedules sale and purchase of power with neighbouring control systems, and monitors
and controls reserve levels. AGC does not regulate short-term natural oscillations in
generator outputs or interconnector ows, as these are controlled through generator
governor action. AGC instead provides the secondary control action, to maintain
steady state values at specic levels. The AGC process can run as frequently as every
two seconds (Areva, 2004). ALARM is a program that alerts and aids the system
operator (or user of the e-terra simulator in this case) in the event of abnormal or
uncommanded equipment operation on the power system.
Power System Dynamic Simulator
The dynamic model consists of a detailed model of the components and topology of the
power system. This power system dynamic simulator models the dynamic behaviour of
generation, load, and circuit breakers. The prime movers and relays are simulated and
system frequency is calculated. Long-term dynamic models are used, with a minimum
step size of 0.5 seconds. Hence, transient responses and intermachine oscillations are
not modelled. The system power ows are calculated based on the network topology
changes, load schedules and prime mover mechanical power.
The dynamic model described above does not model the short-term frequency response,
as modelled in Chapters 2, 3 and 4. In order to model short-term frequency control
in the event of a large power imbalance and the eects of system inertia and dynamic
characteristics of dierent generators, integration of models such as those described in
Sections 2.2, 3.5.1 and 4.3 would be desirable. However, in order to model the short-
term voltages and transmission eects in addition to frequency eects, as necessary for
a large interconnected system, additional models incorporating voltage eects would
be necessary. In addition, while intermachine ocillations on a well designed, electrically
short power system are negligible, as described in Section 2.2.1, for an interconnected
system or a system with large distances between generators, intermachine oscillations in
the event of a power imbalance may become signicant, and therefore require modelling.
Instructional Subsystem
The instructional subsystem is the interface between the user and the simulator. Here,
the loading, initialising, running, pausing and stopping of simulations is controlled by
the user, as well as management of various simulation parameters such as simulation
time and simulation speed. Simulations may be run at speeds ranging from real time
up to 20 times as fast. Numerous events such as a loss of generation, the tripping
of transmission lines and equipment failures can also be set up to occur at either a
set time into the scenario or on the occurrence of a trigger event/situation. When
using the e-terra simulator to model an actual power system, a snapshot of the entire
system at any time may be taken and loaded into the e-terra simulator. This provides
the capability of running simulations much faster than real time to examine dierent
possible future scenarios and aid decision making.
Wind modelling in the e-terra simulator
No wind turbine generator models are currently available for the e-terra simulator. This
is due to the variety of wind turbine technologies and the deciency of any standard
wind turbine generator model. Therefore, no standard prime mover model has been
developed.
Instead, wind generation is represented as an actual or forecast wind power output
time series. Each wind turbine generator is not represented individually on the system.
Rather, each wind farm, consisting of multiple wind turbine generators, is represented
as a single time series, with resolution dependent on the time scale of interest to the
study. This power prole is input to the power system model at the point of wind farm
interconnection. Therefore, wind turbine dynamic behaviour, unless embedded within
the power time series data, is neglected.
The representation of wind generation as a single time series in e-terra simulator allows
the eects of wind generation on conventional system generation, long-term system
frequency and system operations to be assessed. However, the eects of a power system
event such as a loss of generation, loss of transmission lines or equipment failure on the
wind farm can not be examined. Therefore, the modelling of eects of wind turbine
behaviour on other generation, system frequency and system operations during such an
event is also not possible. This could be amended through the inclusion of a dynamic
wind turbine model. However, as the e-terra simulator dynamic models are long-term
models, even with the addition of a dynamic wind turbine model, the inuence of wind
on short-term frequency dynamics is not possible.
5.3.2 Wind Data
Many wind integration studies examine the eects of future integration of substantial
amounts of wind generation. Therefore, availability of actual wind farm power output
data will be limited, or often unavailable, for many of the proposed wind farm sites.
Wind speed and direction data for specic sites is therefore a useful and common source
of data for wind integration studies. However, care needs to be taken when converting
wind speed data into the power output data from wind farms. The conversion of wind
speed into wind power may be accomplished through the use of wind turbine power
curves. However, use of this method results in data suitable only for analysis in time-
frames where changes in wind turbine power output are small and slow enough such
that the wind turbine can be assumed to be in steady state at all times, and dynamics
will not inuence the wind turbine. If the resolution of the wind data is any greater
than approximately 15 seconds (Mullane, 2005), wind turbine dynamics start to come
into play, invalidating the steady state assumption.
In addition, the scaling of the power produced by a single wind turbine to represent the
power output of an entire wind farm is non-trivial. The smoothing eect of multiple
generators dispersed over the site of the wind farm needs to be taken into account for
a realistic power prole, and a number of dierent methods have been employed in
previous studies (Brooks et al., 2003; Zavadil et al., 2004). Norgaard and Holttinen
(2004) proposed a method to generate a qualied estimate of the time series of the
aggregate power generation of multiple wind turbine generators distributed over the
area of a wind farm. Using a single wind speed time series for the site and a standard
power curve for a single wind turbine representative of the wind turbine generators
in the wind farm as inputs, as well as the dimensions of the area of the wind farm,
a power curve which takes into account the smoothing eects of time and space was
developed. Although rapid, small-scale uctuations were not to be captured by the
aggregate power output time series when compared with actual aggregated power out-
put during validation, a number of qualities of the aggregate power were reproduced.
This methodology provides a reasonably accurate approach to modelling the smoothing
eects of multiple wind turbines.
5.3.3 Determination of Wind Variability Costs
Wind variability occurs on all time frames from seconds to days, and can potentially
impact on regulation, load following and the commitment of generation to dierent
extents.
The variability of wind is a characteristic that changes from location to location. As
such, the results of any study are directly linked to the quantity, quality and resolution
of wind data available. However, in addition to the wind data, a number of other
factors will inuence the outcome of any study. These factors include:
Wind penetration level: However variable the wind generation, if penetration
levels are low, then the eects are unlikely to greatly inuence total system vari-
ability. Conversely, if wind generation reaches large penetration levels, the impact
of even small amounts of variability may be signicant.
Magnitude of load variability and correlation between wind and load variability:
If load is highly variable, then load variability eects may dominate total (i.e.
load and wind) variability.
Generation mix on the system: The ability for conventional generation to ramp
to counter system variability and the resultant costs of ramping will play a large
role in any costs attributable to wind in the event of an increase in total system
variability.
Power system characteristics such as inertia will also play a role, particularly
in the short-term frequency control time-frame, as illustrated by the results in
Section 4.5.
Regulation
Regulation deals with the rapid, small-scale uctuations in system load and generation
around the longer time scale trends. Short-term uctuations with a time-scale of several
seconds are generally dealt with by generator governor action. For longer time-scale
uctuations, generally greater than 5 seconds, AGC will respond. As wind power
uctuation are generally small enough to be neglected below ve seconds (Parsons et al.,
2001), the costs imposed/saved on the regulation time-scale due to wind variability can
be calculated based on the additional AGC requirement.
E-terra simulator, equipped with long-term dynamic models for conventional genera-
tors, is capable of accurately simulating power system conventional generation control
in the regulation time-frame.
Wind turbine generators are at present modelled in the e-terra simulator as a single
aggregate power output time series from each wind farm. Therefore, the dynamic
behaviour of the wind turbines to any short term uctuations in wind speed would
need to be encompassed within the data. This is possible as the wind data time series
is the actual power output from wind turbine generators. If the time series is derived
from wind speed data, it is likely that the wind turbine short term dynamics due to
changing wind conditions will not be captured.
However, with wind turbine generators modelled as a power time series input, the
dynamics of the wind turbine generators to changing system condition can not be cap-
tured in any time frame. Therefore, if using the e-terra simulator for a wind integration
frequency control study, the eects of wind turbine generator dynamic behaviour are
not taken into account.
Load Following
Load following in the time scale of minutes to hours is carried out through a com-
bination of AGC and changing the dispatch of the generation currently online. The
cost of load following imposed on the system as a result of wind power variability can
be calculated from the AGC reuirements and the system dispatch requirements. If
the system is run assuming wind power to have a constant output, the costs of load
following attributable to system load may be assessed. Consequently, any dierence
between the costs of running the system with and without variable wind power can be
attributed to wind generation. Depending on the correlation between system load and
wind, and capacity of the wind generation, these costs may vary, to the extent that
cost to the system may be lower if wind variability osets load variability.
The AGC costs of wind will encompass both the regulating and load following time
frame, while the dispatch costs incurred/saved due to wind will be attributable solely
to the load following time-frame.
The e-terra simulator is suitable for the analysis of the inuence of wind generation in
the load following time-frame between 5 and 30 minutes. AGC deals with the system
imbalances between system generation and load in this time frame, and all units within
the model are equipped with AGC. Numerous scenarios are possible to assess the overall
eects of the wind variability. Increasing wind penetrations can be examined.
The e-terra simulator in conjunction with an e-terra commitment and dispatch platform
is suitable for re-dispatching the system to deal with wind power variability.
Unit Commitment
The additional costs due to the variability of wind in the unit commitment time-
scale can be calculated by comparing unit commitment costs with and without wind
variability. Determining the unit commitment without wind variability entails the
commitment and dispatch of the system to meet load with wind generation constant
for the commitment period at its average value. To include wind variability in the
unit commitment, the system is committed and dispatched again, this time with wind
generation varying. The dierence between the cost of the two commitment schedules
is the cost attributable to wind variability in this time frame. Most unit commitment
programs can be used to assess the impact of wind variability on system operations
and costs.
5.3.4 Determination of Wind Unpredictability Costs
Wind unpredictability increases with time, as forecasting becomes more dicult as time
increases. Therefore, in the regulation time frame, wind predictability is only minor
and can be neglected. However, over the longer time scales, wind forecasts become less
reliable and costs of wind unpredictability tend to increase with increasing time.
Determination of the costs associated with the unpredictability of wind consists of
two steps. Initially, the operation and costs of operating the system is determined for
the situation where wind generation is exactly as forecast. This involves dispatching
generation to meet load with the forecast wind power output. This dispatch can then be
run in the e-terra simulator to determine the operating costs of the system, where the
wind time series is the forecast wind time series. Then, to determine the change in costs
of running the system due to unpredictability of wind forecast, the system simulation
is run once again, this time with the actual wind power time series. The eect on other
generation, system frequency and interconnector power ows are then monitored, and
any dierence in system operating costs can be attributed to the unpredictability of
wind.
5.4 Preliminary Results
A methodology for assessing the impact of wind integration on system frequency control
is presented in Section 5.3. This is a generic methodology that can be applied to a wind
integration study, employing the e-terra simulator. This methodology is applied to a
test system to provide some illustrative results. However, due to limited wind data, and
the absence of a unit commitment program, the scope of this study is severely restricted.
As such, this work is incomplete and the results are limited in their usefulness.
5.4.1 Available Wind Data
For this study, wind data from two separate sources was available. From the rst source,
wind speed data of ten-minute resolution for one year (2003) was available, both before
and after being corrected for atmospheric air density. In addition, a simulated forecast
error data set of 365 forty-eight hour forecasts was also available (i.e., one forty-eight
hour forecast per day for one year). Twenty-four hours of ve-minute resolution data
for two hypothetical wind farms, each of capacity 400 MW, was also available from
the same source. The data for each wind farm is the power output from a single wind
turbine at that location scaled up to have a capacity of 400 MW. Therefore, smoothing
eects of multiple wind turbines are not captured. This data was captured during
particularly stormy conditions. There is some correlation between the two wind farms
at certain times of the day, and the correlation coecient between the two wind farms
is 0.272.
From the second source, actual wind data of ten-minute resolution for two one-month
periods, one in summer and one in winter, were available. Both the observed wind
speed and the actual power output in MW were available, as well as the one-hour
forecast for each time.
5.4.2 Scope of the study
Regulation
Given the wind data available, evaluation of the wind generation variability eects on
regulation requirements for the test system under analysis was not feasible, as the data
resolution (5 minutes is the highest resolution data available for this study) was not
high enough.
Load following
The two ve minute time series of wind power data, representing two 400 MW capacity
wind farms, as described above, is used in this study. Therefore, the eect on the system
can be calculated using the e-terra simulator. However, due to unavailability of the
commitment and dispatch platform, AGC was required to take care of all variability
from the wind farms.
Unit commitment
Due to unavailability of the unit commitment and dispatch algorithm required to
change unit commitment and dispatch on the system, the costs of wind variability
in the unit commitment time-frame could not be calculated in this study.
5.4.3 Sample Test System
A test system in the e-terra simulator consisting of three distinct but synchronous
control areas, each containing both conventional generation resources and load, with a
collective generating capacity of just over 12,000 MW was used for the study. Control
area A is interconnected to control area B and control area C is interconnected directly
to control area B, as illustrated in Fig. 5.1. However, there is no direct interconnection
between control areas A and C. Both control areas A and C also each contained a wind
farm and the capacity of each could be varied between 0 and 400 MW. Control Area
A is the home control area, so the incremental cost curves of each generator were
available. However for generation within other control areas (B and C) costs were not
available.
Large thermal generators, several OCGT and CCGT generators and a sizeable amount
of hydroelectric generation comprised the conventional generation in the three control
areas. The generation capacity for each control area is given in Table 5.1. Each
generator is equipped with automatic generation control (AGC). AGC is congured
to maintain system frequency at 60 Hz or in the case of an interconnected system,
maintain frequency at 60 Hz and interconnector ows to the contracted values.
Table 5.1: Test system generation capacity
Control Area A Control Area B Control Area C
Steam Units (MW) 3200 1400 4200
Hydroelectric (MW) 1590 - 1300
OCGTs (MW) - 900 -
CCGTs (MW) 140 - -
Figure 5.1: Test system
5.4.4 Scenario
The test system was set up to examine the eects of wind generation variability on
load-following capabilities and costs of conventional generators, system frequency, in-
terconnector ows and area control error. Table 5.2 gives the load values and wind
farm capacities for the three control areas in the test system. The ow scheduled on
Interconnector AB is 971 MW, from A to B, and the ow scheduled on Interconnector
BC is 705 MW, from C to B.
As the system in use is a test system, no actual load data is available and for the
results presented here, load in each control area is assumed constant. As a result, all
AGC commands issued to conventional generators will be a direct result of changing
wind generation. Although the variability of load is neglected, the results provide a
good illustration of how conventional generation must respond to variability on the
system. In studies of an actual power system, existing load data or load forecasts can
easily be incorporated alongside the actual or forecast wind data and system operating
conditions for that system.
Table 5.2: Test system set-up
Control Area A Control Area B Control Area C
Load (MW) 950 3000 2000
Wind Farm Capacity (MW) 400 - 400
The wind data used for this study is two wind power time series of 5 minute resolution
for 24 hours captured during stormy conditions, as described in Section 5.3.2, each
representative of a 400 MW capacity wind farm. The two time series are illustrated
in Fig. 5.2, and have a correlation coecient of 0.272. The time series from each wind
farm is the power output of a single wind turbine at that location, scaled conservatively
up to 400 MW. No smoothing eects are taken into account, and changes in wind power
output from a wind farm will generally not be as variable as for a single generator. In
the time series for wind farm A, the power output goes from maximum to zero during
one ve minute interval. Although extreme, this scenario is very possible for an entire
wind farm, as a storm front moves through the farm, and wind speed exceeds the
cut-out speed for each generator.
In order to assess the eects on the test system of each wind farm individually and
also both generating at the same time, four dierent scenarios were considered. For
the base case, case (i), each wind farm is generating at constant power output, which
is the average power output for the 24 hour time period in question. The average
power output values are 193 MW and 131 MW for wind farms A and C respectively.
Then cases (ii), (iii) and (iv) examine the eect of variable power output from Wind
Farm A, then Wind Farm C and nally both wind farms. The four dierent cases are
summarised in Table 5.3.
Table 5.3: Test case scenarios
Wind Farm A Wind Farm C
Case (i) Constant Power Constant Power
Case (ii) Constant Power Actual Time Series
Case (iii) Actual Time Series Constant Power
Case (iv) Actual Time Series Actual Time Series
0 6 12 18 24
0
50
100
150
200
250
300
350
400
Time (Hrs)
P
o
w
e
r

(
M
W
)
Wind Plant C
Wind Plant A
Figure 5.2: Wind farm power output time series
The parameters monitored in the study were
System Frequency
Area Control Error
Generator MW Power Output
Power ow on the Interconnectors
Control Signals
Regulation
Costs (Generator Incremental Cost, Average Cost, Hourly Cost)
5.4.5 Results
The eect of wind power variability on frequency is plotted in Fig. 5.3, and clearly
frequency follows closely the net variations in wind generation. When both wind farms
are varying, the frequency will be inuenced by the net power generation from wind, not
the individual power generation. As a result, total frequency variations can sometimes
be lower than the frequency deviations for individual farms (This is illustrated between
hours 14 and 17). In the results of the analysis illustrated here, correlation between
wind farm conditions most of the time results in larger variations in frequency for case
(iv) than for either cases (ii) or (iii), where only one wind farm had variability in the
power output. However, it is important to note that only two wind farms are studied
here, and wind data is reasonably highly correlated at certain times. Greater diversity
would result in more smoothing of frequency eects.
0 6 12 18 24
60
60.02
60.04
60.06
60.08
60.1
Time (hrs)
F
r
e
q
u
e
n
c
y

(
H
z
)
Case (i)
Case (ii)
Case (iii)
Case (iv)
Figure 5.3: System frequency
The ACE for each case is presented for Control Area A in Fig. 5.4, Control Area B in
Fig. 5.5 and Control Area C in Fig. 5.6. The ACE is shown to be highly dependent on
the wind power variability within the control area in question, and less dependent on
the wind power variability of a wind farm located in another control area. However, it
is clear that a wind farm in a neighbouring control area will aect ACE, demonstrating
that wind has an impact on the ACE of neighbouring systems as well as the ACE of
the control area within which it is located. This is clearly illustrated in Fig. 5.5, where
Control Area Bs ACE deviates from zero, even though no wind farm is located within
this control area.
0 6 12 18 24
40
20
0
20
40
60
80
100
120
140
Time (Hrs)
A
C
E

(
M
W
)
Case (i)
Case (ii)
Case (iii)
Case (iv)
Figure 5.4: ACE: Control Area A
The active power output from a sample conventional synchronous generator in each
control area was examined in Fig. 5.7, Fig. 5.8 and Fig. 5.9. It is clear that a much
larger response from conventional generation is required when a wind farm with varying
output is located within the same control area (e.g. the response of generator A for
Cases (iii) and (iv), and response of Generator C for Cases (ii) and (iv)). However, the
power output from each generator is also aected by wind generation connected outside
the immediate control area, albeit to a lesser extent. This is illustrated in Fig. 5.8,
when the power output from Generator B changes, even though there is no wind farm
located within Control Area B. This is a result of AGC being activated due to changes
in the power ow over the interchange and deviations in system frequency.
The corresponding AGC control signals to the generator power outputs in Fig. 5.7,
Fig. 5.8 and Fig. 5.9 were also examined. The control signals were found to be highly
0 6 12 18 24
40
20
0
20
40
60
80
100
120
140
Time (Hrs)
A
C
E

(
M
W
)
Case (i)
Case (ii)
Case (iii)
Case (iv)
Figure 5.5: ACE: Control Area B
correlated with the generator active power changes, as expected. The signals have a
greater magnitude when the generator is located within the same control area as the
wind farm with variable output, which is consistent with the fact that great changes
in generation are required for such situations. However, there is little dierence in
the number of control signals sent to each generator. This implies that all generators
receive control signals, but the magnitude is dependent on the control area within which
they are located. This can be accounted for by the fact that AGC control signals are
a function of the control area ACE, which is dependent on both system frequency and
the interconnector ow. Therefore, when ACE is high, control signals tend to be high.
Conversely, if the control signals for AGC were dependent on frequency alone, there
would be no dierence in control signals to a given generator if the variable wind farm
was located within or outside of the generators control area.
0 6 12 18 24
40
20
0
20
40
60
80
100
120
140
Time (Hrs)
A
C
E

(
M
W
)
Case (i)
Case (ii)
Case (iii)
Case (iv)
Figure 5.6: ACE: Control Area C
5.5 Discussion and Conclusion
A review of a number of wind integration studies was undertaken. As a result, a number
of key elements that inuence the accuracy and validity of wind integration studies can
be concluded. The method of representation of wind generation in the study and the
quality, quantity and type of available wind data will play a major role in the accuracy
of the results. Inaccuracies will arise through imprecise scaling of wind power data,
the use of wind speed instead of actual wind power data and neglect of the eects of
geographical dispersion. High quality and quantity of system data, in particular for
system load, is also vital.
The results of each wind integration study are specic to the power system under
consideration. However, some general conclusions can be drawn from results common
to a large number of studies. The balancing of the entire power system is required
to maintain frequency at nominal system frequency, so each variation in wind power
does not need to be matched by an equal and opposite variation in another resource.
The additional costs imposed on a power system with increasing wind penetration
0 6 12 18 24
320
330
340
350
360
370
380
390
Time (Hrs)
P
o
w
e
r

(
M
W
)
Case (i)
Case (ii)
Case (iii)
Case (iv)
Figure 5.7: Generator power output: Control Area A
will increase with increasing time-frame. As wind penetration increases, the reserve
requirements increase more rapidly and the rate of change of power variations on the
power system in addition to the magnitude of power variations may increase.
A preliminary methodology for a wind integration frequency control study was devel-
oped, which identies and describes the key elements. In order to accurately capture
wind turbine dynamics to both wind uctuations and power system conditions, dy-
namic models of wind turbine generators that are valid in over a long time-frame are
required.
An assessment of the suitability of the e-terra simulator as a platform to carry out such a
wind integration study over a range of timescales is carried out. The platform can assess
frequency control in all time-frames in excess of 5 seconds, although the incorporation
of the e-terra generation platform (a unit commitment platform) is necessary for the
unit commitment time-frame. Analysis indicates that the eect of wind generation on
system frequency and operating conditions within the control area and in neighbouring
control areas can be assessed using the e-terra simulator. However, with the current
representation of wind farms as active power time series, the eects of changing system
0 6 12 18 24
450
460
470
480
490
500
510
Time (Hrs)
P
o
w
e
r

(
M
W
)
Case (i)
Case (ii)
Case (iii)
Case (iv)
Figure 5.8: Generator power output: Control Area B
conditions on the wind farm, and the subsequent interactions can not be modelled.
Some preliminary illustrative results of the impact of actual wind data during stormy
conditions on the system frequency and ACE were presented.
0 6 12 18 24
540
550
560
570
580
590
600
610
620
630
Time (Hrs)
P
o
w
e
r

(
M
W
)
Case (i)
Case (ii)
Case (iii)
Case (iv)
Figure 5.9: Generator power output: Control Area C
Chapter 6
Conclusions
6.1 Synopsis
A validated dynamic model of the isolated electricity system of Ireland, suitable for
the study of frequency control during a power imbalance, is developed in Chapter 2.
With just a single HVDC interconnection to neighbouring electricity systems, which
is not congured to provide primary operating reserve during a low frequency event,
the Ireland electricity system is essentially isolated from the perspective of short term
frequency control.
Each generating unit on the system, with the exception of wind and hydroelectric
generation, is individually modelled and tuned with system data recorded during a
number of low frequency events over the last ve years. The droop characteristics of
some generators on the Ireland system were found to vary, signicantly in several cases,
from the 4% droop characteristic required by the system operators (ESBNG, 2005b;
SONI, 2003a). The system load is represented with a measurement-based dynamic
load model (Welfonder et al., 1989; OSullivan and OMalley, 1996), tuned to actual
system event data. This load model provides a good representation of the frequency
sensitivity of the load during low frequency events on the system.
The system model shows good agreement when the simulated system frequency is com-
pared to actual system frequency events that occurred on the Ireland electricity system.
Thus validated, the system model can be used to predict the frequency behaviour of
the Ireland electricity system during a loss of a generation event. Accurate prediction
Chapter 6. Conclusions 135
of system frequency behaviour has a number of applications in addition to system se-
curity and stability assessment. The frequency characteristics of the Ireland electricity
system derived from the validated system model, used in conjunction with system op-
erator experience, are subsequently used to develop a minimum frequency constraint
for a competitive electricity market dispatch (Appendix C).
A validated model of a combined cycle gas turbine is developed which incorporates
the eects of fast acting inlet guide vanes, ambient temperature and pressure and a
mechanism for varying the frequency dependency of the gas turbine. This CCGT
model is combined with the validated system model to study the impact of CCGTs on
frequency control following a loss of generation event. During such a frequency event,
the frequency nadir was found to fall further as the proportion of base-loaded CCGTs
on the system increased. However, the eect on system frequency depends not only
on the proportion of base-loaded CCGTs, but also on the magnitude of the generation
provided by these units. As the magnitude of CCGT generation increases the eect on
the frequency nadir becomes greater (Appendix B).
Wind turbine generator models for two dierent technologies, xed speed and DFIG
WTGs, are incorporated into the predicted 2010 Ireland system model, to examine
the eect of increasing wind penetration on system frequency control during a loss
of generation event. Wind turbine generators are not currently congured to provide
response to changing system frequency. However, the inertial response of each WTG
technology diers signicantly from the inertial response of a conventional synchronous
generator. As a consequence, rates of change of system frequency will increase as
wind generation displaces an equivalent amount of conventional generation. Results
indicate that the maximum rate of change of frequency following a loss of generation
is independent of wind turbine technology. As xed speed wind turbines displace
conventional synchronous generation, although maximum rate of change of frequency
increases, there is no signicant change in minimum frequency reached following a loss
of generation. This is due to the inertial response provided by xed speed wind turbines,
which is distinctly slower than the inertial response of a synchronous generator. As
DFIG wind turbines displace conventional synchronous generation, the frequency nadir
following a loss of generation reduces. This is due to the negligible inertial response
provided by DFIG wind turbines.
One possible solution to the reduced system inertial response as proportions of WTGs
increase is outlined. Through the addition of a supplementary control loop to the
DFIG WTG, it is possible to provide a response similar to the inertial response a
conventional synchronous generator. The achievable response of a standard DFIG
WTG is constrained by technical limitations, such as current limits and the converter
rating. In this study it was assumed that these limits were not exceeded. In practise, the
implementation of the supplementary control loop and the maintenance of operational
constraints within required limits may be problematic and some design modication,
such as an increase of DFIG converter rating, may be required (Appendix D).
A review of a number of studies, which assess the operational impacts of increasing
wind integration on power systems, in particular frequency control, was carried out.
Subsequently, a preliminary methodology for assessing the impact of wind generation
on frequency control over a range of time-frames was developed. The application
of this methodology employing the Areva T&D e-terra simulator program was then
investigated. This work was carried out during a three month industry placement with
Areva T&D.
6.2 Conclusions
Proportions of both CCGT and wind turbine generators are increasing on power sys-
tems worldwide, and the consequences of these trends on system frequency control of
an island electricity system have been examined in this thesis.
Results indicate that increased frequency excursions from nominal system frequency
will occur with increasing amounts of base-loaded CCGTs on the system. Large fre-
quency excursions result in a degradation of power quality and can have a serious
impact on power system components. Many components, including generating units,
are rated to operate at nominal system frequency and thermal and mechanical damage
can occur during times of low system frequency. Many are therefore equipped with
protection devices. The tripping of components from the system through protection
devices will result in interruption to customers and can impact on system reliability
standards.
The margin between minimum frequency and the on-set frequency level for UFLS
will be reduced as frequency excursions increase. If operating with reduced security
margins, a power system will be more susceptible to disturbance, resulting in a higher
likelihood of customer interruption.
As increasing amounts of wind generation displace conventional generation, results in-
dicate that the rate at which frequency changes during a power imbalance will increase.
Increased rates of change of frequency, particularly during periods of low system load,
could potentially lead to the activation of generator ROCOF protection relays. This
would result in generators tripping o the system, exacerbating the power imbalance,
as experienced recently on the Ireland electricity system (Appendix A). Increased
rates of change of frequency can also result in increased frequency excursions, as the
frequency falls further before being arrested by the POR of the system. The addition
of a supplementary controller to the DFIG WTG is one possible solution to prevent
erosion of system inertial response as DFIG WTGs displace conventional synchronous
generators.
With increasing levels of base-loaded CCGT and wind generation penetration com-
bined, the eects on system frequency control could be signicant. Faster rates of
change of frequency and larger frequency excursions may require system operators to
modify existing reserve policies to maintain system security standards. For example,
increasing DFIG wind turbine usage may necessitate additional availability of static re-
serve to maintain system frequency above a given threshold. In addition, the activation
of non-governor POR through ROCOF relays could be assessed as a possibility.
If no modication to reserve policy is made, increasing wind and base-loaded CCGT
generation could result in faster and larger frequency excursions, with an increasing
likelihood of load shedding due to reduced security margins.
The method of wind representation and the quality, quantity and type of wind data
will inuence the accuracy of any wind integration study. Wind generation will impact
on system frequency control over the range of time frames from regulation to unit
commitment, and future work is required in this area to accurately dene these impacts.
6.3 Scope for future work
The measurement based dynamic load model employed in the validated system model
is tuned to represent the average load behaviour in response to low frequency events.
However, as the proportions of dierent load components vary with time of day, week
and year, the frequency responsiveness of any load is dependent on load mix. Sucient
data for tuning the load model accurately for multiple dierent system conditions was
not available for this study. However, in order to further increase the accuracy of the
system model for all times of day and season, load model tuning for dierent system
conditions would be advantageous. The monitoring of system load at a number of
distinct locations with appropriate equipment would provide additional high resolution
load data. This data, captured for a number of system frequency events over of a
number of dierent times of day, week and year would be necessary to achieve a high
level of load model accuracy. As most standard methods of system stability and security
assessment involve the modelling of extreme events, for greater accuracy, data gathering
at times of extreme system conditions, such as annual minimum and peak load periods,
would provide more accurate load measurements during these periods. Consequently,
increased condence could be placed in the system load model.
At present, wind turbine generators are not typically congured to respond to low
system frequency like conventional synchronous generator. The addition of controllers
to achieve a response similar to the response from conventional synchronous generators
is a possible future development. Investigation to evaluate the potential for its use and
subsequent impacts on system frequency would be necessary.
The impacts of wind generation on short-term frequency control have been investigated
in this thesis. A preliminary methodology for a study into the impact of wind integra-
tion on frequency control over broader range of time-scales is described in Chapter 5.
Additional wind data, in conjunction with a unit commitment platform, would enable
implementation and assessment of this methodology in the future. The impact on fre-
quency control from the regulation to unit commitment time-scales could be examined,
and the relevant costs of increasing wind penetration extracted.
Aero-derivative gas turbines, with their high exibility, are increasingly being proposed
as a possible solution to oset the variability of wind. The combination of exible gas
turbine technology with the variable wind generation could result in a much reduced
net variability. However, aero-derivative gas turbines have a very low inertial con-
stant in comparison with conventional synchronous generation. The displacement of
conventional generation with increasing proportions of wind turbines in conjunction
with aero-derivatives could result in an erosion of system inertia, with similar impacts
on rate of change of frequency as those of increasing proportions of wind generation.
Further analysis of the dynamics of aero-derivative gas turbines and the eect that
increasing proportions of these units will have on system frequency is necessary before
further conclusions may be drawn.
As the generating plant mix of a power system evolves the frequency response charac-
teristic also evolves. In addition to the system inertia varying with dierent generation
resources, dierent sources of POR will also result in a dierent system frequency char-
acteristic in the event of a power imbalance. While increasing amounts of static POR
from demand through schemes such as the interruptible customer scheme on the Ire-
land electricity system will help to arrest the falling frequency, they will not necessarily
aid in the recovery of system frequency to within acceptable limits around nominal fre-
quency. In addition, as increasing distributed generation comes onto the system, the
use of UFLS may become less appropriate, as the loss of distributed generation in ad-
dition to load may occur upon activation of these relays. Therefore, the investigation
of an optimal mix of POR would be useful, to examine the contribution of dierent
technologies to short-term frequency control. Subsequent to this, an investigation of
an appropriate market structure to incentivise the evolution towards this optimal mix
of POR would be required.
References
Areva, 2004. e-terrageneration Programmer/Analyst Guide. Tech. rep., AREVA
T&D Energy Automation & Information Corporation.
Asean Energy, 2005.
URL http://www.aseanenergy.org/
Bagnasco, A., Delno, B., Denegri, G. B., Massucco, S., 1998. Management
and dynamic performances of combined cycle power plants during parallel and
islanding operation. IEEE Transactions on Energy Conversion 13 (2), 194201.
Beddoes, A., Thomas, P., Gosden, M., 2005. Loss of mains protection relay
performances when subjected to network disturbances/events. In: CIRED Inter-
national Conference on Electricity Distribution 2005. Turin, Italy.
Bialek, J., 2003. Are blackouts contagious? IEE Power Engineer 17, 1013.
Blaschke, F., 1972. The principle of eld orientation as applied to the new
transvector closed loop control system in a PWM inverter induction motor drive.
Siemens Review 39 (5), 217220.
Bohrenkmper, G., Reiermann, D., Hhne, G., Lingner, U., 2004. Technology
evolution of the proven gas turbine models V94.2 and V84.2 for new units and
service retrots. In: Power-Gen Europe 2004.
URL http://www.powergeneration.siemens.com/download/pool/Bohrenkaemper
Gerhard.pdf
Brooks, D., Lo, E., Smith, J., Pease, J., McGree, M., 2002. Assessing the impact
of wind generation on system operations at Xcel Energy North and Bonneville
Power Administration. In: WindPower 2002. Washington D.C.
140
References 141
Brooks, D., Lo, E., Zavadil, R., Santoso, S., Smith, J., 2003. Characterizing the
impacts of signicant wind generation facilities on bulk power system operations
planning. Case study, Electrotek Concepts, Inc., Arlington, VA, USA.
CER, 2004a. Commission for Energy Regulation (CER). List of Licensees.
URL http://www.cer.ie/list of licensees details.asp?LicenceType=1
CER, 2004b. Commission for Energy Regulation (CER). Wind Farm Transmis-
sion Grid Code Provisions.
URL http://www.cer.ie/CERDocs/cer04237.pdf
CER, 2004c. Commission for Energy Regulation (CER). Wind Generation Dis-
tribution Code Provisions.
URL http://www.cer.ie/CERDocs/cer04318.pdf
Chien, K., Ergin, E., Ling, C., Lee, A., 1958. Dynamic analysis of a boiler.
Transactions of the ASME 80, 18091819.
CIGRE, 2003. CIGRE Taskforce 38.02.25. Modeling of gas turbines and steam
turbines in combined cycle power plants. CIGRE Technical Report.
Cohen, H., Rogers, G. F. C., Saravanamuttoo, H. I. H., 1996. Gas Turbine Theory
(4rd Edition). Longman.
Concordia, C., Fink, L. H., Poullikkas, G., 1995. Load shedding on an isolated
system. IEEE Transactions on Power Systems 10 (3), 14671472.
DeMello, F. P., 1991. Boiler models for system dynamic performance studies.
IEEE Transactions on Power Systems 6 (1), 6674.
DETINI, 2003. Department of Enterprise Trade and Investment of Northern Ire-
land (DETINI). Report on the NIE response to the loss of two generating units
(520mw) at Kilroot power station on Thursday 4 December 2003.
Doherty, R., 2005. New methods for planning and operating modern electricity
systems with signicant wind generation. Ph.D. thesis, University College Dublin,
Dublin, Ireland.
Doherty, R., Denny, E., OMalley, M., 2004. System operation with a signicant
wind power penetration. In: IEEE Power Engineering Society General Meeting.
Denver, Colorado.
References 142
Doherty, R., Lalor, G., OMalley, M., 2005. Frequency control in competitive
electricity market dispatch. IEEE Transactions on Power Systems 20 (3), 1588
1596.
Ekanayake, J. B., Holdsworth, L., Wu, X., Jenkins, N., 2003. Dynamic modelling
of doubly fed induction generator wind turbines. IEEE Transactions on Power
Systems 18 (2), 803809.
Ekanayake, J. B., Jenkins, N., 2004. Comparison of the response of doubly fed and
xed-speed induction generator wind turbines to changes in network frequency.
IEEE Transactions on Energy Conversion: Accepted for future publication PP,
13.
Elgerd, O. I., 1982. Electric Energy Systems Theory: An Introduction. Second
Edition. McGraw-Hill, Inc.
ELTRA, 2000. Specications for Connecting Wind Farms to the Transmission
Network.
URL http://www.eltra.dk/composite-837.htm
ESBNG, 2003. ESB National Grid (ESBNG). Generation Adequacy Report: 2004
- 2010.
URL www.eirgrid.com
ESBNG, 2004a. ESB National Grid (ESBNG). Forecast Statement 2004-2010.
URL www.eirgrid.com
ESBNG, 2004b. ESB National Grid (ESBNG). Generation Adequacy Report:
2005 - 2011.
URL www.eirgrid.com
ESBNG, 2004. ESB National Grid (ESBNG). Impact of Wind Power Generation
in Ireland on the Operation of Conventional Plant and the Economic Implica-
tions.
URL http://www.eirgrid.com
ESBNG, 2005a. ESB National Grid Download Centre.
URL http://www.eirgrid.com/EirGridPortal/DesktopDefault.aspx?tabid=
Download%20Centre
References 143
ESBNG, 2005b. ESB National Grid (ESBNG). Grid Code Version 1.2.
URL http://www.eirgrid.com/
ESBNG, 2005a. ESB National Grid Pre Conference Workshop on Wind:
Tuesday, 23rd August 2005: Introduction.
URL http://www.eirgrid.com/EirGridPortal/uploads/General%20Documents/
Introduction%20Paul%20Smith%202005.pdf
ESBNG, 2005b. ESB National Grid: System Operator of Republic of Ireland.
URL http://www.eirgrid.com/
ESBNG, 2005c. ESB National Grid, (TSO Ireland). Interconnector Status.
URL http://www.eirgrid.ie/EirGridPortal/DesktopDefault.aspx?tabid=
Interconnector%20Status&TreeLinkModID=1451&TreeLinkItemID=16
EU Directive:77, 2001. Directive 2001/77/EC of the European Parliament and
the Council of 27 September 2001 on the promotion of electricity produced from
renewable energy sources in the internal electricity market. Ocial Journal of
the European Communities, L283, 3340.
EWEA, 2003. The European Wind Energy Association (EWEA) and the Euro-
pean Commissions Directorate General for Transport and Energy (DG TREN).
Wind Energy - The Facts.
URL http://www.ewea.org/06projects events/proj WEfacts.htm
Fink, L., Carlsen, K., 1978. Operating under stress and strain. IEEE Spectrum
15 (3), 4853.
Flynn, D., 2003. Thermal Power Plant Simulation and Control. The Institution
of Electrical Engineers, London.
Flynn, M., OMalley, M., 1999. A drum boiler model for long term power system
dynamic simulation. IEEE Transactions on Power Systems 14 (1), 209217.
Fox, B., McCartney, A., 1988. Emergency control of frequency on the NIE system.
Power Engineering Journal, 195201.
Fox, B., Thompson, J., Tindall, C., 1989. Adaptive control of load shedding
relays under generation loss conditions. In: Fourth International Conference on
Developments in Power Protection. Edinburgh, Scotland, pp. 259263.
References 144
GE, 2004. GE Power Systems Energy Consulting. The Eects of Integrating
Wind Power on Transmission System Planning, Reliability and Operations: Pre-
liminary Overall Reliability Assessment. Tech. rep.
GE, 2005. GE Power Systems Energy Consulting. The Eects of Integrating Wind
Power on Transmission System Planning, Reliability and Operations: System
Performance Evaluation. Tech. rep.
Hajagos, L. M., Berube, G. R., 2001. Utility experience with gas turbine testing
and modeling. IEEE Power Engineering Society Winter Meeting 2, 671677.
Hampton, D., Tindall, C., McArdle, J., 1991. Emergency control of power system
frequency using ywheel energy injection. In: Proceedings of IEE International
Conference on Advances in Power System Control, Operation and Management.
Hong Kong, pp. 662667.
Hannett, L. N., Feltes, J. W., 2001. Testing and model validation for combined
cycle power plants. IEEE Power Engineering Society Winter Meeting 2, 664670.
Hatziargyriou, N., Contaxis, G., Matos, M., Lopes, J. P., Kariniotakis, G., Mayer,
D., Halliday, J., Dutton, G., Dokopoulos, P., Bakirtzis, A., Stefanakis, J., Gi-
gantidou, A., ODonnell, P., McCoy, D., Fernandes, M., Cotrim, J., A.P.Figueira,
2002. Energy management and control of island power systems with increased
penetration from renewable energy resources. In: IEEE Power Engineering So-
ciety Winter Meeting. Vol. 1. pp. 335 339.
Hatziargyriou, N., Contaxis, G., Matos, M., Nogaret, E., Papadopoulos, M.,
Papadias, B., Lopes, J. P., Kariniotakis, G., Halliday, J., Dokopoulos, P., An-
droutsos, A., Stefanakis, J., Dutton, G., Bakirtzis, A., Gigantidou, A., 2000.
Operation and control of island systems - the crete case. In: IEEE Power Engi-
neering Society Winter Meeting. Vol. 2. pp. 10531056.
Hirst, E., 2002a. Integrating wind energy with the BPA power system: Prelim-
inary study. Tech. rep., For: Bonneville Power Administration, Portland, OR,
USA.
Hirst, E., 2002b. Integrating wind output with bulk power operations and whole-
sale electricity markets. Wind Energy 5 (1), 1936.
References 145
Holdsworth, L., Wu, X., Ekanayake, J., Jenkins, N., 2003. Comparison of xed
speed and doubly-fed induction wind turbines during power system disturbances.
IEE Proceedings on Generation, Transmission and Distribustion 150 (3), 343
352.
Hung, W., 1991. Dynamic simulation of a gas-turbine generating unit. Proceedings
of the IEE, Part C 138 (4), 342350.
Hung, W., 2001. Frequency control issues in the England and Wales electricity
grid. GMA Conference on Development and Interaction of Generators and Power
Networks in an Energy De-Regulating Environment.
Hur, N., Jung, J., Nam, K., 2001. A fast dynamic DC-link power-balancing
scheme for a PWM converter-inverter system. IEEE Transactions on Industrial
Electronics 48 (4), 794803.
IEEE, 1973a. IEEE Committee Report. Dynamic models for steam and hydro
turbines in power system studies. IEEE Transactions on Power Apparatus and
Systems PAS-92 (6), 19041915.
IEEE, 1973b. IEEE Working Group on Power Plant Response to Load Changes.
MW Response of Fossil Fueled Steam Units. IEEE Transactions on Power Ap-
paratus and Systems PAS-92, 455463.
IEEE, 1991. IEEE Working Group on Prime Mover and Energy Supply Models
for System Dynamic Performance. Dynamic models for fossil fueled steam units
in power system studies. IEEE Transactions on Power Systems 6 (2), 753761.
for System Dynamic Performance Studies. Hydraulic turbine and turbine control
models for system dynamic studies. IEEE Transactions on Power Systems 7 (1),
167179.
IEEE, 1993. IEEE TaskForce: Load representation for dynamic performance
analysis. IEEE Transactions on Power Systems 8 (2), 472482.
for System Dynamic Performance Studies. Dynamic models for combined cycle
plants in power system studies. IEEE Transactions on Power Systems 9 (3),
16981708.
References 146
IEEE/CIGRE, 2004. IEEE/CIGRE Joint Task Force on Stablity Terms and De-
nitions. Denition and classication of power system stability. IEEE Transactions
on Power Systems 19 (2), 13871401.
Jauch, C., Sorensen, P., Bak-Jensen, B., 2004. International review of grid con-
nection requirements for wind turbines. In: Nordic Wind Power Conference.
Chalmers Univeristy of Technology, Sweden, pp. 16.
URL http://www.elkraft.chalmers.se/Publikationer/EMKE.publ/NWPC04
/papers/JAUCH.PDF
Kim, T. S., 2004. Comparative analysis on the part load performance of combined
cycle plants considering design performance and power control strategy. Energy
29, 7185.
Kirby, B., Dyer, J., Martinez, C., Shoureshi, R., Guttromson, R., Dagle, J.,
2002. Frequency control concerns in the North American Electric Power System.
Tech. Rep. ORNL/TM-2003/41, Oak Ridge National Laboratory, Oak Ridge,
Tennessee, USA.
Kottick, D., Blau, M., Edelstein, D., 1993. Battery energy storage for frequency
regulation in an island power system. IEEE Transactions on Energy Conversion
8 (3), 455459.
Kottick, D., Or, Q., 1996. Neural-networks for predicting the operation of an
under-frequency load shedding system. IEEE Transactions on Power Systems
11 (3), 13501358.
Kundur, P., 1994. Power System Stablity and Control. McGraw-Hill, Inc.
Kunitomi, K., Kurita, A., Okamoto, H., Tada, Y., Ihara, S., Pourbeik, P., Price,
W. W., Leirbukt, A. B., Sanchez-Gasca, J. J., 2001. Modeling frequency depen-
dency of gas turbine output. IEEE Power Engineering Society Winter Meeting
2, 678683.
Kunitomi, K., Kurita, A., Tada, Y., Ihara, S., Price, W. W., Richardson, L. M.,
Smith, G., 2003. Modeling combined-cycle power plant for simulation of fre-
quency excursions. IEEE Transactions on Power Systems 18 (2), 724729.
Kwan, H., Anderson, J., 1970. A mathematical model for a 200W boiler. Inter-
national Journal of Control 12 (6), 977998.
References 147
Lalor, G., OMalley, M., 2003. Frequency control on an island power system
with increasing proportions of combined cycle gas turbines. In: IEEE Powertech
Conference. Vol. 4. Bologna, Italy, pp. 228234.
Lalor, G., Ritchie, J., Flynn, D., OMalley, M., 2005. The impact of combined-
cycle gas turbine short-term dynamics on frequency control. IEEE Transactions
on Power Systems 20 (3), 14561464.
Machowski, J., Bialek, J., Bumby, J., 1997. Power System Dynamics and Stabil-
ity. John Wiley & Sons, Chichester, England.
Mak, T., Law, C., 1991. Spinning reserve and under-frequency load shedding
strategies for the interconnected china light power system. In: Proceedings of
IEE International Conference on Advances in Power System Control, Operation
and Management. Hong Kong, pp. 542548.
Mangan, P., 2005. Security of Supply Issues in an isolated system: the island of
Ireland. Tech. rep., ESB National Grid.
URL http://www.eirgrid.com/EirGridPortal/uploads/General%20Documents/
Pat%20Mangan%202005.pdf
Manwell, J., McGowan, J., Rogers, A., 2002. Wind Energy Explained: Theory,
Design and Application. John Wiley & Sons Ltd, Chichester, England.
McDonald, J., Kwatny, H., 1970. A mathematical model for reheat boiler-turbine-
generator systems, IEEE Paper 70 CP221 PWR. In: IEEE Power Engineering
Society Winter Meeting. New York.
Milligan, M., 2003. Wind power plants and system operation in the hourly time
domain. In: WindPower 2003. Austin, Texas, USA.
Mullane, A., 2005. Personal Communication.
Mullane, A., OMalley, M., 2005. The inertial-response of induction-machine
based wind-turbines. IEEE Transaction on Power Systems 20 (3), 14961503.
Mullane, A. P., 2004. Advanced Control of Wind Energy Conversion Systems.
Ph.D. thesis, National University of Ireland, University College Cork, Ireland.
NERC, 2004. North American Electric Reliability Council. Technical Analysis of
the August 14, 2003 Blackout.
URL www.nerc.com
References 148
NERC, 2005. North American Electric Reliability Council (NERC), Operating
Manual: Standard BAL-001-0 Real Power Balancing Control Performance.
URL ftp://www.nerc.com/pub/sys/all updl/standards/rs/BAL-001-0.pdf
NGC, 2003. National Grid Company PLC. Investigation into the Loss of Supply
Incident aecting parts of South London at 18.20 on Thursday, 28 August 2003.
URL www.nationalgrid.com/uk/
Norgaard, P., Holttinen, H., 2004. A multi-turbine power curve approach. In:
Nordic Wind Power Conference. Chalmers University of Technology.
Ordys, A. W., Pike, A. W., Johnson, M. A., Katebi, R. M., Grimble, M. J., 1994.
Modelling and Simulation of Power Generation Plants. Springer-Verlag, London.
OSullivan, J., OMalley, M., 1996. Identication and validation of dynamic global
load model parameters for use in power system simulation. IEEE Transactions
on Power Systems 11, 851857.
OSullivan, J., OMalley, M., 1999. A new methodology for the provision of re-
serve in an isolated power system. IEEE Transactions on Power Systems 14 (2),
519524.
OSullivan, J., Power, M., Flynn, M., OMalley, M., 1999. Modelling of frequency
control in an island system. In: IEEE Power Engineering Society 1999 Winter
Meeting. Vol. 1. pp. 574579.
OSullivan, J. W., 1996. Modelling and identication of emergency reserve with
applications to isolated power systems. Ph.D. thesis, University College Dublin,
Dublin, Ireland.
PaciCorp, 2003. Integrated Resource Plan 2003.
URL http://www.pacificorp.com/Navigation/Navigation23807.html
Papazoglou, T. M., Gigandidou, A., 2003. Impact and benets of distributed wind
generation on quality and security in the case of the cretan EPS. CIGRE/IEEE
PES International Symposium on Quality and Security of Electric Power Delivery
Systems, 193197.
Parsons, B., Milligan, M., Zavadil, R., Brooks, D., Kirby, B., Dragoon, K., Cald-
well, J., 2003. Grid impacts of wind power: A summary of recent studies in the
united states. In: European Wind Energy Conference. Madrid, Spain.
References 149
Parsons, B., Wan, Y., Kirby, B., 2001. Wind farm power uctuations, ancillary
services and system operating impact analysis activities in the United States.
Tech. rep., NREL.
URL http://www.nrel.gov
Pourbeik, P., 2002. The dependence of gas turbine power output on system fre-
quency and ambient conditions. In: CIGRE Session 2002. No. 38-101. Paris,
France.
Pourbeik, P., Koessler, R. J., Dickmander, D. L., Wong, W., 2003. Integration of
large wind farms into utility grids (part 2 - performance issues). In: IEEE Power
Engineering Society General Meeting. Vol. 3. Toronto, Canada, pp. 15251528.
Ramey, D., Skooglund, J., 1970. Detailed hydro-governor representation for sys-
tem stability studies. IEEE Transactions on Power Apparatus and Systems PAS-
89 (1), 106112.
Rowen, W. I., 1983a. Gas turbine airow control for optimum heat recovery.
Transactions of the ASME 105, 7279.
Rowen, W. I., 1983b. Simplied mathematical representations of heavy-duty gas
turbines. ASME Journal of Engineering for Power 105 (1), 865869.
Rowen, W. I., 1992. Simplied mathematical representations of single shaft gas
turbines in mechanical drive service. Turbomachinery International, 2632.
SEI, 2004. Sustainable Energy Ireland (SEI). Operating Reserve Requirements
as Wind Power Penetration Increases in the Irish Electricity System.
URL http://www.sei.ie/uploads/documents/upload/publications/Ilex-Wind
-Reser rev2FSFinal.pdf
Siemens, 2001. Poolbeg GSE Add-On Combined Cycle and Auxiliary Plant: Per-
formance Test Report. Technical report, Siemens Power Generation (KWU).
Slootweg, J., Polinder, H., Kling, W., 2003. Representing wind turbine electrical
generating systems in fundamental frequency simulations. IEEE Transactions on
Energy Conversion 18 (4), 516524.
Smith, J., DeMeo, E., Parsons, B., Milligan, M., 2004. Wind power impacts on
electric power system operating costs: Summary and perspective on work to date.
In: WindPower 2004. Chicago, Illinois, USA.
References 150
SONI, 2003a. System Operator for Northern Ireland (SONI). Grid Code.
URL www.soni.ltd.uk/gridcode.asp
SONI, 2003b. System Operator for Northern Ireland (SONI). Seven Year Trans-
mission Statement 2003/04 - 2009/10. Tech. rep.
SONI, 2005. NI Wind Quarterly Report July 05.
URL http://www.soni.ltd.uk/upload/WindPower Quartly Report Jul 05.pdf
Tapia, A., Tapia, G., Ostolaza, J., Saenz, J., 2003. Modelling and control of
a wind turbine driven doubly fed induction generator. IEEE Transactions on
Energy Conversions 18 (2), 194204.
Thiringer, T., Luomi, J., 2001. Comparison of reduced-order dynamic models of
induction machines. IEEE Transactions on Power Systems 16 (1), 119126.
Thompson, J. G., Fox, B., 1994. Adaptive load shedding for isolated power sys-
tems. IEE Proceedings on Generation, Transmission and Distribution 141 (5),
491496.
Transpower, 2001. Transpower, Frequency Standards Working Group Prelimi-
nary Report to the Grid Security Committee. Performance of thermal plant at
low frequencies: Appendix 6.
URL http://www.gsp.co.nz/
UCTE, 2003. Union for the Coordination of Tranmission of Electricity (UCTE).
Interim report of the Investigation Committee on the 28 September 2003 Blackout
in Italy.
URL http://www.ucte.org (Publications)
UN, 1997. The United Nations Framework Convention on Climate Change. The
Kyoto Protocol.
URL http://unfccc.int/resource/docs/convkp/kpeng.pdf
Undrill, J. M., Garmendia, A., 2001. Modeling of combined cycle plants in grid
simulation studies. IEEE Power Engineering Society Winter Meeting 2, 657663.
Welfonder, E., Weber, H., Hall, B., 1989. Investigations of the frequency and
voltage dependence of load part systems using a digital self-acting measuring
and identication system. IEEE Transactions on Power Systems 4 (1), 1925.
References 151
Zavadil, R., King, J., Xiadong, L., Ahlstrom, M., Lee, B., Moon, D., Finley, C.,
Alnes, L., Jones, L., Hudry, F., Monstream, M., Lai, S., Smith, J., 2004. Wind
integration study - nal report: For Xcel Energy and the Minnesota Department
of Commerce. Tech. rep., Enernex Corporation and WindLogics, Inc.
Zhang, Q., So, P. L., 2000. Dynamic modelling of a combined cycle plant for
power system stability studies. IEEE Power Engineering Society Winter Meeting
2, 15381543.
Appendix A
Frequency Disturbance Event
On the 5
th
August 2005, a frequency disturbance event occurred on the Ireland elec-
tricity system, with a minimum system frequency of 48.41 Hz reached. The system
frequency during the event is illustrated in Fig. A.1.
0 100 200 300 400 500 600 700 800
48.4
48.6
48.8
49
49.2
49.4
49.6
49.8
50
Time (s)
F
r
e
q
u
e
n
c
y

(
H
z
)
Figure A.1: Recorded system frequency on Ireland electricity system
Appendix A. Frequency Disturbance 153
The power ow on the main connection between the NIE and ESB power systems
generaly ows from north (NIE) to south (ESB) under normal operating conditions. In
addition, the HVDC interconnection is generally importing active power from Scotland,
which enters the NIE system. Therefore, in the event of a loss of synchronism between
the ESB and NIE electricity systems, the NIE system will experience an active power
surplus and the ESB system will experience an active power decit. To prevent system
frequency from rising excessively on the NIE system, a rapid reduction in active power
is necessary in the event of the loss of the main NIE-ESB connection. Therefore, in
the event of a loss of the main NIE-ESB connection a signal is sent immediately to
the HVDC interconnection to reduce active power import by a pre-dened amount.
This occurs almost instantaneously, helping to prevent the NIE system frequency from
rising too high.
On August 5
th
, 2005, the NIE-ESB connection was operating with approximately
370 MW of active power owing from north to south and the HVDC interconnection
was importing approximately 115 MW (Mangan, 2005). At 10.22 hours, a spurious
signal was sent to the HVDC Interconnection from the NIE-ESB main connection, al-
though the NIE-ESB was operating normally. This signal instigated a step change in
the active power on the HVDC interconnection. The active power changed from im-
porting approximately 115 MW to exporting approximately 165 MW, a step change of
about 280 MW (Mangan, 2005). The main NIE-ESB connection remained in operation
and the system frequency on the synchronous Ireland power system fell to 49.52 Hz,
and POR recovered the system frequency to 49.86 Hz.
However, two minutes later, a spurious signal was once again issued from the main NIE-
ESB connection to the HVDC interconnection. As a result, there was a step change in
the active power ow on the HVDC interconnection of approximately 280 MW, from
exporting approximately 165 MW to exporting approximately 445 MW at maximum
(Mangan, 2005). The active power ow on the HVDC interconnection then settled at
the maximum rate active power ow of 400 MW. As a consequence, the frequency on
the Ireland electricity system declined rapidly. This resulted in two generating units,
with total generation of approximately 475 MW (ESBNG, 2005a), tripping from the
power system, due to the activation of ROCOF protection relays.
Therefore, the Ireland power system experienced a loss of active power of over 1000 MW
in a time-scale of just over ve minutes. Under-frequency load shedding relays tripped
Appendix A. Frequency Disturbance 154
load from the system and system frequency fall was arrested at a minimum of 48.41
Hz. The system frequency was recovered to 49.5 Hz within 11 minutes of the initial
event on the system.
Appendix B
The Impact of Combined Cycle Gas
Turbine Short Term Dynamics on
Frequency Control
156

The Impact of Combined Cycle Gas Turbine
Short Term Dynamics on Frequency Control

Gillian Lalor, Julia Ritchie, Damian Flynn and Mark J. OMalley

Abstract A model suitable for studying the short-term
dynamic response of a combined cycle gas turbine to a system
frequency deviation is developed. The model is used in
conjunction with a larger system model to study the impact of
increasing levels of CCGT generation on frequency control of a
small island system. The study considers single contingencies
and does not consider severe cascading type events. A
consequence of the results is that as the number and proportion
of base loaded combined cycle gas turbines increases, frequency
control may become more challenging. The results indicate that
with additional CCGTs on the system large frequency excursions
will be more likely and the transmission system operators on the
island of Ireland should review their frequency control strategies
in the future so as to avoid the shedding of customers.

Index Terms-- Frequency control, power generation control,
power system security, combined-cycle gas turbine.
I. NOMENCLATURE
a = Scaling factor for frequency sensitivity of gas turbine
exhaust temperature calculation
b = Constant, such that a + b = 1
igv = Inlet guide vane angle ()
K = 1/Droop
K
i
= Inlet guide vane controller constant (s)
L
igv
= Inlet guide vane position (per unit)
N = Speed (per unit)
P
a
= Ambient pressure (mbar)
P
gt
= Gas turbine power output (per unit)
SP = Set point (per unit)
T
a
= Ambient temperature (C)
T
cd
= Compressor discharge time constant (s)
T
gf
= Gas fuel time constant (s)
T
i
= Inlet guide vane controller integration rate (s)
T
igv
= Inlet guide vane actuator time constant (s)
T
r
= Rated turbine exhaust temperature (C)
T
s
= Speed controller time constant (s)
T
t
= Temperature controller integration rate (s)
T
v
= Valve positioner time constant (s)

This work has been conducted in the Electricity Research Centre,
University College Dublin, which is supported by the Electricity Supply
Board (ESB), ESB National Grid, Commission for Energy Regulation (CER),
Cylon Controls and Enterprise Ireland.
Gillian Lalor and Mark J. OMalley are with Electricity Research Centre,
Department of Electronic and Electrical Engineering, University College
Dublin, Ireland. (e-mail: gill.lalor@ee.ucd.ie, mark.omalley@ucd.ie).
Julia Ritchie and Damian Flynn are with the School of Electrical and
Electronic Engineering, The Queens University of Belfast, N. Ireland (e-
mail: j.a.ritchie@qub.ac.uk, d.flynn@qub.ac.uk.)
T
x
= Exhaust gas temperature (C)
W
f
= Fuel flow (per unit)
W
x
= Exhaust gas flow (per unit)
II. INTRODUCTION
T
he dynamic characteristics of combined cycle gas turbines
(CCGTs) have become an issue of considerable interest
over the last ten years. This is due to the increasing
proportions of CCGT plant that are being brought online in
the majority of electricity systems worldwide. Higher
efficiency, greater flexibility and lower emissions than many
conventional thermal generators, combined with progressively
shorter installation times and reducing installation costs, are
the basis for this move towards CCGT generation.
Understanding the dynamic behaviour of CCGT units is
crucial to maintaining system reliability and security in
electricity systems. This is particularly true as the move
towards competitive markets means system operators now
have little or no control over the type and location of new
plant investment.
Maintaining the standards of security and quality of supply
of any electricity system are of utmost importance to the
system operator. Recent system blackouts in countries such as
the USA, Canada, United Kingdom and Italy highlight the
importance of system security and reliability [1]-[3]. When an
incident occurs on the system, maintaining the system
frequency within the stipulated limits is a major priority, and
if these limits are breached, then the magnitude of the
excursion needs to be restricted, and the frequency returned to
within the limits as quickly as possible.
Of particular relevance to this study is the August 1996
blackout of Peninsular Malaysia that occurred as the result of
a serious generation loss. The electricity system in the
Peninsular and Sahab regions of Malaysia, operated by the
Tenaga Nasional Berhad (TNB) power utility, has a peak load
approaching 14,000 MW, with approximately 29% of the
generation consisting of CCGT and gas turbine units [4]. The
behaviour of gas turbines and CCGT units in response to the
frequency disturbance contributed significantly to the severity
of the event [4]. As a consequence of the incident, several
modifications to improve the response of both gas turbine
(GT) and CCGT controllers to large frequency excursions
were incorporated.
In large, interconnected electricity systems, frequency
deviations from nominal tend to be small. This is due to the
relatively high inertia of the system, and the fact that any
sudden supply/demand imbalances are generally small in
comparison with the total size of the system. The largest
157

infeed in a small electricity system is likely to be a much
higher proportion of the total generation while the system
inertia is considerably less. Consequently, the effect of an
incident, such as the loss of generation, on the system
frequency is much more notable. An understanding of the
response characteristics of CCGT generators to frequency
events is therefore essential for small systems, as their
response could potentially have a relatively large influence on
the severity of the event. While frequency deviations on larger
systems are generally less sizeable, as the proportion of
CCGTs on such systems increases, their influence will
become more significant.
The Ireland electricity system operates at 50 Hz, with a
peak load of approximately 6065 MW. The system consists of
two AC interconnected power systems, operated by Northern
Ireland Electricity (NIE) and ESB National Grid (ESBNG).
There is a single HVDC interconnection between Ireland and
Scotland, with a capacity of 500 MW. At present, CCGT plant
comprise almost 27% of the total installed generating capacity
on the island, and this figure should continue to rise further,
with over 32% penetration expected by the end of 2005
[5], [6]. At times of low demand, a single CCGT can comprise
more than 15% of the total generation. Therefore, the impact
of the dynamic behaviour of CCGTs in response to frequency
events on the system needs to be assessed carefully.
Furthermore, CCGT ratings tend to be large, and therefore,
with the addition of CCGTs to the system, the size of the
largest infeed, and, as a consequence, the size of the largest
possible contingency has increased. Severe frequency events
are rare on the Ireland system with only one event in the past
ten years resulting in the unanticipated shedding of normal
customers [7]. However, as large CCGTs replace a number of
smaller conventional plant, major frequency events may
become more likely and the impact on the system needs to be
studied. In particular the impact on system integrity and the
possibility of interruption to customers needs to be assessed.
The objective of this paper is to study the impact of the
dynamics of combined cycle gas turbines to frequency
deviations on an island electricity system. The paper also
aims to quantify the effect these units will have on the overall
system short-term response to events on the system.
Preliminary results were reported in [8]. The study focuses on
the response of the system to large single contingencies.
Cascading events, such as those that have occurred elsewhere
[1], [3], [4] are outside the scope of this paper.
A more detailed explanation of the relevant characteristics
of CCGTs is presented in Section III. The details of the CCGT
dynamic model being used in this study are given in Section
IV. Subsequently, in Section V, a model of the Ireland
electricity system is described, followed by the results and
discussion of the findings of this study in Section VI.
III. DYNAMIC RESPONSE OF COMBINED CYCLE GAS TURBINES
Combined cycle gas turbines integrate the technologies of
both the gas turbine and the steam turbine. The exhaust gases
from the gas turbine are fed into the heat recovery steam
generator (HRSG), which produces a supply of steam to drive
the steam turbine. The gas turbine component of the combined
cycle plant is very similar to that of an open cycle gas turbine
(OCGT).
While both OCGT and CCGT technologies have a
maximum allowable temperature imposed by the turbine blade
materials, any variation in the temperature of the exhaust
gases entering the HRSG will affect its efficiency and thus the
efficiency of the steam turbine. Therefore, in order to achieve
optimal efficiency in CCGTs, the exhaust gas temperature
should be maintained at the maximum allowable level.
The exhaust temperature is maintained at this optimal level
by controlling the air and fuel flows. Variable inlet guide
vanes (IGVs), fitted at the entrance to the compressor, control
the incoming airflow. As the gas turbine runs up the IGVs are
positioned to ensure a smooth run up of the air compressor
(avoiding stall zones) until full speed at no load. Thereafter
the IGVs move from their minimum on load position to their
maximum opening in line with the admission of fuel to
maintain the programmed target exhaust gas temperature.
In order to maintain constant outlet temperature it is
necessary to adjust the airflow as the fuel flow changes.
However, as the gas turbine operating point approaches base
load (100% of rated output) the IGVs are fully open, and
airflow cannot be increased further.
Combined cycle gas turbines, in general, are fast
responding units. The speed controller responds within a very
short time to falling system frequency and results in more fuel
being injected into the combustor. However if fuel flow
increases too rapidly, the IGVs may be unable to maintain the
correct air-fuel ratio. The temperature controller will thus
reduce the fuel flow to a level in keeping with the rate at
which the IGVs are opening.
Some CCGTs are equipped with fast acting actuators on the
IGVs, significantly improving the responsiveness to frequency
events. On these units, system frequency is monitored and in
the event of a large deviation, an anticipated increase in outlet
temperature, due to future changes in fuel flow, will be
minimised through feed-forward action suitably increasing
airflow.
In order to better understand the dynamic behaviour of
CCGT units to frequency disturbances, it is important to
recognise the effects that changing system frequency will
introduce. The CCGT will provide an inertial response,
proportional to both the magnitude and rate of the frequency
change. This response is vital in helping to maintain system
security by alleviating the rate at which the frequency is
falling. In addition, as the system frequency falls, the
compressor slows down, since it is synchronised to the
system. A reduction in compressor speed leads to a drop in the
pressure ratio across the compressor, and thus the airflow into
the combustion chamber is reduced. Consequently the
pressure ratio across the gas turbine is reduced, causing a
decline in the power output.
If the unit is operating at partial load, the IGVs are not fully
open, and so it is possible to open these further, offsetting the
reduction in airflow across the compressor. The speed at
which the IGVs can respond determines how quickly this can
take effect. However, when the CCGT is operating at base
(100%) load, the IGVs are now in the fully open position, and
cannot be further adjusted to increase the airflow through the
158

compressor. The reduction in airflow from the compressor
thus causes an increase in the fuel to air ratio, which directly
determines the turbine inlet temperature, and as a result, the
exhaust temperature. Therefore, any rise in the fuel to air ratio
results in an increase in the exhaust temperature. As the
exhaust temperature rises, the temperature control system
quickly overrides other inputs and reduces the fuel flow to
restore the correct fuel to air ratio. As such, the power output
is reduced due to the reduction in fuel flow. These phenomena
can also be significant when a CCGT is operating slightly
below base load.
The power output of the gas turbine also depends on the
ambient atmospheric temperature and pressure. Variations in
the atmospheric temperature from the ambient temperature for
which the gas turbine is rated can result in significant changes
in the maximum power output achievable [9], [10]. As
ambient temperature increases, the temperature differential
across the compressor is reduced, requiring less work to attain
the rated compressor exit temperature. Consequently, the
pressure ratio across the compressor and, therefore, across the
gas turbine is reduced, leading to a reduction in the maximum
power output.
Combined cycle gas turbines are primarily designed to
operate at base load i.e. the condition for maximum efficiency
[11]. As the operating point is reduced from base load, the
efficiency of the gas turbine is similarly reduced, leading to
higher per unit fuel costs. Both the compressor and turbine
components are designed to operate at specific rated
conditions, and any deviation from these optimal conditions
will result in a decrease in their respective efficiencies.
Furthermore, operation of the compressor depends on several
parameters such as ambient conditions, air mass flow and
shaft speed. As the operating point of the compressor moves
away from its rated operating point, and air mass flow is
reduced, the compressor surge margin is also reduced. Since
the speed of rotation of the compressor also affects the surge
margin, if the unit is operating at partial loads and a reduction
in the system frequency occurs, the possibility of encountering
problems such as compressor surge is increasing [12].
Emission levels of both carbon dioxide (CO
2
) and oxides of
nitrogen (NOx) are minimised at base load. However, as the
operating point moves away from rated conditions, the level
of NOx emissions per unit of power produced increases, while
below approximately 60-70% of rated load, these emissions
increase dramatically [12].
When operating in sliding pressure mode, the power output
from the steam turbine depends directly on the exhaust gas
flow from the gas turbine, and as a consequence, the steam
turbine has minimal stored energy available. Unit response
depends entirely on that of the gas turbine, with a typical
delay of at least twenty seconds before changes in the gas
turbine fuel flow result in a notable change in the power
output of the steam turbine. It is therefore reasonable to
neglect steam turbine dynamics when considering the
response of CCGT units for a short time period (20 s).
IV. THE CCGT MODEL
The objective is to develop a CCGT model suitable for
studying the dependency of CCGTs on system frequency,
which can then be tuned to represent individual CCGT units
on a power system, in order to evaluate their dynamic
response to a system frequency event, and examine the
subsequent impact on system frequency control.
A. Previous Work
A great deal of work has been carried out in developing
accurate dynamic models of gas turbines, and it is this work,
which generally forms the foundation for most combined
cycle gas turbine models. The basis of many gas turbine and
combined cycle gas turbine studies is the model of the gas
turbine developed by Rowen [13], [14]. The initial model [13]
was later expanded to include the effects of variable inlet
guide vanes and of ambient temperature [14], along with the
dependency of airflow through the compressor on the system
speed. This model was simpler than some that had previously
been in use, however, the model nevertheless was considered
capable of capturing the dynamic characteristics of the gas
turbine in a reasonably accurate manner.
Bagnasco et al. [15] incorporated the gas turbine model
[13], [14] into a model for a combined cycle power plant,
through the addition of a HRSG and a simplified steam
turbine. Alternative combined cycle plant models have been
developed using detailed physical relationships to represent
each component individually, thus producing a more detailed
model [16].
The dependency of the power output from a combined
cycle gas turbine on system frequency, and also the effects of
ambient conditions are two very important features that
require careful consideration. Kunitomi et al. [9], [17] present
a model that incorporates the frequency dependency of the gas
turbine while also recognising the impact of ambient
conditions on the performance of the unit. This detailed model
is based on [13], [14], but the turbine thermodynamics are
replaced with new equations derived from physical principles,
while frequency dependency is estimated from the ambient
temperature dependency. The model developed is suitable for
use in long-term dynamic simulations, although clearly one
drawback with the more extensive models is the level of detail
required.
A comprehensive study has recently been published by
CIGRE, based on work carried out by a large committee from
various industrial and academic backgrounds [10]. An
extensive examination of combined cycle gas turbines is
presented, with analysis of the various characteristics that set
these units apart from more traditional generating plant. A
dynamic model for a CCGT is developed, based on the broad
experiences of the committee, and previously published work
in this area, for use in power system simulation studies.
B. Assumptions
Assumptions made in this study include: fuel flow is
negligible with respect to airflow, so that the volume of
exhaust gases may be assumed equal to the volume of air
flowing through the compressor, and the pressure ratios are
the same across both the compressor and the gas turbine (i.e.
no air is extracted for cooling). The timescale of interest in
this study is up to 20 s following a system disturbance, so that
the response of the steam turbine can be considered negligible.
159

Acceleration control generally only comes into play during
unit start up and shut down. As steady state conditions are
initially assumed, the acceleration control loop can be
neglected.
C. The model structure
The gas turbine model used in this study has been adapted
from that developed by Rowen [13], [14], by making
reference to more recent CCGT models [10], [15], [16]. An
outline of the structure is shown in Fig. 1, with relevant
equations given in Appendix I.
The required inputs to the model are the steady-state
set-point of the unit, and the ambient temperature and
pressure. The effect of ambient temperature on the rating of
the gas turbine is incorporated as a correction factor,
developed from historical data, which is applied to the unit
set-point input. This differs from previous models, where
ambient temperature was either neglected or incorporated
inside the model structure, for example within the exhaust
temperature calculations [13], [14]. The effect of ambient
pressure on the gas turbine output, which is not addressed in
some models, was also incorporated in a similar way. All
parameters are given in per unit, with the exception of
temperature and pressure, which are measured in C and mbar
respectively.
There are two key control loops on a gas turbine: speed
control and temperature control. Under normal operating
conditions, speed control regulates the fuel supply, and takes
the form of a simple droop governor.
The outputs from both the speed and temperature
controllers are fed into a minimum block, such that the lesser
of the two signals determines the fuel flow into the gas
turbine, within allowable limits. There is a no-load fuel
requirement in a gas turbine, generally about 23% of
maximum, so the controllers regulate the fuel flow between
this minimum point and the maximum. In previous models
[13]-[15], fuel flowing from the fuel pump into the gas turbine
was obtained as a function of the system speed. For this study,
the fuel supply is assumed independent of frequency, and
depends only on the control signal.
In the model illustrated in Fig. 1, the system speed input to
the turbine calculation of exhaust temperature, F1,
incorporates a scaling mechanism to regulate the frequency
sensitivity of the unit. This was developed to take into account
the variations in frequency sensitivity conditional on
technology and age of CCGT units, which were evident on
examination of responses of the different units. The system
speed, N, is multiplied by a scalar, a, and then a constant, b, is
added, such that a+b=1. Since the actual exhaust temperature
depends on the airflow into the combustor, the exhaust
temperature from the turbine calculation, F1, is modulated
using the calculated airflow through the compressor. The
temperature of the exhaust gases is then measured, with the
appropriate time delays incorporated.
The airflow is regulated using an IGV controller. The
calculated exhaust temperature is compared with the rated
exhaust temperature such that if there is a difference, the IGVs
control airflow in order to bring the temperature back to the
rated value. However, as airflow also depends on the speed of
the compressor, the expected airflow, due to IGV position, is
Fig. 1. The gas turbine model
160

modulated using the actual speed, yielding the calculated
flow.
The model is also equipped with an over-firing capability,
which allows the temperature limits to be increased for a short
period of time during a frequency transient.
V. THE SYSTEM MODEL
The electricity system of Ireland, consisting of the NIE and
ESBNG systems, is studied here. The generating capacity of
the current Ireland electricity system consists of a combination
of thermal generators, open cycle gas turbines, hydroelectric
generators, combined cycle gas turbines and a single pumped
storage station. Several ac interconnectors between Northern
Ireland and the Republic of Ireland are in service. The main
interconnection consists of a two 275 kV circuits, each with a
capacity of 600 MW and of length 50 km. There are also two
additional 110 kV lines, with capacity of 120 MW, connecting
the systems at two separate locations along the interface
between the systems [18]. A HVDC interconnector now exists
between Ireland and Scotland, but is not currently configured
to provide reserve in a short time frame.
The electricity system of Ireland is tightly meshed and
electrically short, with the relative impedances between nodes
quite small. Therefore during a major contingency involving
the loss of significant generation the system will remain in
synchronism and the frequency deviation will be very similar
at all points on the system. This is borne out by system
studies carried out by the Transmission System Operators and
by measurements during major events (e.g. sudden generation
deficit of over 450 MW of generation and system frequency
dip to 49.2 Hz). The system is designed and operated so that
in the event of the loss of the largest infeed there will be no
consequential events (i.e. the protection does not trip out any
other devices, for example lines). Historical data shows that
such a cascading event has never occurred. In particular
during a major loss of generation there are noticeable power
swings on the interconnections between the NIE and ESBNG
systems, due to the sharing of reserve [19], [20]. These power
swings have never caused any additional tripping of lines.
Therefore, for frequency control studies, a single busbar
model has traditionally been employed [21]-[23] and is
appropriate for this study i.e. the impact of combined cycle
gas turbines short-term dynamics on frequency control.
The dynamic model of the Ireland system is based on two
previous models, one for the ESBNG system [22] and one for
the NIE system [24], with considerable enhancements
introduced. The two models have been combined and updated
to correspond to a representation of the entire Ireland
electricity system in 2004, with as much detail and accuracy
as is achievable, given available knowledge and data. Since
the development of the original models, new generating plant
has been introduced onto the system and older
decommissioned plant has been removed. Combined cycle gas
turbines were previously not modelled on the system, but are
now fully represented. Another major addition to the model is
the replacement of a simple ramping response with dynamic
models for the individual gas turbines.
When a system event occurs, such as the loss of a large
generator, a model simulates the frequency response for 20 s
subsequent to the disturbance. At present, a frequency event is
defined as a deviation of frequency below 49.7 Hz [25].
Demand customers who can withstand the unplanned and
instantaneous interruption of some or all of their demand
(triggered by an under-frequency relay below 49.3 Hz) supply
interruptible load service. Customers providing this service
should be able to withstand 10 to 20 such interruptions per
annum, with typical interruptions lasting for approximately
five minutes. Historically interruptible load service is called
on infrequently (less than five times per annum) [26]. In
extreme events where the frequency falls below 48.9 Hz load
shedding of customers begins. Fortunately load shedding of
customers is extremely rare with only one such incident in the
past 10 years [7].
Dynamic models have been developed for each generating
unit on the Ireland system in addition to a representative load
model. Prior to a disturbance on the system, power generation
and system demand are assumed to be in equilibrium.
Subsequently, as a result of a power imbalance on the system,
the model simulates the change in power output from each
generating unit, and also any variation in load consumption,
including the effects of tripping interruptible load service,
when available, if a frequency of 49.3 Hz is reached, and load
shedding, if the frequency reaches the appropriate thresholds,
which start at 48.9 Hz. The net power imbalance between
generation and demand drives a feedback loop, where the
system frequency is calculated by integrating the observed
difference, dependent on the combined inertia of the load and
individual generating units. The model was developed and is
implemented using Matlab Simulink.
Both ESBNG and NIE system operators perform
scheduling and dispatch independently, taking into account
the contracted flows on the ac interconnectors. However, the
entire Ireland electricity system shares the provision of
operating reserve. Of particular interest in this study is
primary spinning reserve, i.e. reserve provided in the first 15
seconds following a frequency event. The following
assumptions are made about the system: the primary spinning
reserve requirement on the Ireland system is 75% of the
largest infeed. At present, the largest infeed on the system is
422 MW, and hence the primary spinning reserve requirement
is 317 MW - the DC interconnector has a capacity of
500 MW, but the maximum traded on the link is 400 MW.
This primary spinning reserve requirement, mainly obtained
from generators and interruptible load service, depending on
time of day, is divided between ESBNG and NIE such that the
former provides 67% (211 MW) while the latter provides 33%
(105 MW). Since the model determines the response from
each unit, it is possible to neglect units that provide no
reserve, with the exception of their contribution to the overall
system inertial response. All generating units on the Ireland
system use droop governors, and are expected to operate with
a droop of 4% [25][27].
VI. RESULTS AND DISCUSSION
A. Scope of this study
Three different operating scenarios have been examined in
this study, comprising the maximum and minimum system
161

loads, which respectively occur during winter daytime, and
summer night. A third scenario, reflecting the minimum
daytime load has also been considered. While the first two
scenarios capture the extremes of system operation, it is also
instructive to examine the minimum daytime load.
For each of the three scenarios described above, two cases
are examined. The system is dispatched based on a merit order
to meet demand and reserve requirements, with all dispatched
CCGTs operating at base load. In the first case, the frequency
dependency of all dispatched CCGTs is removed.
Consequently, the power output from the base loaded CCGTs
is constant, with the exception of the inertial response. For the
second case system dispatch is identical, with unchanged
system inertia, the only difference being the frequency
dependency in the CCGT models. In each case, the same
generating units, and interruptible load service if appropriate,
provide the primary reserve requirement.
For each scenario, the most serious single disturbance to the
system is examined, which involves tripping the largest infeed
to the system. During the winter peak, this is likely to be
422 MW, while during the summer night valley and summer
day valley scenarios, this figure is reduced to 400 MW,
accounting for the reduction in power output of this largest
infeed, a CCGT, due to a higher ambient temperature. System
operators on the island of Ireland do not consider multiple
independent events occurring concurrently. Following an
event system operators restore the system back to a secure
operating state as quickly as possible. Therefore N-2
contingencies are not considered here and the tripping of the
largest infeed is considered a likely and severe event that can
occur on the system. These are not the most severe events that
can occur [1] - [4]. The Ireland electricity system has never
experienced the loss of transmission capacity during a low
frequency event, therefore cascading events caused by
tripping of transmission lines during the loss of the largest
infeed are not considered. Consequently this study uses an
optimistic model of the transmission system, which is assumed
to remain intact with no unfortunate or unexpected
transmission system protection consequences considered.
B. Tuning the model and model response
The combined cycle gas turbine model developed was tuned
to represent each individual CCGT unit on the Ireland system,
using available data from a number of system frequency
events (approximately fifty) that have occurred on the system
over the last five years. Post and pre event system frequency
and power output data from units was available, sampled at
1 Hz. Technical data for each individual generating station
was available, and also the ambient conditions at the time of
each event were also obtained.
In order to complete each CCGT model it was necessary to
determine the effect of ambient conditions on unit output. This
relationship was determined using historical hourly data,
comprising ambient conditions, IGV position and generated
power. The sensitivity of a typical gas turbine to ambient
conditions is illustrated in Fig. 2, with output increasing as
ambient pressure rises, but falling as ambient temperature
increases.

Fig. 2. The change in maximum rated power output of a typical gas turbine
for (a) Temperature deviations from rated ambient temperature and (b)
pressure deviations from rated ambient pressure.

Although the relationship between maximum power output
and ambient temperature is generally agreed to be highly
complex and non-linear [10], the temperature-power output
curve obtained in this study was found to be approximately
linear within the temperature range present for the available
data. The pressure-power output relationship was found to be
linear, as expected [10].
Having established the above relationships, CCGT model
validation was performed by driving the model with the
observed frequency trace for a particular event. The ambient
temperature and pressure were also introduced as inputs to the
model. The simulated response, incorporating both the
ambient temperature and pressure corrections, was then
compared to the actual response of the unit. Fig. 3 illustrates
such a comparison between the actual power output of a
CCGT and the corresponding simulation model. It can be
clearly seen that when frequency effects are included in the
CCGT model that a much better representation of the actual
power output is achieved, despite the limited resolution of the
actual data.
The frequency dependency of CCGTs on the system was
found to vary both with age and technology. Improvements in
CCGT technology are on going, and this is evident in the
improved responses of newer units. However, many CCGTs
have already been installed in electricity systems worldwide,
with varying degrees of technology and, unless modified, their
responsiveness will depend on the technology at the time of
installation.
A small number of CCGTs on the Ireland system
incorporate fast acting inlet guide vanes, and these have been
represented in the model by considerably reducing the time
constant on the IGV actuator. This approximation was seen to
capture the characteristics of such fast acting IGVs to a level
of accuracy that the inclusion of an additional control loop to
represent them was not considered necessary.
162

Fig. 3. Change in power output (including the inertial response) of a typical
near base loaded CCGT in response to a frequency event on the system.
(a) System frequency, (b) Response of the actual CCGT unit, (c) Simulated
response of CCGT model, and (d) Simulated response of CCGT model with
frequency dependency of model removed.

The significant role of fast acting IGVs can be better
understood by examining Fig. 4, which considers the response
of a base loaded CCGT to a 0.5 Hz frequency drop compared
to the response of the same CCGT when operating at 95% of
base load.

Fig. 4. Simulated power output of the GT component of a typical CCGT to a
frequency drop of 0.5 Hz (inertial response neglected). (i) Frequency trace
(ii) the CCGT operating at base load and (iii) the CCGT operating at 95%
capacity with (a) conventional IGVs and (b) fast-acting IGVs.

When operating at base load, the power output from the
CCGT declines with the system frequency, and remains below
rated output as long as the frequency is less than nominal.
When operating at partial loads, however, the CCGT response
to changing frequency is rapid, and the influence of fast-acting
IGVs is clearly illustrated. In both cases, the power output
initially increases in response to falling frequency. With
conventional IGVs, however, as the fuel flow increases, the
exhaust temperature will increase causing the temperature
controller to reduce the fuel flow and hence the power output
of the gas turbine. The temperature controller then only allows
the fuel flow to increase as the airflow increases, so the
response of the unit is limited by the operational speed of the
IGVs. With fast-acting IGVs, the airflow increases much more
quickly, and so the response of the CCGT is no longer
constrained in the same way. The maximum response is
instead curtailed by the reduction in airflow due to reduced
compressor speed.
C. Impact of tripping the largest infeed on system frequency
For each of the three considered scenarios, the effect on
system frequency following the loss of the largest generating
infeed to the system was modelled. For the winter peak
scenario, illustrated in Fig. 5, the load is 6065 MW and CCGT
generation comprises 33.5% of the total.

Fig. 5. Winter peak with 422 MW trip (a) base loaded CCGTs comprise
33.5% of generation, and (b) all CCGTs have frequency dependency removed.

During the winter peak, typically between 17.00 and 19.00
hours, there is no interruptible load service on the system, and
the primary reserve requirement must be entirely met by
generator spinning reserves.
It can be seen that the frequency nadir falls a further 15%
when all CCGT generation are operating at base load and the
frequency dependency of their response is modelled,
demonstrating that a small drop in the power output, due to
frequency, of a single CCGT in isolation becomes more
significant as the number of base loaded CCGTs increases.
The load on the system for the summer night valley
scenario is set at 2170 MW, with 49.6% of generation
provided by CCGTs. There are no interruptible load service
on the system and three 73 MW pumped storage units are
operating in pumping mode, thereby providing almost
instantaneous reserve when triggered by falling frequency.
The remaining primary reserve is met by spinning reserves
from online (mainly thermal) generators. The loss of a
400 MW generator during the summer night valley is shown
in Fig. 6.
163

In this case, when system inertia is very low, it can be seen
that the system frequency falls very fast, with an initial rate of
change of frequency of 0.495 Hz/s, and frequency drops
below 49 Hz. The frequency nadir falls by an additional 5.4%
when the CCGTs are base loaded and the frequency
dependency of their response is modelled, if load tripping is
not incorporated. This additional fall in frequency is critical
when the system is already severely stressed. The frequency
falls below the upper threshold for under-frequency load
shedding of 48.9 Hz, leading to 8% of load being shed, as
illustrated in Fig. 6. This summer night valley scenario with
low system load and inertia indicates that with additional
CCGTs on the system events of this nature will be more likely
and the transmission system operators on the island of Ireland
should review their frequency control strategies in the future
so as to avoid the shedding of customers.

Fig. 6. Summer night valley with 400 MW trip (a) base loaded CCGTs
comprise 49.6% of generation, with no tripping of load, (b) base loaded
CCGTs comprise 49.6% of generation, with load tripping of 8% of system
demand at 48.9 Hz and (c) all CCGTs have frequency dependency removed.

However, the difference between the two frequency nadirs
is not as large as for the winter peak load scenario. This
illustrates that although the percentage generation provided by
base loaded CCGTs has grown from 33.5% to 49.6%
penetration, the magnitude of generation provided by these
units is less, and therefore the effect on system frequency is
smaller.
Finally, of the 3150 MW of generation on the Ireland
system for the summer day valley scenario, 34.2% consists of
CCGT generation. During this scenario, 60 MW of the
primary reserve on the system is in the form of interruptible
load service, which is activated if the frequency falls to
49.3 Hz. The results of the summer day valley scenario
simulation are presented in Fig. 7. The frequency nadir falls
6.3 % further when the CCGT generating units are base
loaded and the frequency dependency of their response is
modelled.
In Fig. 8, the effect on the frequency nadir of increasing
proportions of base loaded CCGT generating units, with the
frequency dependency of their response modelled, is
illustrated for a typical winter day load, with 60MW of
interruptible load service and the remaining 256 MW of
spinning reserve requirement met by thermal generators. It is
shown to be non-linear, with the effect becoming more
pronounced as the percentage of base-loaded CCGTs
increases. This exponential fall in frequency as more CCGTs
are added is due to frequency falling further, which causes
unit outputs to fall to a greater extent, in turn causing the
frequency to fall further.

Fig. 7. Summer day valley with 400 MW trip (a) base loaded CCGTs
comprise 34.2 % of generation and (b) all CCGTs have frequency dependency
removed.

The technical problems of operating a CCGT at maximum
output may be overcome by a number of methods. Over-firing
is one option, but is currently not allowed for units on the
Ireland system. De-rating the CCGT is also possible, but the
overall efficiency of the unit will be reduced. Other options
include increasing the primary operating reserve requirement,
to offset the negative power response of base loaded CCGTs,
or reducing the size of the largest infeed to the system, thus
limiting the largest possible single contingency. However, a
significant cost can be associated with each solution, and
detailed cost/benefit analysis will be required to quantify and
compare these costs, which is not undertaken here.

Fig. 8. Sensitivity of system frequency nadir to increasing proportions of
CCGTs (a) The percentage increase in the frequency nadir as the proportion of
base loaded generation increases and (b) the percentage increase if the
increase was linear.
164

VII. CONCLUSIONS
A validated model of a CCGT is developed which
incorporates the effects of fast acting inlet guide vanes,
ambient temperature and pressure and a mechanism for
varying the frequency dependency of the gas turbine. This
CCGT model is combined with a validated system model to
study the impact of CCGTs on frequency control following
the loss of generation. As large base loaded CCGTs replace
existing conventional generation in the future, frequency
control may become more challenging. As the events studied
here are likely, are not the most severe and are based on an
optimistic model of the transmission system, the results
indicate that with additional CCGTs on the system, large
frequency excursions will be more likely and the transmission
system operators on the island of Ireland should review their
frequency control strategies in the future so as to avoid the
shedding of customers.
VIII. APPENDIX
Model equations:

F1: T
x
= {[T
r
-453*((a*N+b)
2
+4.21*((a*N+b)+4.42)
*0.82*(1-W
f
)]+722*(1-(a*N+b))+1.94*(Maxigv-igv)}

F2: Torque = 1.3*(W
f
-0.23)+0.5*(1-N)

F3: W
x
= N*(L
igv
)
0.257

IX. ACKNOWLEDGMENTS
The authors gratefully acknowledge the useful discussions
and interactions with colleagues in ESB National Grid, in
particular J. OSullivan, M. Power, D. Barry and K.
OConnor, colleagues in ESB Power Generation, in particular
A. Egan and N. Tarrant, T. Wilson of Viridian, colleagues in
NIE, in particular M. Preston, colleagues in Synergen, and
also colleagues in the ERC, especially A. Mullane, S. Rourke,
R. Doherty, E. Denny and A. Keane.
X. REFERENCES
[1] U.S. Canada Power System Outage Task Force, Interim Report:
Causes of the August 14
th
Blackout in the United States and Canada,
November 2003, https://reports.energy.gov/.
[2] National Grid Transco, Investigation into the Loss of Supply Incident
affecting parts of South London at 18.20 on Thursday, 28 August
2003, September 2003. www.nationalgrid.com/uk/.
[3] Union for the Co-ordination of Transmission of Electricity (UCTE),
Interim report of the Investigation Committee on the 28 September
2003 blackout in Italy, October 2003, www.ucte.org (Publications)
[4] Asean Energy website, http://www.aseanenergy.org/.
[5] ESB National Grid, Generation adequacy report 2004-2010,
http://www.eirgrid.com/.
[6] Commission for Energy Regulation (CER), List of licensees
http://www.cer.ie/list_of_licensees_details.asp?LicenceType=1
[7] Power, M., Personal communication, Manager Power System Control,
ESB National Grid, December, 2004.
[8] Lalor, G. and OMalley, M., Frequency Control on an Island Power
System with Increasing Proportions of Combined Cycle Gas Turbines,
IEEE Powertech Conference, Bologna, June 2003.
[9] Kunitomi, K., Kurita, A., Okamoto, H., Tada, Y., Ihara, S., Pourbeik, P.,
Price, W.W., Leirbukt, A.B. and Sanchez-Gasca, J.J., Modeling
frequency dependency of gas turbine output, IEEE Power Engineering
Society Winter Meeting, 2001, Vol. 2, pp. 678-683.
[10] CIGRE Taskforce 38.02.25, "Modeling of gas turbines and steam
turbines in combined-cycle power plants," CIGRE, Technical Report,
April 2003.
[11] Hung, W., Frequency Control Issues in the England and Wales
Electricity Grid, GMA Conference on Development and Interaction
of Generators and Power Networks in an Energy De-regulating
Environment, Munich, Feb. 2001.
[12] Transpower, Frequency Standards Working Group Preliminary Report
to the Grid Security Committee, Performance of Thermal Plant at Low
Frequencies, Appendix 6, www.gsp.co.nz/.
[13] Rowen, W.I., Simplified mathematical representations of heavy-duty
gas turbines, ASME, Vol. 105(1), Journal of Engineering for Power,
Series A, October 1983, pp. 865-869.
[14] Rowen, W.I., Simplified mathematical representations of single shaft
gas turbines in mechanical drive service, Turbomachinery
International, July/August 1992, pp. 26-32.
[15] Bagnasco, A., Delfino, B., Denegri, G.B. and Massucco, S.,
Management and dynamic performances of combined cycle power
plants during parallel and islanding operation, IEEE Transactions on
Energy Conversion, Vol. 13, No. 2, June 1998, pp. 194-201.
[16] Working Group on prime mover and energy supply models for system
dynamic performance studies, Dynamic models for combined cycle
plants in power system studies, IEEE Transactions on Power Systems,
Vol. 9, No. 3, August 1994, pp. 1698-1708.
[17] Kunitomi, K., Kurita, A., Tada, Y., Ihara, S., Price, W.W., Richardson,
L.M. and Smith, G., Modeling combined-cycle power plant for
simulation of frequency excursions, IEEE Transactions on Power
Systems, Vol. 18, No. 2, May 2003, pp. 724- 729.
[18] ESB National Grid (TSO Ireland), Forecast Statement 2004 - 2010,
http://www.eirgrid.ie/EirGridPortal/uploads/Publications/FS0410.pdf.
[19] Department of Enterprise, Trade and Investment Northern Ireland
(DETINI), Report on the NIE response to the loss of two generating
units (520MW) at Kilroot Power Station on Thursday 4 December
2003, http:// www.detini.gov.uk/cgi-bin/downutildoc?id=399.
[20] ESB National Grid, (TSO Ireland), Interconnector Status,
http://www.eirgrid.ie/EirGridPortal/DesktopDefault.aspx?tabid=Interco
nnector%20Status&TreeLinkModID=1451&TreeLinkItemID=16
[21] OSullivan, J., "Modelling and identification of emergency reserve with
applications to isolated power systems" Ph.D. dissertation, Dept.
Electronic and Electrical. Eng., University College Dublin, Dublin,
1996.
[22] O'Sullivan, J. and O'Malley, M. J., Identification and validation of
dynamic global load model parameters for use in power system
simulation, IEEE Transactions on Power Systems, Vol. 11, 1996, pp.
851-857.
[23] O'Sullivan, J. and O'Malley, M. J., Economic dispatch of a small utility
with a frequency based reserve policy, IEEE Transactions on Power
Systems, Vol. 11, 1996, pp. 1648-1653.
[24] Thompson, J. G. and Fox, B., Adaptive load shedding for isolated
power systems, IEE Proceedings on Generation, Transmission and
Distribution, 1994, Vol. 141, No. 5, pp. 491-496.
[25] ESB National Grid, Grid Code, http://www.eirgrid.com/.
[26] ESB National Grid, System Operations: Ancillary Services,
http://www.eirgrid.ie/EirGridPortal/DesktopDefault.aspx?tabid=Ancilla
ry%20Services.
[27] System Operator for Northern Ireland, Grid Code,
http://www.soni.ltd.uk/gridcode.asp

XI. BIOGRAPHIES
Gillian Lalor received a B.E. degree in Mechanical
Engineering from University College Dublin in 2001.
She is currently conducting research for a Ph.D. at
University College Dublin, with interests in power
system modelling and control.

165

Julia Ritchie received a B.Eng. degree in Electrical
and Electronic Engineering from The Queens
University of Belfast in 2000. She is currently
studying for a Ph.D. in the area of intelligent control
and modelling of power plants and power systems at
The Queens University of Belfast.

Damian Flynn is a lecturer in Power Engineering at
The Queen's University of Belfast. His research
interests involve an investigation of the effects of
embedded generation sources, especially renewables,
on the operation of power systems. He is also
interested in advanced modelling and control
techniques applied to power plant. He is a member of
the IEEE.

Mark OMalley received B.E. and Ph.D. degrees
from University College Dublin in 1983 and 1987,
respectively. He is currently a Professor in University
College Dublin with research interests in power
systems, control theory and biomedical engineering.
He is a senior member of the IEEE.
Appendix C
Frequency Control in Competitive
Electricity Market Dispatch
167
Abstract-- The dynamic and inertial characteristics of
electricity systems are evolving with consequences for system
frequency control. This issue is particularly important for smaller
more isolated electricity systems where simple heuristic reserve
targets may no longer be sufficient to ensure the security of the
system. In a market environment this issue needs to be dealt with
in a systematic way compatible with the method of market
dispatch to ensure least cost operation of the system. This work
describes how frequency control constraints can be derived and
included into a market dispatch algorithm. Comparisons are
made with a conventional reserve target dispatch approach for a
real isolated power system. The impact on market prices, cost and
the number and size of problem contingencies is assessed. The
effect of increasing amounts of wind capacity is also examined.
Index Terms-- Frequency control, linear programming, power
generation dispatch, power system security.
I. INTRODUCTION
ispatching an electricity system in a way that ensures the
security of the system has always been a priority of
system operators. This priority has not changed during the
transition from vertically integrated utilities to competitive
market based operation. Controlling the system frequency at
all times to avoid load shedding is a critical aspect of system
security [1], and while this is easily achieved in large heavily
interconnected systems, it is much more problematic in smaller
isolated systems. Electricity systems are evolving rapidly and
the inertial and dynamic characteristics of many new sources
of generation differ from that of conventional plant in the past
[2]. The increasing penetration of wind generation in systems
has caused concern about the availability of the turbines
stored kinetic energy to the system [3]. In isolated systems,
which already have a relatively small inertial base, these
changes may cause significant problems to operators trying to
ensure system security. Typically, many power systems have
tried to ensure frequency control by the use of a simple reserve
constraint in an economic dispatch algorithm [4],[5]. This
constraint generally stated that the reserve on the system
University College Dublin which is supported by Electricity Supply Board
(ESB), ESB National Grid, Commission for Energy Regulation, Cylon and
Enterprise Ireland.
R. Doherty, G. Lalor and M. OMalley are with the Electricity Research
Centre, Department of Electronic and Electrical Engineering, University
College Dublin, Dublin 4, Ireland (Ph: +353 (0)1 7161857; e-mail:
ronan@ee.ucd.ie; gill@ee.ucd.ie; mark.omalley@ucd.ie ).
should equal a certain percentage of the largest infeed or load.
This percentage was determined heuristically in order to be
sufficient to avoid load shedding but does not account
specifically for the dynamic characteristics of the power
system. Given the evolving characteristics of systems it may be
the case that for isolated systems the use of a single heuristic
reserve target (where the sum of the reserves on the system
equals a predetermined target value) is not sufficient to ensure
system security and consideration should also be given to other
factors such as the alteration of the size of the largest
contingency and the stored kinetic energy on the system.
Economic factors are the main drivers in modern electricity
systems and operators no longer enjoy the same redundancy in
systems as they had in the past. Cost reduction can be achieved
if ancillary services are simultaneously considered when
dispatching the system [6] and many systems now handle
frequency control issues inherently within their markets to try
to reduce the costs of system security [7]. Some developed
electricity markets use linear programming (LP) dispatch
algorithms to co-optimise energy and reserves as a means to
ensure system reliability in a least cost manner [8],[9],[10].
Work based on the Taiwan power system [11] and work based
on the Republic of Ireland power system [12] use system
frequency models to examine the behaviour of the systems
after the loss of large units. The amount of reserve carried by
the systems is assessed to see if it is sufficient to meet the
systems frequency criteria. These methodologies can help to
ensure system security but are generally not compatible with
modern market dispatch algorithms. Crucially they also lack
the facility to alter other important factors such as the size of
problematic contingencies and the amount of kinetic energy
connected to the system as part of a least cost solution.
The evolving dynamic characteristics of modern electricity
systems calls for a new approach to ensuring frequency control
that is compatible with existing market clearing engines. In this
paper a co-optimised market clearing algorithm is presented
which incorporates two new frequency-based security
constraints, a rate of change of frequency (Rocof) constraint
and a minimum frequency constraint. These constraints are
derived in terms of the variables that have a direct impact on
the frequency. These variables are also control variables in the
dispatch. In this way the system can be dispatched in a least-
cost manner and in accordance with frequency control criteria.
In section II the basis of a LP market dispatch is described
into which will be incorporated the frequency-based
constraints. Section III shows how the constraints are derived,
Frequency Control in Competitive Electricity
Market Dispatch
Ronan Doherty, Member, IEEE, Gillian Lalor, Member, IEEE, and Mark OMalley, Senior Member IEEE
D
168
and includes details of the dynamic modeling and linearisation
stages associated with the minimum frequency constraint.
Details of the system scenarios and simulations are given in
section IV. Results shown in section V include illustrations of
the performance of the frequency-based constraints and
comparisons are made with the performance of a conventional
dispatch method with a simple reserve target. Market prices,
cost and the impact of increasing amounts of wind generation
are also examined in section V.
II. MARKET CLEARING FORMULATION
In any economic dispatch the aim is to dispatch the system
in a least-cost manner subject to constraints. Here a LP market
clearing formulation is used to co-optimise energy, reserve and
kinetic energy. The control variables in the dispatch
formulation are:
P
i
power from each unit, and thus the size of the
contingency resulting from the loss of the unit.
R
i
primary reserve from each unit.
KE
i
stored kinetic energy provided by each unit
L the system load
The two frequency based constraints will be derived in terms
of these control variables, thus allowing the LP algorithm to
meet the constraints in a least-cost manner. Unit operating
points, reserve levels and kinetic energy are co-optimised to
find a least cost solution on an hourly basis. Units are assumed
to have linear bids for energy and reserve. Units are assumed
to have a zero bid for kinetic energy, as there is little or no
incremental cost associated with its provision. Price responsive
load characteristics can easily be included in the formulation,
but for simplicity have been excluded here.
A. Formulation
The aim is to minimise the objective function:
1 1
N N
i i i i
i i
min bp P br R
= =
+
| |
|
\ .

(1)
where bp
i
and br
i
are the energy and reserve bids of generator i
and N is the number of generators. Neglecting losses, this is
subject to the load balancing constraint.
1
N
i
i
P L
=
=
(2)
The unit characteristics are included with equations (3)-(8).
0
i i
P Pmax s s (3)
0
i i
R Rmax s s (4)
0
i i
KE KEmax s s (5)
1
i i i
i
P R Pmax
Rslope
s (6)
0
i
i i
i
Rmax
P R
Pmin
+ s (7)
0
i
i i
i
KEmax
P KE
Pmin
+ s (8)
The nature of these constraints are illustrated in Fig. 1.
Fig. 1. Generator reserve and kinetic energy characteristics.
The issue of ensuring that the units are not dispatched in the
infeasible region 0 < P
i
< Pmin
i
is dealt with in section IV. To
complete the market clearing formulation two frequency-based
security constraints are derived and included. These
constraints must be linear in nature and be expressed in terms
of the control variables. Details of how these constraints are
derived for a specific isolated system are shown in the next
section.
III. FREQUENCY CONTROL CONSTRAINTS
The isolated system used as the subject of this work is the
all Ireland electricity system. It consists of both the Northern
Ireland and Republic of Ireland systems. After consultation
with the Republic of Ireland system operator it was assumed
that the system could be defined as secure if the loss of a
single infeed did not cause a Rocof greater than 0.25 Hz/s or a
minimum frequency less than 49.3 Hz.
A. Rocof Constraint
The rate of change of system frequency is dependent on the
size of the unit lost and the amount of available stored kinetic
energy on the system [13]. The rate of change of system
frequency after the loss of a unit can be expressed as (9).
0
2
system b
f P df
dt H S
= (9)
where P is the size of the infeed lost, H
system
is the system
inertial constant, f
0
is the nominal system frequency and S
b
is
the MVA rating of the system. Therefore, the rate of change of
frequency on a 50 Hz system can be expressed in terms of the
kinetic energy as follows:
25
system
df P
dt KE
= (10)
where KE
system
is the total stored kinetic energy on the system.
In the formulation only the kinetic energy supplied from the
units is deemed dispatchable and therefore it must be separated
from KE
system
in the constraint. For a maximum allowable
Rocof of 0.25 Hz/s the constraint is formulated as follows.
1
100
N
i k L
i
i k
KE P KE k G
=
=
> e
(11)
Power
Pmin
Pmax
KEmax
Kinetic
Energy
Pmin
Pmax
Rmax
Power
Reserve
Rslope > -1
169
0
50
100
150
200
250
300
350
400
450
500
0 5 10 15 20
Time (s)
R
e
s
p
o
n
s
e

(
M
W
)
49.2
49.3
49.4
49.5
49.6
49.7
49.8
49.9
50
50.1
F
r
e
q
u
e
n
c
y

(
H
z
)
Black Box Model Validated Model Frequency
where G is the set of generators and KE
L
is the kinetic energy
supplied from the load. It can be seen that the Rocof constraint
is inherently linear and can be expressed in terms of the
control variables. It can be seen that a reserve target cannot
replicate the impact of the Rocof constraint on system security,
as it does not account for the kinetic energy of the system.
B. Minimum Frequency Constraint
The minimum frequency constraint is also a function of the
size of a single possible contingency and the kinetic energy on
the system but, unlike the Rocof constraint, the minimum
frequency constraint is also a function of the response of the
generators providing reserve and the dynamic response of the
load. In order to examine such a system a dynamic frequency
model is required. A fully validated dynamic frequency model
of the all-Ireland system [14],[15],[16] developed over many
years is used here to derive the minimum frequency constraint.
To be meaningfully incorporated into economic dispatch, the
MW amount of reserve that a unit is dispatched for must be
related to the response of the unit to a frequency event. Here
the amount of reserve that a unit is dispatched for corresponds
to the maximum amount that the unit is able to provide 5
seconds after a frequency event.
In order to accurately derive the minimum frequency
constraint a frequency model of the system needs to be run
many times to create a very large database of events that have
either a Rocof of 0.25 Hz/s or frequency fall to 49.3 Hz. The
fully validated system frequency model is a large and complex
model, which is valid for a large range of frequency events.
However, the full system frequency model is computationally
intensive and would require a very long time to create a large
database of events. It is for this reason that a simplified model
of the system was used. This is shown in Fig 2. The dynamic
load response in the simplified system model is assumed to be
2.5% for a one hertz drop in frequency [17], and the kinetic
energy of the load, KE
L
, is assumed to be 2.5L in MWs. All
kinetic energy connected to the system is lumped into the
model of the connecting system.
Fig. 2. Outline of simplified system frequency model.
The simplified system model incorporates a black box model
to simulate the generator reserve responses. The black box
model was tuned using the fully validated frequency model
and is only valid for frequency events that result in a minimum
frequency of 49.3 Hz. Fig. 3 shows the black box model where
R denotes the total reserve dispatched on the system.
Fig. 3. Black box model of generator reserve responses.
Fig. 4 shows the response of 252 MW of reserve to a relatively
slow frequency deviation for both the black box model and the
full system model and Fig. 5 below compares black box model
with the full system model for 403 MW of reserve during a
faster frequency deviation. It was found that the black box
model provides a good approximation to the fully validated
system models generator responses for frequency deviations
to 49.3 Hz and over the relevant range of system reserve.
Fig. 4. Response of models to slow frequency drop with 252 MW of reserve.
Fig. 5. Response of models to fast frequency drop with 403 MW of reserve.
The simplified system model is run many times for different
sizes of contingencies, reserve levels and load levels. For each
combination of inputs the kinetic energy on the system is
altered through an iterative process until the resulting
frequency output either has a rate of change of frequency
0.25 Hz/s or a minimum frequency of 49.3 Hz. A large
database of over 20,00 event scenarios was produced this way.
0
100
200
300
400
500
600
0 5 10 15 20
Time (s)
R
e
s
p
o
n
s
e

(
M
W
)
49.2
49.3
49.4
49.5
49.6
49.7
49.8
49.9
50
50.1
F
r
e
q
u
e
n
c
y

(
H
z
)
Black Box Model Validated Model Frequency
-0.000864R
2
+ 2.869248R
2s + 1
Rate Limiter
Slew Rate = 0.12
Af AP
Loss of Unit
Reserve
Response
Load
Response
Connecting System
(Kinetic Energy)
Af
P
k
AP
Gen
AP
Load
_
+
+
170
100
200
300
400
500
150
200
250
300
350
400
450
500
0
1
2
3
4
x 10
4
Size of Contingency
(MW)
Primary Reserve
(MW)
Minimum
Kinetic Energy
needed from
Generation
(MWs)
Min Freq
Constrain
Infeasible
Region
Rocof Constraint
Min Freq
Constraint
Reserve
(MW)
While creating the database the load values, size of
contingency values and system reserve values were each varied
over a wide range so that the database encapsulates all likely
single event scenarios.
Fig. 6 below shows the minimum amount of kinetic energy
needed from the generation to meet the two constraints as a
function of the size of the infeed lost and the reserve on the
system. In Fig. 6 it should be noted that the load dimension has
been excluded and that the constraints are shown for a load
level of 5000 MW. Event scenarios that require unfeasible
amounts of kinetic energy from the generation, i.e. more than
40,000 MWs, are not considered and are shown as zero and
labelled as the infeasible region.
Fig. 6. Illustration of the frequency-based constraints.
It can be see that the minimum frequency constraint is not
inherently linear in nature. However, it was found that the
minimum frequency constraint remained convex over the full
range of values examined and therefore can be approximated
for the LP dispatch with a number of linear functions. The set
of scenarios that were identified as having binding minimum
frequency constraint are approximated for the relevant region
with 5 separate, 4-dimensional linear functions. This was done
using functions in the MATLAB optimisation toolbox to
minimise the error between the linear functions and the actual
data form the database of event scenarios. The linear
approximations were based on dividing the size of contingency
and reserve plane into 5 pieces. It was found that the linear
functions gave a very good approximation to the constraint
with only very small errors. The minimum frequency
constraint as approximated in terms of the control variables by
the linear functions is given in (12) and (13), where j ranges
between 1 and 5.
It can be seen that the minimum frequency constraint and
Rocof constraint are better suited to ensuring system security
than a conventional reserve target. They consider other
variables (such as the kinetic energy and the size of
contingency) that affect the system frequency and allow them
to be controlled in the dispatch as part of a least-cost solution.
Subsequently, the frequency-based constraints are well suited
to systems with changing dynamic characteristics.
,1 , 2 ,3 , 4
1 1
,
N N
i j j k j i j
i i
i k i k
KE C L C P C R C k G j
= =
= =
> + + + e

(12)
-5.33 230.93 -124.89 -834.79
-5.56 248.35 -150.80 -580.40
-4.69 182.40 -67.61 -832.85
-4.84 198.17 -76.98 -2887.29
-4.48 175.78 -52.08 -4143.23
C =
(
(
(
(
(

(13)
IV. SYSTEM AND SIMULATIONS
The all-Ireland electricity system has, at present, an
installed capacity of approximately 7500 MW and just one
HVDC interconnector to Scotland. The all-Ireland system is
currently undergoing significant changes with increased
amounts of HVDC interconnection to Wales planned for the
near future along with an increasing amount of new combined
cycle gas turbine (CCGT) plant. There is also expected to be a
rapid increase in the amount of wind capacity being installed
on the island [18]. Consideration has also been given to the
implementation of a new market structure [19]. The system has
four 73 MW pumped storage units at Turlough Hill which
plays an important role in the dynamic security of the system.
Frequency control has always been an important issue on
the all-Ireland system, as it is an isolated system where a single
unit can be providing up to 10%-15% of the generation at
certain times. The prospect of increasing amounts of HVDC
interconnection and wind power capacity has given rise to
further concern about frequency control as the possible
provision of reserve and stored kinetic energy from these
technologies is still unclear.
A. System and Scenarios
Market simulations were carried out using MATLAB.
Three test days in 2004 and 2010 on the all-Ireland system
were considered. The 2010 system assumes new generation
and HVDC interconnection. The installed wind capacity for
2004 and 2010 was assumed to be 270 MW and 2000 MW
respectively. A wind production profile was applied and taken
to be the same for each of the test days. It is assumed that the
HVDC interconnection and wind capacity do not provide any
reserve or kinetic energy to the system. It is assumed that the
wind capacity has the ability to fault ride-through. The load
levels for the 2004 scenario were based on historical data [20]
and load levels for the 2010 scenario are based on a load
growth of 4% per year [21]. The three test days are:
1. July business day, representing a low load case.
2. May business day, representing an average load case.
3. January business day, representing a high load case.
Details of the interconnection and unit characteristics were
supplied by the system operator in the Republic of Ireland.
Units bids are assumed to be the short run cost of the
provision of energy and reserve. The availability and operation
of the energy constrained hydro generators and pumped
171
storage unit was based on the historic operation on the test
days in 2003 [20]. The contribution of the pumped storage
station to the frequency control of the system was fully
modelled and included in the simulation.
B. Performance Versus Conventional Method
The performance of the dispatch model with the
frequency-based constraints as described by (1)-(8) and (11)-
(13) is compared to a conventional dispatch model with just a
simple reserve target R
T
. The conventional reserve target
dispatch has no frequency-based constraints and does not
consider kinetic energy. Instead there is a reserve target of 316
MW, this is similar to the amount of primary reserve held on
the all-Ireland system present. This is a heuristic target based
on an assumed largest infeed of 400 MW in both 2004 and
2010 and an assumed load response of 2.5% per hertz. The
conventional reserve target method is fully described by (1)-
(4),(6),(7) and (14)
1
316
N
T i
i
R R
=
= =
(14)
C. Ensuring Feasibility
The problem of ensuring feasibility in co-optimised LP
market dispatch is well documented [22],[23]. The minimum
operating point of generators requires that discrete decisions
be made about the on or off status of the units. In the past this
problem was often dealt with by unit commitment techniques.
However, in an increasing amount of deregulated electricity
markets, central unit commitment algorithms are not being
used and market participants are expected to self-commit. As
this issue is not the main focus of study in this paper a simple
approach is adopted. The dispatch algorithms are run firstly
with the set of P
i
allowed to vary between 0 and Pmax
i
in (3).
This ensures optimality but does not ensure feasibility as some
of the units may be dispatched below their minimum operating
points. With linear bids for energy and reserve, and the nature
of constraints (7) and (8), it is assumed that units that the
algorithms have dispatched on in the infeasible region are
necessary for the dispatch. The dispatch algorithms are run a
second time with all the units that were deemed necessary for
the dispatch turned on by allowing their set of P
i
vary between
Pmin
i
, the actual minimum operating points and Pmax
i
. The
remainder of the units are made unavailable. The dispatch
algorithms now return a feasible dispatch. Although this
approach has the potential to over commit generation it was
found to give good results for almost all the scenarios tested
with only a very slight increase in the value of the objective
function between the first and second run of the algorithm.
D. The Lagrangian and Marginal Prices
The Lagrange multipliers can be used to provide a measure
of the sensitivity of an objective function to changes in a
constraint [24]. Here they are used to derive the marginal price
for energy, reserve and kinetic energy. The full Lagrangian
functions and the market prices for both dispatch methods are
shown in the appendix.
V. RESULTS AND DISCUSSION
A. Performance of Conventional Reserve Target Dispatch
The conventional reserve target approach as described by
(1)-(4),(6),(7) and (14) was applied to all test days outlined in
section IV. Fig. 7 and Fig. 8 show the number of problem
contingencies that would occur during the May 2004 and May
2010 test days using conventional reserve target dispatch
method
1
. Shown is the number of units, which, if tripped
during the hour in question would cause a Rocof greater than
0.25 Hz/s or cause a minimum frequency less than 49.3 Hz. It
can be seen that for the May 2004 test day there are up to 6
different units that cause the Rocof and the minimum
frequency security criteria to be broken. The lower level of
response and kinetic energy provided by the load during the
night hours coupled with the fewer units and consequently less
kinetic energy supplied from the units causes the number of
problematic contingencies to increase during the night hours.
0
1
2
3
4
5
6
7
1 3 5 7 9 11 13 15 17 19 21 23
Hour
N
u
m
b
e
r

o
f

P
r
o
b
l
e
m

C
o
n
t
i
n
g
e
n
c
i
e
s
Rocof Constraint
Minimum Frequency Constraint
Fig. 7. Number of problem contingencies during May 2004 test day.
0
2
4
6
8
10
1 3 5 7 9 11 13 15 17 19 21 23
Hour
N
u
m
b
e
r

o
f

P
r
o
b
l
e
m

C
o
n
t
i
n
g
e
n
c
i
e
s

Fig. 8. Number of problem contingencies during May 2010 test day.
For the May 2010 test day it can be seen that the number of
problematic contingencies has in general reduced due to the
increase load and subsequent increase in the number of units
on-line. However, there are up to 9 separate units that can
cause the security criteria to be broken. This is due to the
projections that several new large CCGT plant will be built in
the coming years. The July test days were shown to be the days
with the most problem contingencies due to its low load level
while the January test days had the fewest problem
contingencies. However, even the January test days showed
hours where 6 separate contingencies that could break the
frequency control criteria.
1
It should be noted that the dispatch methods presented here differ from
those used by the system operators on the all-Ireland system at present.
172
B. Performance of Dispatch with Frequency Based
Constraints
The dispatch model with the frequency based constraints as
described by (1)-(8) and (11)-(13) was applied to the scenarios
described in section IV. Individual hours were run and
dispatches obtained. Due to the constraints (11)-(13), all
dispatches met the frequency control criteria. The dispatched
operating points of the units, the reserve levels and the kinetic
energy levels were then fed back into the simplified system
frequency model to check dynamically the performance of the
dispatch. It was found that the dispatch model produced
accurate dispatches in terms of meeting the frequency-based
constraints in a least cost fashion for all the scenarios tested.
This shows that the linearisation of the minimum frequency
constraint, (12), (13) has not compromised the accuracy of the
constraint. This also leads to the conclusion that specific
frequency control criteria can be met in a least-cost fashion
with a LP dispatch algorithm.
C. Cost
Analysis was carried out on the cost of dispatching the
system using the dispatch with frequency-based constraints
and comparisons were made with the cost of dispatching the
system with the conventional reserve target method. It was
found that over all the test day scenarios the dispatch with
frequency-based constraints only caused an average increase
of 0.3 % to the cost of dispatching the system. The January test
days even showed a decrease in the cost of dispatching the
system using this approach.
D. Meeting the Security Constraints with the Conventional
Reserve Target Dispatch
Analysis was carried out to determine the ability of the
conventional reserve target method to satisfy the security
criteria by increasing the heuristic reserve target for the test
day above 316 MW. For the January 2004 test day it was
found that the system needed to carry 573 MW of reserve in
order to just satisfy the minimum frequency constraint in every
hour. This causes a 5.5% increase in the cost of dispatching
the system for the day. The January 2010 test day required
438 MW of reserve to be carried to satisfy the minimum
frequency constraint. This resulted in a 1.1% increase in the
cost of dispatching the system. For the May and July test days
the dispatch algorithm experienced significant difficulty in
solving due to the large amount of reserve needed to satisfy the
minimum frequency constraint. However, it is expected that
these test days would experience a greater increase in cost than
the January test day in the same year. It was found that the
conventional reserve target method could not satisfy the Rocof
constraint for any of the test days.
E. The Effect of the Frequency Based Constraints on the
Largest Infeeds
One of the main differences between the conventional
reserve target method and the dispatch with frequency based
constraints is the ability of the frequency based constraints to
dispatch large problematic units at lower levels to help meet
the constraints as part of a least cost solution. Fig. 9 shows the
operating point of the largest units on the system, a 400MW
CCGT, for the January 2004 test day. It can be seen that the
conventional dispatch method dispatched the system at its
maximum during all hours of the day.
200
250
300
350
400
450
1 3 5 7 9 11 13 15 17 19 21 23
Hour
C
C
G
T

O
p
e
r
a
t
i
n
g

P
o
i
n
t

(
M
W
)
With Frequency Based Constraints
Conventional Reserve Target
Fig. 9. Operation of the largest unit during the January 2004 test day.
The dispatch with the frequency-based constraints
determines that the least cost operation of the system given the
constraints results in the units being dispatched at lower
operating points. Fig. 10 shows the operating point of the
CCGT using both dispatch methods for the January 2010 test
day. It can be seen that the conventional dispatch method
dispatched the system at its maximum all day except for two
hours where it is dispatched at a lower point in order to supply
reserve.
200
250
300
350
400
450
1 3 5 7 9 11 13 15 17 19 21 23
Hour
C
C
G
T

O
p
e
r
a
t
i
n
g

P
o
i
n
t

(
M
W
)
Fig. 10. Operation of the largest unit during the January 2010 test day.
It can be seen that the operation of the CCGT using the
dispatch with frequency-based constraints results in higher
operating points for the 2010 test day than for the 2004 test
day. For the 2004 there are only two infeeds of 400 MW and
the frequency-based constraints determine that the least cost
operation of the system results in the curtailment of these units
rather than the provision of increased amounts of reserve and
kinetic energy to facilitate the loss of the units at higher
operating points. For the 2010 test day there are assumed to be
five infeeds of a similar size as the CCGT. For this case it can
be seen from Fig. 10 that the frequency-based constraints
determines that the least cost operation of the system results in
the system carrying increased amounts of reserve and kinetic
173
energy to allow all five units to be dispatched at higher
operating points.
F. Market Prices
Fig. 11 shows the market prices for both methods of
dispatch for the July test day in 2004. It should be noted that
the conventional reserve target dispatch does not produce a
marginal price for kinetic energy. It was found that for all test
days in 2004 and 2010 the two methods of dispatch give
similar marginal prices for energy. The marginal price of
reserve is generally lower for the dispatch with frequency-
based constraints as it often dispatches the system with a
smaller largest contingency and, as a result, a smaller amount
of reserve. The marginal price of kinetic energy ranges from
between 2c and 35c over all the test days and is generally
higher during the lower load hours. Although the possible
benefits of a competitive market for kinetic energy are limited,
this marginal price could be paid to participants providing
kinetic energy. This would provide a reflective price signal of
the value of the kinetic energy to the system. Developers of
sources of generation that may not provide any natural inertial
response, e.g. wind capacity and HVDC interconnection can
use this price signal to determine the feasibility of enabling
their technology to provide an inertial response [3], [25].
Fig. 11. Market prices for July 2004 test day for (a) Energy, (b) Reserve and
(c) Kinetic Energy
G. Effect of Increasing Wind Capacity and Benefits of
Curtailment
In future years the all-Ireland system is likely to experience
a large increase in the amount of wind capacity connected to
the system [18]. There is also concern about the impact wind
capacity will have on small systems if its stored kinetic energy
is not available to the system. The variable nature of the wind
production and its subsequent replacement of generation that
does provide kinetic energy to the system makes the effective
application of heuristic reserve constraints more difficult.
In [26] it is suggested that at high wind capacities
significant curtailment of wind generation may be necessary
for various reasons to ensure security of the system. The three
test days were run on the 2010 system using the dispatch with
frequency-based constraints considering increasing amounts of
wind generation. Each test day was run with a constant amount
of wind generation available throughout the day and it was
again assumed that the wind generations stored kinetic energy
was not available to the system. Results found that the January
2010 and May 2010 test days did not require any curtailment
of wind generation to ensure short-term frequency control of
the system, even with 2000 MW of wind generation. The July
2010 test day did however require very slight curtailment
during some hours with very high available wind generation.
The dispatch algorithm generally found that the least cost
system dispatch with high levels of wind penetration involved
significantly reducing the size of the largest contingency rather
than curtailing the wind generation.
H. Application to Other Electricity Systems
The frequency control constraints developed in this work
are based on the all Ireland electricity system. It was found that
the constraints resulted in a convex boundary, thus allowing
their incorporation into a LP dispatch algorithm. Proof of
convexity or the conditions under which convexity is
conserved cannot be formally found for the general case due to
the nature of how the minimum frequency constraint is
derived. However, it is likely that the constraints will also be
convex for many other isolated systems, although actual
modeling of each systems dynamic characteristics would be
necessary to examine this. In this way the suitability of the
dispatch methodology presented here could be assessed for
each system.
The methodology presented is mainly relevant for smaller
more isolated systems as it is generally assumed that larger and
more interconnected system will not have the same challenges
with frequency control. However, the effects of
interconnection between systems may also be incorporated
into the methodology. This would involve modeling separately
the response of neighboring systems to frequency events and
then equating this response in terms of the 4 control variables
used here. This would not encapsulate all the interactions of
interconnected power systems, but would allow a simple
representation of interconnection to be included.
0
5
10
15
20
M
a
r
g
i
n
a
l

P
r
i
c
e

f
o
r

R
e
s
e
r
v
e

(
/
M
W
)
0
5
10
15
20
25
30
35
40
M
a
r
g
i
n
a
l

P
r
i
c
e

o
f

E
n
e
r
g
y

(
/
M
W
h
)
0
0.1
0.2
0.3
0.4
1 3 5 7 9 11 13 15 17 19 21 23
Hour
M
a
r
g
i
n
a
l

P
r
i
c
e

f
o
r

K
i
n
e
t
i
c

E
n
e
r
g
y

(
/
M
W
s
)
(a)
(b)
(c)
174
VI. CONCLUSION
This work has shown that frequency-based security
constraints can be derived accurately and successfully
incorporated into a LP dispatch algorithm. It has been shown
that this approach provides a least-cost solution to meeting the
frequency control criteria. It was found that a conventional
reserve target dispatch method may not be sufficient to meet
such criteria in a small isolated system. For such a system,
reducing the size of contingencies that the system is exposed to
plays an important role in securely dispatching the system in a
least cost manner. It was found that curtailment of wind
generation, even at times of high wind production and low
load, was generally not necessary to meet the frequency-based
security constraints in a least cost manner.
VII. APPENDIX
The Lagrangian function for the conventional reserve target
dispatch as described by (1)-(4),(6),(7) and (14) is as follows:
( ) ( )
1 1 1 1
, , ,
1 1 1
, 1,
1 1
2,
, , ,
( ) ( ) ( )
1
( ) ( )
N N N N
i i i i P i R T i
i i i i
N N N
Pmin i i i Pmax i i i Rmin i i
i i i
N N
Rmax i i i Rchar i i i i
i i i
Rchar i
P R bp P br R L P R R
Pmin P P Pmax R
R Rmax P R Pmax
Rslope

= = = =
= = =
= =
| | | |
I = + + +
| |
\ . \ .
+ + +
+ +
+

1
( )
N
i
i i
i i
Rmax
P R
Pmin
=
(15)
The Lagrangian function for the dispatch model with the
frequency based constraints as described by (1)-(8) and
(11)-(13) is as follows:
( ) ( )
1 1 1
,
1 1
5
, , ,1 ,2 ,3 ,4
1 1 1 1
, ,
1
, , , ,
(100 )
( )
( ) (
N N N
i i i i P i
i i i
N N
Rocof k k Load i
k i
i k
N N N
minfreq j k j j k j i j i
j k i i
i k i k
N
Pmin i i i Pmax i i i
i
P R KE bp P br R L P
P KE KE
C L C P C R C KE
Pmin P P Pmax

= = =
= =
=
= = = =
= =
=
| |
I = + +
|
\ .
+
+ + + +
+ +

,
1 1
, ,
1 1
, 1,
1 1
2, 1,
1 1
) ( )
( ) ( )
1
( ) ( )
( ) ( )
N N
Rmin i i
i i
N N
Rmax i i i KEmin i i
i i
N N
KEmax i i i Rchar i i i i
i i i
N N
i i
Rchar i i i KEchar i i i
i i i i
R
R Rmax KE
KE KEmax P R Pmax
Rslope
Rmax KEmax
P R P KE
Pmin Pmin

= =
= =
= =
= =
+
+ +
+ +

+ + + +

(16)
Using the standard Kuhn-Tucker optimality conditions the
marginal cost of energy, MC
E
and the marginal cost of reserve
MC
R
for the conventional reserve target dispatch is shown in
(17) and (18).
E P
MC = (17)
R R
MC = (18)
For the dispatch with the frequency-based constraints the
marginal cost of energy can be again given by (16). The
marginal cost of reserve and the marginal cost of kinetic
energy MC
KE
are shown in (19) and (20).
5
, , ,3
1 1
N
R minfreq j k j
j k
MC C
= =
=
(19)
5
, , ,
1 1 1
N N
KE rocof k minfreq j k
k j k
MC
= = =
=

(20)
VIII. ACKNOWLEDGMENT
The authors gratefully acknowledge the assistance of ESB
National Grid during this work. The authors also wish to thank
Shane Rourke for his advice and expertise.
IX. REFERENCES
[1] A. J. Wood and B. F. Wollenberg, Power Generation Operation and
Control, New York: Wiley, 1996.
vol. 9, no. 3, August 1994, pp. 1698-1708.
[3] L. Holdsworth, J Ekanayake, and N. Jenkins, Power system frequency
response from fixed speed and doubly-fed induction generator based
wind turbine, Wind Energy, vol. 7, 2004, pp. 21-35.
[4] W. O. Stadlin, Economic allocation of regulating margin, IEEE
Transactions on Power Apparatus and Systems, vol. PAS-89, no. 4,
1971, pp. 1776-1781.
[5] D. R. Bobo, D. M. Mauzy and F. J. Trefny, Economic generation
dispatch with responsive spinning reserve constraints, IEEE
Transactions on Power Systems, vol. 9, no. 1, 1994, pp. 555-559.
[6] M. Gibescu and C.-C. Liu, Optimization of ancillary services for
system security, in Proceedings of Bulk Power System Dynamics and
Control IV Restructuring, Symposium, Greece, 1998, pp. 351-358.
[7] I. Arnott, G. Chown, K. Lindstrom, M. Power, A. Bose, O. Gjerde, R.
Morfill, N. Singh, Frequency control practices in market
environments, CIGRE/IEEE PES International Symposium, Quality
and Security of Electric Power Delivery Systems, 2003, pp. 143-148.
[8] National Electricity Market of Singapore, Wholesale Electricity
Market Report 2003, [Online]. Available:
http://www.emcsg.com/upload/contentimages/3617_175_Wholesale_El
ectricity_Market_Report_2003.pdf.
[9] Transpower New Zealand, Reserve Management Tool, [Online].
Available: http://www.transpower.co.nz/?id=4452.
[10] NEMMCO Australia, Operating Procedure, [Online]. Available:
http://www.nemmco.com.au/powersystemops/so_op3705v027.pdf.
[11] C. C. Wu and N. Chen, Frequency-based method for fast-response
reserve dispatch in isolated systems, IEE Proceedings on Generation,
Transmission and Distribution, vol. 151, no. 1, 2004, pp. 73-77.
[12] J. O'Sullivan and M. J. O'Malley, A new methodology for the provision
of reserve in a isolated power system, IEEE Transactions on Power
Systems, vol. 14, no. 2, 1999, pp. 519-524.
[13] P. Kundur, Power System Stability and Control, EPRI, New York:
McGraw Hill, 1994.
[14] G. Lalor, J. Ritchie, S. Rourke, D. Flynn and M. J. OMalley, Dynamic
Frequency Control with Increasing Wind Generation, in Proceedings
IEEE PES General Meeting, Denver, Colorado 2004.
[15] J. OSullivan, "Modelling and identification of emergency reserve with
applications to isolated power systems" Ph.D. dissertation, Dept.
Electronic and Electrical. Eng., University College Dublin, Dublin,
1996.
[16] J. G. Thompson and B. Fox, Adaptive load shedding for isolated power
systems, IEE Proceedings on Generation, Transmission and
[17] J. O'Sullivan and M. J. O'Malley, Identification and validation of
simulation, IEEE Transactions on Power Systems, vol. 11, no. 2, 1996,
pp. 851-857.
175
[18] Department of Public Enterprise, Green Paper on Sustainable Energy,
Irish Government. Publications, 1999. [Online]. Available:
http://www.dcmnr.gov.ie/display.asp/pg=557.
[19] Commission for Energy Regulation, Market Arrangements for
Electricity Margadh Aibhlise na ireann (MAE), 2003, [Online].
Available: http://www.cer.ie/cerdocs/cer03230.pdf.
[20] ESB National Grid, Download Centre, 2003, [Online]. Available:
http://www.eirgrid.com/EirGridPortal/DesktopDefault.aspx?tabid=Dow
nload%20Centre.
[21] ESB National Grid, Generation adequacy report 2004-2010, [Online].
Available: http://www.eirgrid.com/EirGridPortal/uploads/Publications/
GAR 2004.pdf.
[22] Commission for Energy Regulation, Self Commitment Versus
Centralised Commitment, 2003, [Online]. Available:
http://www.cer.ie/cerdocs/cer03299.pdf.
[23] Commission for Energy Regulation, A Co-Optimised Energy-Reserve
Market Frequently Asked Questions, 2003, [Online]. Available:
http://www.cer.ie/cerdocs/cer03300.pdf.
[24] P. E. Gill, W. Murray and M. H. Wright, Practical Optimization,
London: Academic Press, 1997.
[25] G. Fujita, K. Ezaki, T. Nakano, R. Yokoyama, K. Koyanagi, T
Funabashi, Dynamic characteristic of frequency control by rotary
frequency converter to link wind farm and power system, in
Proceedings IEEE Power Tech Bologna, Italy, 2003, ref. BPT03-318.
[26] ESB National Grid, Options for Operational Rules to Curtail Wind
Generation, 2004, [Online]. Available:
http://www.cer.ie/CERDocs/cer04247.pdf.
X. BIOGRAPHIES
Ronan Doherty received a B.E. degree in Electronic
Engineering from the University College Dublin in 2001.
He is currently conducting research for a Ph. D. degree in
the Electricity Research Centre, University College Dublin
with research interests in power system operations and
markets, and integration of renewable energy sources. He is
a member of the IEEE.
Gillian Lalor received a B.E. degree in Mechanical
Engineering from University College Dublin in 2001. She is
currently conducting research for a Ph.D. at University
College Dublin, with interests in power system modelling
and control. She is a member of the IEEE.
Mark OMalley received B.E. and Ph. D. degrees from
University College Dublin in 1983 and 1987, respectively.
He is currently a Professor in University College Dublin and
director of the Electricity Research Centre with research
interests in power systems, control theory and biomedical
engineering. He is a senior member of the IEEE.
Appendix D
Frequency Control and Wind
Turbine Technologies
177
Frequency Control and Wind Turbine Technologies
Gillian Lalor, Student Member, IEEE, Alan Mullane, Member, IEEE, and Mark OMalley, Senior Member, IEEE
AbstractIncreasing levels of wind generation has resulted in
an urgent need for the assessment of their impact on frequency
control of power systems. Whereas increased system inertia is
intrinsically linked to the addition of synchronous generation to
power systems, due to differing electromechanical characteristics
this inherent link is not present in wind turbine generators.
Regardless of wind turbine technology, the displacement of
conventional generation with wind will result in increased rates
of change of system frequency. The magnitude of the frequency
excursion following a loss of generation may also increase.
Amendment of reserve policies or modication of wind turbine
inertial response characteristics may be necessary to facilitate
increased levels of wind generation. This is particularly true in
small isolated power systems.
Index TermsFrequency control, power system security,
power system control, wind energy, wind power generation.
I. NOMENCLATURE
A = Swept area of wind turbine (m
2
)
= Blade pitch (
)
C
p
= Power coefcient
i = Current (A)
K
1P
, K
2P
, K
1I
, K
2I
= Gains
K = Supplementary control loop constant
L
m
= Per phase mutual inductance (H)
L
r
= Per phase rotor inductance (H)
L
s
= Per phase stator inductance (H)
L
= L
2
m
L
r
L
s
(H)
= Flux linkage (Wb)
P = Number of machine poles
P
a
= Accelerating aerodynamic power (MW)
p = Differential operator
= Density of air (kg/m
3
)
T
em
= Electromagnetic torque (N m)
T
emref
= Reference electromagnetic torque (N m)
T
ref
= Reference torque (N m)
T
sc
= Supplementary control loop torque (N m)
= Time constant (s)
u = wind speed (m/s)
v = Voltage (V)
= Reference frame angular velocity (rad/s)
s
= Shaft speed (rad/s)
r
= Rotor electrical angular velocity (rad/s)
Subscripts
d, q = Direct, Quadrature axis component
This work has been conducted in the Electricity Research Centre, University
College Dublin, which is supported by Electricity Supply Board (ESB)
Networks, ESB Power Generation, ESB National Grid, Commission for
Energy Regulation, Cylon, Airtricity and Enterprise Ireland.
G. Lalor, A. Mullane and M. OMalley are with the Electricity Research
Centre, University College Dublin, Dublin 4, Ireland. (Ph: +353 (0)1 7161857,
Email: gill@ee.ucd.ie, alan.mullane@ee.ucd.ie, mark.omalley@ucd.ie.
I, P = Integral, Proportional
r, s = Rotor, Stator
II. INTRODUCTION
The electricity industry worldwide is turning increasingly to
renewable sources of energy to generate electricity. Environ-
mental concerns about fossil fuelled conventional generators,
the desire to increase the diversity and security of fuel supply,
and increasing fossil fuel costs are all motivating factors
behind this upward trend. Global targets for the reduction of
carbon dioxide (CO
2
) and other greenhouse gases have been
introduced [1] and separate targets for minimum quantities of
electricity generated from renewable energy sources have also
been established in many parts of the world [2]. Wind is the
fastest growing and most widely utilised of the emerging re-
newable energy technologies in electricity systems at present,
with a total of approximately 40 GW installed worldwide at
the beginning of 2004 [3]. However, the majority of wind
resources are still untapped, and the potential for energy from
wind generation is vast. Constant advances in technology and
the relatively low capital costs when compared with other
forms of renewable energy are signicant contributing factors
to the ongoing rapid growth in the proportion of electricity
being generated from wind.
The ease with which wind generation is integrated into
existing electricity systems depends on a number of fac-
tors, both technical and regulatory [3]. Technical aspects that
need to be considered include the amount and location of
wind generation, wind turbine technology and the size and
characteristics of the electricity system. Regulatory issues
such as compliance with existing system operator rules and
regulations also arise. Of the technical aspects that need to be
considered, system inertia plays an extremely important role
as it determines the sensitivity of system frequency to supply
demand imbalances. The lower the system inertia, the faster
the frequency will change if a variation in load or generation
occurs. Large interconnected electricity systems generally have
sizeable system inertia and large frequency deviations from
nominal are rare. However, frequency deviations on small,
isolated electricity systems, when they occur, tend to be more
sizeable. When connecting wind turbines to small isolated
power systems their contribution to system inertia must be
considered. The addition of synchronous generation to a power
system intrinsically increases the system inertial response.
This intrinsic increase does not necessarily occur with the
addition of wind turbine generators due to their differing
electromechanical characteristics. Therefore, the displacement
of conventional synchronous generation with wind generation
may result in an erosion of system inertial response resulting
178
in increased rates of change of frequency (ROCOF) and
larger frequency excursions. In small isolated systems these
phenomena are particularly challenging due to low system
inertia. Where wind turbines are found to severely impact on
system inertial response, system operators need to consider
altering their frequency control strategies to avoid large rates
of change of frequency and/or large frequency excursions.
Ireland is experiencing a rapid increase in wind generation,
and has recently completed the development of transmission
and distribution grid codes for wind [4], [5]. The main issues
that are addressed, similar to other wind grid codes developed,
are the specic requirements for frequency, voltage and fault
ride through behaviour [4], [6]. As a small system with a peak
load in the region of 6,100 MW at present and relatively low
inertia, frequency control on the Ireland system is an important
issue. This paper examines the impact of increasing wind pene-
tration on frequency control on the Ireland electricity system.
A brief outline of common wind generation technologies is
given in Section III. The wind turbine generator models used
in this study are presented in Section IV, and the test system
is described in Section V. The effects on system frequency
control of increasing wind generation and the opportunities
for the introduction of supplementary control on wind turbine
generators are presented in Section VI. Conclusions are pre-
sented in section VII.
III. WIND GENERATION TECHNOLOGY
Wind turbine generators (WTGs) can be divided into two
basic categories: xed speed and variable speed. A xed speed
WTG generally uses a squirrel-cage induction generator to
convert the mechanical energy from the wind turbine into
electrical energy. During a frequency excursion, the relation-
ship between the system frequency and the electromagnetic
torque of any induction machine will determine the inertial
response. As there is a strong coupling between the squirrel
cage induction generator stator and the power system, and
due to the low nominal slip of 1 - 2% [7], any deviations
in system speed will result in a change in rotational speed.
This linking of rotor speed with system speed gives rise to an
inertial response from the xed speed WTG when the system
frequency falls.
Variable speed WTGs can offer increased efciency in
capturing the energy from wind over a wider range of wind
speeds, along with better power quality and the ability to
regulate the power factor, by either consuming or producing
reactive power. Doubly fed induction generator (DFIG) and
multi-pole synchronous generator are popular types of variable
speed WTGs. Both forms of generator have power electronic
converters between the electrical machine and the power
system. The multi-pole synchronous generator allows variable
speed operation by employing a back-to-back AC/DC/AC
converter attached to the stator of the synchronous machine.
As a result, the stator is isolated from and, consequently,
unaffected by any changes in the frequency of the power
system. Therefore, the power output from the WTG does not
change and no inertial response is obtained during a frequency
event.
The power electronic converter in the DFIG, on the other
hand, is attached to the rotor. The rotor is connected to the
power system through this back-to-back AC/DC/AC converter,
while the stator is connected directly to the power system.
The net power output from the DFIG is the sum of the
power outputs from both the stator and the rotor [8], [9].
During a system frequency excursion, any inertial response
provided by the DFIG depends on the relationship between the
electromagnetic torque of the machine and system frequency.
This relationship, in turn, depends upon the type of controllers
used in the converter, and the parameters of the controllers
[10].
IV. WIND TURBINE GENERATOR MODELLING
The operation and behaviour of a induction generator is well
documented [11], and various models have been developed and
evaluated. The model used for the xed speed wind turbines in
this study is the fth order d-q model of [12], which was found
to have very good accuracy in simulating the rotor speed,
electromagnetic torque, both active and reactive power, and
stator currents. An inertial constant of 3.5 s is assumed for
the xed speed WTG models used in this study [8].
The fth order d-q induction machine model is also em-
ployed in the case of the DFIG wind turbine. However,
the addition of the power electronics converter and related
controllers to this induction machine model is necessary to
model the behaviour of the DFIG. Field-orientated control
(FOC) is a prevalent form of controller setup utilised in the
control of doubly fed induction generators [10], [13]. This type
of controller is implemented in the DFIG model for this study
and further details of the model used for the DFIG WTGs may
be found in [10].
Control of electromagnetic torque allows for the control
of the speed, the basis of the variable speed nature of the
DFIG wind turbine. Field-orientated control allows for the in-
dependent control of electromagnetic torque and stator reactive
power. By the selection of an appropriate reference frame in
the controller of the back-to-back AC/DC/AC converter, it is
possible to have independent control of the electromagnetic
torque through the control of a single variable: the rotor
current, i
dr
. The equation for the electromagnetic torque, T
em
,
written in terms of rotor currents, reduces from
T
em
=
3P
4
L
m
L
s
(
qs
i
dr

ds
i
qr
) (1)
through the appropriate choice of the reference frame, such
that
ds
= 0, to
T
em
=
3P
4
L
m
L
s
(
qs
i
dr
) (2)
The control of i
dr
may also be simplied considerably
through the careful design of the d- and q-axes voltages, such
that
v
dr
= v
dr
(
r
)
L
m
L
s
qs
+ (
r
)
L
L
s
i
qr
(3)
and
v
qr
= v
qr
+
L
m
L
s
p
qs
(
r
)
L
L
s
i
dr
(4)
179
where v
dr
and v
qr
are auxiliary signals in the controller
reference frame and are the outputs from the d-axis and q-
axis proportional integral current controllers,
v
dr
= (i
dr
i
dr
)
K
1P
+
K
1I
p
(5)
v
qr
=
qr
i
qr
K
2P
+
K
2I
p
(6)
where K
1P
, K
1I
, K
2P
and K
2I
are the proportional and inte-
gral gains for the d and q axes current controllers respectively,
and i
dr
and i
qr
are the reference d and q axes rotor currents.
A more detailed description of equations (3-6) is provided in
[12].
Therefore, the full model of the DFIG wind turbine consists
of the standard fth order induction machine d-q model and
the FOC, modelled using equations (3)-(6) above. This model,
combined with the system model described in Section V, can
be used to examine the inuence that increasing proportions
of DFIG wind turbines will have on system frequency control.
Wind model input assumptions vary from constant torque to
constant power [13], [14]. The frequently made assumption of
constant torque means any changes in shaft speed will result
in a change in captured aerodynamic power. The constant
power assumption is also sometimes applied. The captured
aerodynamic power is given by equation (7), and depends
on air density, , swept area of wind turbine, A, the power
coefcient, C
p
and wind speed, u.
P
a
=
1
2
AC
p
(
s
, u, )u
3
(7)
The motivation behind the constant power assumption is that
C
p
remains constant for any changes in shaft speed,
s
. This is
achieved by varying the blade pitch, . The effect on the output
of the WTG models is examined for the both the assumption
of constant power and constant torque.
V. TEST SYSTEM
A. The Ireland Electricity System
The Ireland electricity system is a small isolated system,
consisting of two ac interconnected 50 Hz power systems
operated by Northern Ireland Electricity (NIE) and Electricity
Supply Board National Grid (ESBNG). The system currently
has a peak load of approximately 6,100 MW, and system load
is forecast to grow by between 3% and 4% per annum for
the next six years [15], [16]. Reheat and non-reheat thermal
generators, simple cycle gas turbines, combined cycle gas tur-
bines and limited hydroelectric generation, along with a single
pumped storage station comprise the conventional generation
on the Ireland system at present. The system also has a single
HVDC interconnection to Scotland, with a capacity of 500
MW. However, it should be noted that it is currently not
operated above 400 MW.
At present, the installed wind capacity on the Ireland system
of approximately 300 MW [17] is rapidly increasing and is
expected to reach 1,000 MW within the next three years. In
order to meet the targets of 13.2% of all electricity in the
Republic of Ireland and Northern Irelands proportion of the
UK target of 10% of all electricity to come from renewable
sources by 2010 [2], a total of at least 1,300 MW of wind
generation is required on the Ireland system by 2010. This
signies a much more rapid increase in wind generation than
predicted system load over the next ve years, resulting in
a large increase in the proportion of wind generation on the
Ireland system. However, if the rate of installation of wind
generation continues to follow current trends on the Ireland
system, the capacity of wind generation in 2010 could easily
exceed that required to meet the EU targets.
B. Frequency Control
In accordance with the system grid codes [18], [19] fre-
quency regulation can be provided by each generator on the
system, by virtue of a droop governor, with a droop setting
of 4%. Primary operating reserve (POR), the reserve available
between 5 and 15 seconds subsequent to an event, corresponds
to 75% of the largest infeed onto the Ireland system. At
present, the largest infeed is 422 MW, thus making the
primary reserve requirement 317 MW. This primary reserve
requirement is divided such that the ESBNG and NIE systems
provide 67% (211 MW) and 33% (106 MW) respectively.
Sources of POR include spinning reserve from generating
units online and static reserve, and the proportion of the
POR provided by spinning and static reserve sources varies
with time of day. Static reserve consists of blocks of reserve
that are available almost instantaneously when tripped by the
system frequency falling below the predetermined frequency
setting of each block. These frequency settings range between
49.5 Hz and 49.3 Hz. For example, at night, up to four 73
MW units of the pumped storage station may be operating in
pumping mode, which trip when the predetermined frequency
levels are reached, to provide static reserve. In general, the
contribution of the pumped storage station to POR depends
on the operational mode in which it is running, and varies
depending on system conditions and time of day.
C. System Model
A single busbar model of the Ireland electricity system is
used in this study to model the response of the Ireland system
frequency to supply/demand imbalances, as illustrated in Fig.
1. The system model was developed and is simulated using
Matlab/Simulink. This model has been extensively validated
for up to 20 s following a frequency excursion with historical
data over many years and full details are available in [20]
[24]. The xed speed WTG [12] and DFIG WTG [10] models
were incorporated into the system model. Simulated system
frequency is used as input to generator and load models, and is
calculated in the connecting system (Fig. 1) by the integration
of the power imbalance [21], [25]. At nominal frequency,
each generator model is in steady state, and power output
is constant. When a power imbalance occurs, a frequency
deviation from nominal results. A response from each gen-
erator, dependant on the steady state set-point, droop setting
and inertial characteristics is modelled.
180
Fig. 1. Schematic of system model. Details of individual models may be
found in [10], [12], [20][24].
D. Scenarios
The predicted Ireland electricity system of 2010 is examined
in this study to investigate the impact of increasing wind
capacity on frequency control. The validated system model
[20][24], has been extended to take into account new gen-
erating plant coming onto the system [15], [16], [26]. The
system load is increased in line with predicted values. The
sources of generation in the predicted 2010 electricity system
are assumed to be similar to the current system, although a
number of older thermal units will be decommissioned, and all
new generation connected to the system, with the exception of
wind generation, will be either combined cycle gas turbines or
open cycle gas turbines. Transmission effects are neglected in
this study, as it is assumed that the transmission system will
be adequately enhanced for additional conventional generating
plant and wind generation connecting to the system. It is also
assumed that by 2010 an additional HVDC interconnector
with a 500 MW capacity will link the Ireland system to
Wales, and that both this interconnector and the current HVDC
interconnector with Scotland will be capable of providing
limited frequency response [27] of up to 50 MW. When
available, it is modelled as static reserve, which is triggered
when the frequency reaches 49.5 Hz. For 2010, it is assumed
that the largest infeed and the POR remain at 422 MW and
317 MW respectively.
The system is examined for increasing installed wind ca-
pacities under three different scenarios. These scenarios are
carefully chosen to represent the most extreme situations that
could occur on the Ireland electricity system.
1) Winter Peak (WP): This is the predicted peak load that
will occur in 2010 on the system, and occurs during a winter
business day. In order to meet this peak load, a substantial
amount of generating plant on the system is online, and system
inertia is near maximum. Therefore, from an inertial response
perspective, this case represents a best case scenario.
2) Summer Night Valley (SNV): This is the predicted min-
imum load on the Ireland system in 2010, and occurs during
a summer night. System inertia is very low at this point, and
from an inertial response perspective this case represents a
worst case scenario.
3) Summer Day Valley (SDV): This represents the expected
minimum daytime load, which occurs during a summer non-
business day. While the system inertia is not as low as in the
SNV case, the frequency response may be signicantly differ-
ent as the SNV case has substantial static reserve response.
The range of wind penetration levels from 0 to 2000 MW
was examined for each scenario. Due to insufcient wind
generation data in existence for the Ireland system, analysis of
the likelihood of different wind generation levels coinciding
with the three scenarios above was not possible. Therefore,
no assumption about the coincidence of any particular wind
penetration with different load levels is made.
E. Simulating Procedure
A merit order dispatch is employed in the system model,
determined using historical dispatch data and forecast dis-
patches for future years [26]. Initially, the system model is in
steady state, with generation and load balanced, and frequency
constant at 50 Hz. At time t=1 s, the largest infeed to the
system model, a 422 MW generator, is tripped, resulting in a
power imbalance. System frequency and the resultant response
of each generator are simulated for 15 s subsequent to the
tripping. For each scenario outlined in Section V-D, the impact
of this disturbance on the system frequency for increasing
wind penetration and for different wind turbine technologies is
examined. It is assumed that the capacity of wind generation
on the system displaces an equivalent amount of conventional
generation, and that this wind generation is not constrained
downwards to provide governor response. The conventional
generation providing POR within each scenario remains the
same regardless of wind penetration.
VI. RESULTS AND DISCUSSION
A. Response of wind turbine technologies to system frequency
deviations
The simulated system frequency trace resulting from the
loss of the largest infeed during the WP scenario is illustrated
in Fig. 2 (I). The frequency falls to 49.58 Hz, reaching the
nadir or minimum frequency approximately 3.5 s after the
start of the event.
TABLE I
COMPARISON OF INERTIAL RESPONSE FROM VARIOUS GENERATORS
Generator Type Max. Change Time to
in Power (%) Max. (s)
Conventional Synchronous 4.03 0.80
Fixed Speed WTG, constant power 1.35 2.50
Fixed Speed WTG, constant torque 1.20 2.20
DFIG WTG, constant power 0.01 2.50
DFIG WTG, constant torque 0.01 2.20
The inertial responses of a conventional synchronous gen-
erator with locked governor (purely the inertial response), an
induction machine based xed speed wind turbine generator
and a DFIG based wind turbine generator to this frequency
event are shown in Fig. 2(II) and summarised in Table I.
The inertial constant of the synchronous generator is taken
181
0 5 10 15
49.6
49.7
49.8
49.9
50
F
r
e
q
u
e
n
c
y

(
H
z
)
(I)
0 5 10 15
1
0
1
2
3
4
(II)
Time (s)
C
h
a
n
g
e

i
n

P
o
w
e
r
(
%

o
f

C
a
p
a
c
i
t
y
)
(a)
(b)
(c)
(d)
Fig. 2. (I) Simulated system frequency resulting from the loss of the largest
infeed during WP. (II) Comparison of xed speed WTG and DFIG responses
to the low frequency event in (I): Change in power output (%) of (a) a
synchronous machine with inertial constant of H=4.2 s, (b) the xed speed
WTG, assuming constant power, (c) the xed speed WTG, assuming constant
torque and (d) the DFIG, for constant power and constant torque.
to be 4.2 s, in comparison with 3.5 s [8] for the xed
speed WTG inertial constant, since conventional synchronous
generators generally have higher inertial constants than WTGs.
It can be seen that the conventional synchronous generator
produces the maximum inertial response in the shortest time.
The response of the induction machine based xed speed wind
turbines is slower and lower, due to the reduced coupling of
induction generator rotational speed to system speed compared
with the synchronous generator and the smaller inertial con-
stant. The assumption of either constant torque or constant
power has a small impact on the maximum inertial response,
with models assuming constant power producing a larger
response. The DFIG WTGs show negligible inertial response,
for both constant power and constant torque, as the DFIG
controllers effectively decouple the generator rotational speed
from the system speed [10]. From a system frequency control
perspective, the constant torque assumption is the slightly
more pessimistic approach, and was therefore assumed in the
subsequent results (Sections VI-B and VI-C).
B. System frequency control with increasing wind penetration
The impact of increasing proportions of wind generation
on the system frequency during a variety of low frequency
events was examined. Proportions of DFIG vs. xed speed
wind turbines ranging from 0 to 100% were examined to
represent numerous possibilities of installed wind capacity in
2010.
The loss of the largest infeed (422 MW) during the 2010
winter peak, summer night valley and summer day valley
were simulated. It should be noted that the probability of a
generation trip coinciding with such extreme conditions is
very unlikely, but possible, and the general operating point
of the system lies between these extremes. The important
characteristics of the observed system frequency responses,
namely maximum ROCOF and frequency nadir (minimum
TABLE II
MAXIMUM ROCOF FOLLOWING LOSS OF LARGEST INFEED (422MW)
FOR VARIOUS OPERATING SCENARIOS, WIND TURBINE PENETRATIONS
AND WIND TURBINE TECHNOLOGY TYPE.
WTG Proportions Wind Penetration (MW)
DFIG Fixed 0 500 1000 1500 2000
WP
0% 100% 0.29 0.30 0.32 0.34 0.36
25% 75% 0.29 0.30 0.32 0.34 0.36
50% 50% 0.29 0.30 0.32 0.34 0.36
75% 25% 0.29 0.30 0.32 0.34 0.36
100% 0% 0.29 0.30 0.32 0.34 0.36
SNV
0% 100% 0.48 0.54 0.62 0.72 0.87
25% 75% 0.48 0.54 0.62 0.72 0.87
50% 50% 0.48 0.54 0.62 0.72 0.87
75% 25% 0.48 0.54 0.62 0.72 0.87
100% 0% 0.48 0.54 0.62 0.72 0.87
SDV
0% 100% 0.44 0.48 0.54 0.61 0.69
25% 75% 0.44 0.48 0.54 0.61 0.69
50% 50% 0.44 0.48 0.54 0.61 0.69
75% 25% 0.44 0.48 0.54 0.61 0.69
100% 0% 0.44 0.48 0.54 0.61 0.69
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0.2
0.3
0.4
0.5
0.6
0.7
0.8
M
a
x
i
m
u
m

r
a
t
e

o
f

c
h
a
n
g
e

o
f

f
r
e
q
u
e
n
c
y

(
H
z
/
s
)
(a)
(b)
(c)
Fig. 3. Effect of increasing wind penetration (100% DFIG) on maximum
rate of change of frequency following the loss of the largest infeed (422 MW)
during the (a) WP scenario, (b) SNV scenario and (c) SDV scenario.
frequency), are summarised in Table II, Fig. 3, Fig. 4, Fig.
5 and Fig. 6.
As increasing amounts of installed wind generators displace
conventional generation, regardless of whether DFIG WTGs
with negligible inertial response or xed speed WTGs with
slow inertial response, the net effect is a reduction of system
inertial response. This results in increased rates of change of
frequency during frequency transients. It can be seen from
Table II that maximum ROCOF is sensitive to the amount
of installed wind. Table II also illustrates that the type of
wind turbine technology installed has little inuence on the
maximum ROCOF. Fig. 3, demonstrating the 100% DFIG
case, further illustrates the increase in maximum ROCOF with
increasing wind penetration. It should be noted that identical
responses to those shown in Fig. 3 will be observed for any
182
of the proportions of xed speed to DFIG wind turbines listed
in Table II.
Considering that the ROCOF protection setting for genera-
tors connected to the Ireland electricity system is 0.5 Hz/s, it
can be seen from Fig. 3 that these protection settings would
not be exceeded for the winter peak scenario. This is due to
the large amount of conventional generation online. For the
summer night and summer day valley scenarios, protection
settings would be exceeded at penetrations of wind of ap-
proximately 140MW and 700MW respectively. Above these
penetrations, there is a possibility that additional generation
will trip, exacerbating the initial power imbalance which
caused the low frequency event.
It should be noted here that the maximum ROCOF rates
presented in Table II and Fig. 3 represent maximum simulated
rates based on a ltered differentiation of the frequency signal.
The detailed operation of ROCOF relays installed on the
system would have to be known in order to determine if
generation would trip for the scenarios examined [28]. In
particular the time period over which the rate of change of
frequency calculation is made would have to be examined
in detail. Regardless of the operation of ROCOF relays the
general trend of increased rates of change of frequency will
be observed as wind displaces conventional generation.
While Table II illustrates that maximum ROCOF is inde-
pendent of wind turbine technology, the frequency nadir is
found to depend on the type of wind turbine technology. An
example of the effect of WTG technology type on frequency
nadir during the SDV scenario is depicted in Fig. 4, showing
system frequency responses for no wind and each of 2000 MW
of xed speed and DFIG wind generation. While the ROCOF
is initially similar for both WTG types, the frequency nadirs
reached differ substantially. The xed speed WTG case results
in a frequency nadir reaching the same frequency level as for
the no wind case, while the frequency nadir in the case of the
DFIG WTGs is considerably lower.
0 5 10 15
49.3
49.4
49.5
49.6
49.7
49.8
49.9
50
Time (s)
F
r
e
q
u
e
n
c
y

(
H
z
)
(a)
(b)
(c)
Fig. 4. Simulated system frequency following the trip of largest infeed (422
MW) during the SDV scenario with (a) No wind generation, (b) 2000 MW
xed speed WTGs and (c) 2000 MW DFIG WTGs.
While Fig. 4 shows the frequency responses for a single
penetration level of wind, the effect of increasing penetrations
of both xed speed and DFIG WTGs on frequency nadir
following the loss of the largest infeed (422 MW) is shown
in Fig. 5 and Fig. 6. Both SDV (Fig. 5) and SNV (Fig.
6) scenarios are shown in order to examine the inuence
of increasing wind penetrations on static reserve usage. As
outlined in Section V-D, the mix between static and dynamic
reserve varies for SDV and SNV.
0 200 400 600 800 1000 1200 1400 1600 1800 2000
49.3
49.32
49.34
49.36
49.38
49.4
F
r
e
q
u
e
n
c
y

N
a
d
i
r

(
H
z
)
(a)
(b)
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
50
100
150
200
250
S
t
a
t
i
c

R
e
s
e
r
v
e

T
r
i
p
p
e
d

(
M
W
)
(c)
Fig. 5. Frequency nadir (Hz) and static reserve tripped (MW) following the
loss of the largest infeed (422 MW) for increasing wind penetration during
the Summer Day Valley (SDV) scenario, where (a) frequency nadir for 100%
xed speed WTG case, (b) frequency nadir for 100% DFIG WTG case and
(c) static reserve tripped (for both cases).
0 200 400 600 800 1000 1200 1400 1600 1800 2000
49.35
49.37
49.39
49.41
49.43
49.45
F
r
e
q
u
e
n
c
y

N
a
d
i
r

(
H
z
)
(a)
(b)
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
50
100
150
200
250
(c)
0 200 400 600 800 1000 1200 1400 1600 1800 2000
0
50
100
150
200
250
S
t
a
t
i
c

R
e
s
e
r
v
e

T
r
i
p
p
e
d

(
M
W
)
(d)
Fig. 6. Frequency nadir (Hz) and static reserve tripped (MW) following the
loss of the largest infeed (422 MW) for increasing wind penetration during the
Summer Night Valley (SNV) scenario, where (a) frequency nadir for 100%
xed speed WTG case, (b) frequency nadir for 100% DFIG WTG case, (c)
static reserve tripped for 100% xed speed WTG case and (d) static reserve
tripped for 100% DFIG WTG case.
During the SDV scenario, following the loss of the largest
infeed, a constant amount of static reserve (50 MW) is tripped
in all cases, so Fig. 5 primarily illustrates the differing effect on
frequency nadir of the two WTG technology types. When xed
speed WTGs comprise the wind generation, the frequency
nadir rises by a small amount as wind penetration increases,
due to the distinctive slower inertial response of the xed
speed WTG. However, if DFIG WTGs comprise the wind
generation, the frequency nadir falls further as wind pene-
tration increases, due to the negligible response of the DFIG
WTGs to changing system frequency. Thus, given constant
183
capacity and dynamic/static mix of POR, the frequency nadir
is detrimentally affected if DFIG WTGs displace conventional
generators, and is largely unaffected if xed speed WTGs
displace conventional generation.
Both the inuence of increasing wind penetrations on static
reserve and the impact of static reserve on frequency nadir
can be seen in Fig. 6. When xed speed WTGs comprise the
wind turbine generation, 123 MW of static reserve is tripped
in all cases, and the frequency nadir does not fall below the
0 MW wind penetration case. However for DFIG WTGs, the
tripping of an additional 73 MW of static reserve is required
to prevent the frequency nadir from falling signicantly below
the 0 MW wind penetration case. As such, it may be concluded
that increasing DFIG WTG penetration requires increasing
availability of static reserve to maintain the frequency nadir
above a given threshold.
C. Supplementary response from DFIG
The magnitude of the inertial response of a DFIG depends
on the extent by which the rotational speed changes in re-
sponse to changing system frequency. As the DFIG is designed
to provide accurate control of rotational speed, the coupling
of rotational speed to system frequency and the resulting
inertial response is largely removed [10]. However through
the addition of a simple supplementary loop to the controller
of the DFIG, it is possible to congure the DFIG to change
the electromagnetic torque in proportion to rate of change of
frequency [14].
As a direct differentiation of measured system frequency is
undesirable due to susceptibility to noise, the supplementary
control loop torque, T
sc
, is determined using the following
equation (8) as illustrated in Fig. 7.
T
sc
+ pT
sc
= 2Kp (8)
The supplementary control loop torque T
sc
is added to the
reference torque of the DFIG WTG, T
ref
, to provide the
reference electromagnetic torque, T
emref
. This signal, in the
same way as a natural inertial response, will impart energy to
the system as frequency drops, and absorb energy from the
system as frequency increases, thus increasing the response
of the DFIG above that which would be normally observed.
The magnitude of the response will depend on the controller
parameters (supplementary control loop constant K and time
constant ). The achievable inertial response of the DFIG
WTG is limited by operational constraints such as current
limits. In this study it is assumed the these limits are not
exceeded. It must be noted that in practise, however, some
design modications may be required to prevent deviation
outside operational constraints.
Fig. 7. Supplementary control loop for DFIG WTG controller.
A comparison of the inertial response during a frequency
event from a synchronous generator, a xed speed WTG and
a DFIG WTG with the supplementary control loop included
is illustrated in Fig. 8(II). Instead of a negligible response to
system frequency, as shown in Fig. 2(II), the addition of the
supplementary controller results in a DFIG response similar to
the inertial response of the conventional generator. The size of
the response is dependent on the value of parameter K within
the supplementary control loop, which was chosen to be 3.5
for this study.
0 5 10 15
49.6
49.7
49.8
49.9
50
(I)
F
r
e
q
u
e
n
c
y

(
H
z
)
0 5 10 15
1
0
1
2
3
4
(II)
Time (s)
C
h
a
n
g
e

i
n

P
o
w
e
r
(
%

o
f

C
a
p
a
c
i
t
y
)
(a)
(b)
(c)
(d)
(e)
Fig. 8. (I) Simulated system frequency resulting from the loss of the largest
infeed during WP. (II) Comparison of xed speed WTG and DFIG responses
to the low frequency event in (I): Change in power output (%) of (a) a
synchronous machine with inertial constant of H=4.2 s, (b) the xed speed
WTG, assuming constant power, (c) the xed speed WTG, assuming constant
torque and (d) the DFIG with supplementary control loop included (K = 3.5),
assuming constant power and (e) the DFIG with supplementary control loop
included (K = 3.5), assuming constant torque.
The system frequency following the loss of the largest
infeed during the SDV scenario is illustrated in Fig. 9. This g-
ure is similar to Fig. 4 but includes 2000 MW of DFIG WTGs
with supplementary control. The addition of the supplementary
control loop to the DFIG WTG results in an improvement in
system frequency response, both in terms of the rate of decline
of the frequency and the maximum frequency excursion that
occurs, as indicated in Fig. 9. However, these effects depend
on both the value of the constant K and the time constant
within the controller.
While these results indicate that the addition of the sup-
plementary control loop to the DFIG WTG can potentially be
benecial to system frequency response, the ease of integration
into the current technology is not assessed here.
VII. CONCLUSION
As a consequence of differing inertial response characteris-
tics, rates of change of system frequency will increase as wind
generation displaces conventional generation. The maximum
rate of change of frequency following a loss of generation is
independent of wind turbine technology.
As xed speed wind turbines displace conventional synchro-
nous generation, there is no signicant change in minimum
frequency reached following a loss of generation. This is due
to the inertial response provided by xed speed wind turbines.
As DFIG wind turbines displace conventional synchronous
generation, the frequency nadir following a loss of generation
184
0 5 10 15
49.3
49.4
49.5
49.6
49.7
49.8
49.9
50
Time (s)
F
r
e
q
u
e
n
c
y

(
H
z
)
(a)
(b)
(c)
(d)
Fig. 9. Simulated system frequency following the trip of largest infeed (422
MW) during the SDV scenario with (a) No wind generation, (b) 2000 MW
xed speed WTGs, (c) 2000 MW DFIG WTGs and (d) 2000 MW DFIG
WTGs, with supplementary control loop added
reduces. This is due to the negligible inertial response provided
by DFIG wind turbines.
To allow for increased penetration of wind, a change in
reserve policy may be required. For example, increasing DFIG
wind turbine usage may necessitate additional availability of
static reserve to maintain system frequency above a given
threshold. This is particularly important for small power
systems where system inertia is low.
A possible solution to the lack of DFIG wind turbine inertial
response is through the addition of a supplementary control
loop to provide an inertial response similar to a conventional
synchronous generator. The achievable inertial response of a
standard DFIG WTG is limited by operational constraints,
such as current limits. In this study it was assumed that these
limits were not exceeded. In practise, the implementation of
the supplementary control loop and the maintenance of oper-
ational constraints within required limits may be problematic
and some design modication, such as an increase of DFIG
converter rating, may be required.
ACKNOWLEDGMENT
and interactions with colleagues in Electricity Supply Board
National Grid (ESBNG), Northern Ireland Electricity (NIE),
The Queens University of Belfast (QUB) and the Electricity
Research Centre (ERC) in University College Dublin (UCD).
She is a member of the IEEE.
REFERENCES
[1] The United Nations Framework Convention on Climate
Change. (1997) The Kyoto Protocol. [Online]. Available:
http://unfccc.int/resource/docs/convkp/kpeng.pdf
[2] Directive 2001/77/EC of the European Parliament and the Council of
27 September 2001 on the promotion of electricity produced from
renewable energy sources in the internal electricity market, Ofcial
Journal of the European Communities, L283, pp. 3340, October 2001.
[3] The European Wind Energy Association (EWEA) and the European
Commissions Directorate General for Transport and Energy (DG
TREN). (2003) Wind Energy - The Facts. [Online]. Available:
http://www.ewea.org/06projects events/proj WEfacts.htm
[4] Commission for Energy Regulation (CER). (2004, July) Wind
Farm Transmission Grid Code Provisions. [Online]. Available:
http://www.cer.ie/CERDocs/cer04237.pdf
[5] . (2004, October) Wind Generation Distribution Code Provisions.
[Online]. Available: http://www.cer.ie/CERDocs/cer04318.pdf
[6] ELTRA. (2000) Specications for Connecting Wind Farms to the Trans-
mission Network. [Online]. Available: http://www.eltra.dk/composite-
837.htm
[7] A. P. Mullane, Advanced Control of Wind Energy Conversion Systems,
Ph.D. dissertation, National University of Ireland, University College
Cork, Ireland, March 2004.
[8] J. B. Ekanayake, L. Holdsworth, X. Wu, and N. Jenkins, Dynamic
modelling of doubly fed induction generator wind turbines, IEEE
Transactions on Power Systems, vol. 18, no. 2, pp. 803809, 2003.
[9] P. Pourbeik, R. J. Koessler, D. L. Dickmander, and W. Wong, Integra-
tion of large wind farms into utility grids (part 2 - performance issues),
in IEEE Power Engineering Society General Meeting, vol. 3, Toronto,
Canada, July 2003, pp. 15251528.
[10] A. Mullane and M. OMalley, The inertial-response of induction-
machine based wind-turbines, IEEE Transaction on Power Systems, in
press, 2005.
[11] X. Xu, R. M. Mathur, J. Jiang, G. J. Rogers, and P. Kundur, Modeling
Effects of System Frequency Variations in Induction Motor Dynamics
Using Singular Perturbations, IEEE Transactions on Power Systems,
vol. 15, no. 2, pp. 764770, May 2000.
[12] T. Thiringer and J. Luomi, Comparison of reduced-order dynamic
models of induction machines, IEEE Transactions on Power Systems,
vol. 16, no. 1, pp. 119126, Feb. 2001.
[13] A. Tapia, G. Tapia, J. Ostolaza, and J. Saenz, Modelling and control of a
wind turbine driven doubly fed induction generator, IEEE Transactions
on Energy Conversions, vol. 18, no. 2, pp. 194204, 2003.
[14] J. B. Ekanayake and N. Jenkins, Comparison of the response of doubly
fed and xed-speed induction generator wind turbines to changes in
network frequency, IEEE Transactions on Energy Conversion, vol. 19,
no. 4, pp. 800802, December 2004.
[15] Electricity Supply Board (ESB) National Grid: Transmission
System Operator Ireland. (2003) Generation Adequacy Report:
2004 - 2010. [Online]. Available: www.eirgrid.com
[16] System Operator for Northern Ireland (SONI), Seven Year Transmis-
sion Statement 2003/04 - 2009/10, Tech. Rep., 2003.
[17] Sustainable Energy Ireland. [Online]. Available: http://www.irish-
energy.ie/home/index.asp
[18] Electricity Supply Board (ESB) National Grid. (2002, October) Grid
code. [Online]. Available: http://www.eirgrid.com/
[19] System Operator for Northern Ireland (SONI). Grid Code. [Online].
Available: www.soni.ltd.uk/gridcode.asp
[20] J. OSullivan and M. OMalley, Identication and validation of dynamic
global load model parameters for use in power system simulation, IEEE
Transactions on Power Systems, vol. 11, pp. 851857, 1996.
[21] J. OSullivan, M. Power, M. Flynn, and M. OMalley, Modelling of
frequency control in an island system, in IEEE Power Engineering
Society 1999 Winter Meeting, vol. 1, 1999, pp. 574579.
[22] J. G. Thompson and B. Fox, Adaptive load shedding for isolated
power systems, IEE Proceedings on Generation, Transmission and
Distribution, vol. 141, no. 5, pp. 491496, 1994.
[23] G. Lalor, J. Ritchie, S. Rourke, D. Flynn, and M. OMalley, Dynamic
Frequency Control with Increasing Wind Generation, in IEEE Power
Engineering Society General Meeting, Denver, Colorado, June 2004.
[24] G. Lalor, J. Ritchie, D. Flynn, and M. OMalley, The Impact of
Combined Cycle Gas Turbine Short Term Dynamics on Frequency
Control, IEEE Transactions on Power Systems, In press, 2005.
[25] O. I. Elgerd, Electric Energy Systems Theory: An Introduction. Second
Edition. McGraw-Hill, Inc., 1982.
[26] Electricity Supply Board (ESB) National Grid: Transmission
System Operator Ireland. (2004) Forecast Statement 2004-2010.
[Online]. Available: www.eirgrid.com
[27] Sustainable Energy Ireland (SEI). (2004) Operating Re-
serve Requirements as Wind Power Penetration In-
creases in the Irish Electricity System. [Online]. Avail-
able: http://www.sei.ie/uploads/documents/upload/publications/Ilex-
Wind-Reser rev2FSFinal.pdf
[28] A. Beddoes, P. Thomas, and M. Gosden, Loss of mains protection relay
performances when subjected to network distrbances/events, in CIRED
International Conference on Electricity Distribution 2005, Turin, Italy,
June 2005.
185
Gillian Lalor received a B.E. degree in Mechani-
cal Engineering from University College Dublin in
2001. She is currently conducting research for a
Ph.D. at University College Dublin, with interests
in power system modelling and control. She is a
member of the IEEE.
Alan Mullane obtained a B.E. in Electrical and
Electronic Engineering in 1998 and a Ph.D. in
Electrical Engineering in 2003, both from the De-
partment of Electrical and Electronic Engineering
at University College Cork. In 2004 he joined the
Electricity Research Center as a postdoctoral re-
search fellow. His research interests include non-
linear modelling and control of dynamic systems,
with particular interest in simulation and control
of wind-turbines and their integration into electrical
networks. He is a member of the IEEE.
Mark OMalley received B.E. and Ph. D. degrees
from University College Dublin in 1983 and 1987,
respectively. He is currently a Professor Univer-
sity College Dublin and director of the Electricity
Research Centre with research interests in power
systems, control theory and biomedical engineering.
He is a senior member of the IEEE.
Appendix E
Frequency Control on an Island
Power System with Increasing
Proportions of Combined Cycle
Gas Turbines
187
Frequency Control on an Island Power System
with Increasing Proportions of Combined Cycle
Gas Turbines
Gillian Lalor, and Mark OMalley, Senior Member, IEEE

AbstractAs combined cycle gas turbines (CCGTs) comprise an
ever-increasing proportion of generating capacity on an island
electricity system, the importance of understanding the effects of
their characteristic behaviour on the system, particularly on
frequency control, becomes crucial. A suitable mathematical
model of a CCGT is developed to study its response following a
frequency disturbance. This model is then integrated into a
larger model, representative of the Irish electricity system, and
the effects of increasing proportions of CCGT generation are
examined. The results of this study are then considered in
conjunction with the impact of increased amounts of wind power
and of interconnection via HVDC to a larger system.

Index TermsCombined cycle gas turbine, frequency control,
island electricity system.
I. NOMENCLATURE
SP = Set Point (per unit)
K = 1/Droop
T
s
= Speed Controller Time Constant
T
r
= Ratio Valve Positioner Time Constant
T
gc
= Gas Control Valve Positioner Time Constant
T
f
= Gas Fuel Time Constant
T
cd
= Compressor Discharge Time Constant
T
t
= Temperature Controller Integration Rate
T
i
= Inlet Guide Vane Controller Integration Rate
K
i
= Inlet Guide Vane Controller Constant
T
av
= Inlet Guide Vane Actuator Time Constant
P
gt
= Gas Turbine Power Output (per unit)
T
or
= Allowable Over-firing Temperature (C)
T
a
= Ambient Temperature (C)
T
rc
= Rated Turbine Exhaust Temperature Corrected for
Ambient Temperature (C)
N = Speed (per unit)
T
x
= Exhaust Gas Temperature (C)
W
x
= Exhaust Gas Flow (per unit)
IGV = Inlet Guide Vanes Angle ()

Board (ESB), ESB National Grid, Commission for Energy Regulation (CER)
and Cylon Controls.

G. Lalor and M. OMalley are with the Electricity Research Centre,
Department of Electronic and Electrical Engineering, University College
Dublin, Belfield Ireland (Ph: +353 1 7161857) (e-mail: gill@ee.ucd.ie
Mark.OMalley@ucd.ie).
L
igv
= Inlet Guide Vane Position (per unit)
T
c
= Heat Capacitance Time Constant
T
b
= Boiler Storage Time Constant
P
st
= Steam Turbine Power Output (per unit)
II. INTRODUCTION
T
he Irish electricity system is an island system, comprising
of two AC interconnected power systems, operated by
Northern Ireland Electricity and ESB National Grid. The only
interconnection between this island system and the larger UK
system is the HVDC Moyle Interconnector. The generating
capacity of the system is predominantly composed of thermal
generating units, gas turbines, hydro units and one pumped
storage unit. However, with the advantages of combined cycle
gas turbines (CCGTs) over conventional units, such as
increased efficiency and reduced emissions, combined with
the fact that CCGTs are comparatively quick to install, it is no
surprise that many companies are considering CCGTs as the
best option for expansion. Recently, three large CCGTs have
been installed on the Irish system, and at present, there are
plans for several more. This leads to the question of what
impact these changes will have on the existing electricity
system, especially from the perspective of frequency control.
It has been recognised that when a significant share of the
generation spread is made up of CCGTs, their response
critically affects the system response to disturbances that
could lead to large frequency deviations [1], and as numbers
increase, their response becomes increasingly important.
When the frequency of the system falls, it has several
effects on the CCGT unit. One consequence is that the unit
delivers an inertial response. The size of this response depends
on both the rate and extent of the frequency drop. This
response is of utmost importance to the system, with the
inertial responses of the generating units, along with the load
response on the system, invaluable in alleviating the initial
frequency fall [2].
Another effect of the falling system frequency on CCGT
units is a fall in the speed of the compressor. Consequently the
output pressure of the compressor is reduced, and there is also
a reduction in airflow. Reduced output pressure of the
compressor leads to a decrease in the pressure ratio across the
gas turbine and results in a lower power output from the gas
turbine. The reduction of airflow into the combustor increases
the fuel to air ratio, leading to a rise in temperature. However,
188
due to operational constraints on the machine, the upper
temperature at which the turbine may operate is limited. This
limit can be broken for short periods of time but can impact on
the lifespan of the machine. The temperature controller comes
into play in such a situation and the fuel flow is reduced
accordingly. So clearly the maximum power output of the gas
turbine element of the CCGT during a low frequency event
will be constrained by the temperature control. This results in
a further reduction in power output by the CCGT. Unlike the
inertial response of the CCGT to falling frequency, this drop
in power output with declining frequency is far from
beneficial. During a frequency transient CCGTs may also
suffer from compressor surge, which can adversely affect the
power output [3].
If a CCGT is operating at base load, the unit is not capable
of producing and maintaining any increase in power output.
In fact, if the frequency falls, the temperature controller will
actually reduce fuel demand and the power output of the gas
turbine will drop. Consequently the output of the unit remains
at a value less than its rated output until the frequency of the
system returns to normal.
When operating below base load, a CCGT unit is capable of
providing an increase in power output in response to falling
frequency. The effectiveness of this response is modulated by
the characteristics described above and the maximum
achievable power is lower than at nominal frequency. In
certain circumstances the response can be initially negative
[3]. One option available for preventing the initial reduction
in power output of the CCGT is to modify the unit by
including fast acting inlet guide vanes. These work so as to
oppose the natural reduction in airflow due to the reduction in
compressor speed. This will only be an improvement in the
situation where the unit is running below base load. However
the practise of modifying the CCGT to prevent its natural
response adds additional cost. Unless some incentive to
modify the units is provided, it is unlikely that the response of
future installations of CCGTs will be improved.
If the response of a combined cycle gas turbine to a fall in
frequency can be a decline in the power output, then the
introduction of multiple CCGTs is a worrying prospect for
any system operator of an isolated system. The response from
the unit, instead of helping to reduce the drop in frequency
and rate at which it is falling, could actually exacerbate the
problem, leading to possible tripping and load shedding [4].
The frequency nadir, in the event of a contingency on the
Irish system, is generally reached after about 5 seconds.
Therefore the initial response of the system is vitally
important. On a larger system, this phenomenon would have a
negligible consequence on the overall system frequency.
However this is not the case on this system.
The reason for developing the CCGT component for the
model of the Irish system [2] is to be able to see to what extent
the initial response of the CCGTs affects the frequency. This
is necessary to determine any requirements and alterations to
the existing grid code and market [5] that may be necessary as
larger numbers of these units are proposed and ultimately
come online. In the present Grid Code, a CCGT operating at
base load is expected to maintain constant active power output
for system frequency changes down to 49.5Hz [6].
The impact of an increasing proportion of the Irish
generating capacity coming from CCGTs is a serious enough
prospect. However when the effects of bringing wind
generation onto the system compound this, an in-depth
knowledge of the likely responses of individual units are more
important than ever.
III. CCGT MODEL
A. Modelling Combined Cycle Gas Turbines
Several CCGT models have been developed in recent years,
with varying degrees of complexity and success. Most are
based on the original gas turbine model produced by Rowen
[7][8]. The research shows that there is a lot of interest in the
behaviour of these units. However most research is carried out
based on the assumption that the CCGTs are being integrated
into large interconnected electricity systems. There has been
considerably less research undertaken into the primary
frequency response. Generally longer time periods are
considered with a greater emphasis on the overall response.
A basic mathematical model of the CCGT is to be used in
the study. It is derived from the general gas turbine model
developed by Rowen [7][8] with reference to [9][10][11] and
[12] for the heat recovery steam generator and steam turbine
section of the model.
As each CCGT on the Irish system has its own particular
characteristics, the model needs to be modified to suit each
individual unit. Various modifications have been made to
some of the units on the Irish system during manufacturing
and installation to improve their frequency response, and these
need to be taken into account. One of the more significant
modifications is the inclusion of fast acting inlet guide vanes,
and these are present on one of the existing units. Different
configurations are to be tested to see the individual responses
and to determine exactly which model is suitable for each unit.
Each CCGT unit will be integrated into the overall system
model [2].
However as these modifications lie predominantly within
the parameter values for each unit, rather than in the overall
structure, an outline of the basic model is possible.
B. Overview of Model
The CCGT modelled in this paper consists of a single shaft,
constant speed gas turbine, with modulating inlet guide vanes
(IGV), and a simple heat recovery steam generator (HRSG)
and steam turbine. The IGV control the airflow into the
compressor and as a result the pressure ratios and turbine
exhaust temperature. The HRSG efficiency is optimal when
the temperature of the exhaust gases is at its maximum
allowable level. At partial loads the IGV close sufficiently to
maintain the temperature at this limit, consequently
maximising power output for a given fuel input.
In this model it is assumed that the pressure ratios across
the compressor and turbine are equal (no cooling air is
extracted from the compressor). Also since the fuel flow is
189
very much smaller than the airflow, the gas flow through the
turbine is regarded to be the equivalent to airflow through the
compressor. Since the ambient temperature appreciably affects
the performance of the gas turbine, a correction to incorporate
ambient conditions is built-in to the rated and calculated
exhaust temperatures and gas flow.
It is assumed that prior to any frequency disturbance the
CCGT is operating in a steady state position. In the model
described below, all signals are in per unit, with the exception
of temperatures, which are measured in degrees Celsius
C. Gas Turbine Model
A schematic of the gas turbine model and its controls is
shown in Fig. 1.
1) Fuel System and Turbine
The fuel signal selector takes the outputs of gas turbine
controllers and selects whichever signal requires the least fuel
to determine the fuel signal. A set of fuel limits ensures that
the fuel signal remains within physically attainable
boundaries.
In a gas turbine, there is a no-load fuel requirement that is
necessary at all times for internal load and to maintain a flame
in the combustor, and this requirement is assumed fixed,
irrelevant of the operating point of the gas turbine. To take
this into account, the fuel signal coming from the selector
controls the variable fuel supply only, and is scaled
accordingly and added to the no-load fuel requirement.
In models previously developed [7,8,9,10,11] the fuel
signal is modulated by the speed to get the actual fuel flow.
This is to take into account the effect of the frequency on the
fuel pump. However, having spoken to various sources in
industry, it seems that the fuel supply currently in use is
essentially independent of frequency, and so it is assumed that
the difference between the fuel signal and the fuel flow is
negligible.
The fuel is fed into the combustor through a set of two
valves. The first valve controls the pressure between the two
valves, and the second regulates the actual fuel flow into the
combustor. Two first order transfer functions with relevant
time constants represent the valve positioners and a third
represents the delay in fuel flow through the fuel system. A
short time delay is also incorporated in the fuel system, to take
into account the time delay associated with the combustion
reaction.
The torque in the turbine is then calculated (F2). In the
restricted operating range of this model, the gas turbine torque
has been found to be linear with respect to fuel flow and
turbine speed. The mechanical power output of the gas turbine
is the product of the torque and the speed. An additional first
order lag associated with the compressor discharge volume is
incorporated before the turbine torque is calculated.
2) Exhaust Temperature
There is a transport delay associated with gases in the
turbine and exhaust system and this is included prior to any
temperature calculations. The temperature of the exhaust
gases from the turbine is then determined (F1), the
characteristics of which can be assumed to be linearly
dependent on speed, fuel flow and ambient temperature in this
operating range. This exhaust temperature assumes that the
IGV are fully open however. The exhaust temperature is then
divided by the exhaust flow, to obtain the corrected exhaust
temperature, taking the actual airflow into account.
Fig. 1 Outline of Gas Turbine model and controls
190
The corrected exhaust temperature is measured using an
arrangement of thermocouples. Temperature measurement is
not instantaneous and the time lags associated with the
radiation shield and the thermocouples are represented in the
temperature measurement section.
3) Gas Turbine Controllers
The gas turbine section of a combined cycle gas turbine has
three discrete control loops regulating the fuel signal, and
ultimately the power output of the CCGT. These control loops
are speed control, temperature control and acceleration
control. Acceleration control, which will generally only
influence the fuel signal during start-up, is neglected in this
model. This is due to the assumption that the CCGT is initially
operating at steady state.
a) Speed Control
Under the Grid Code in the Irish Electricity System [6], all
speed controllers are droop governors, with a droop setting of
four percent required. The speed governor is fed by the
frequency error, which is the difference between nominal and
actual frequency. The output of the speed governor, added to
the set point of the unit, is fed into the fuel signal selector.
b) Temperature Control and Inlet Guide Vane Control
The measured exhaust temperature is compared with the
maximum allowable exhaust temperature and the error
generated is fed into the temperature controller. The
temperature controller produces a signal, which is then fed
into the fuel signal selector. The temperature controller has no
impact on the fuel signal unless the output is negative.
The measured exhaust temperature is compared with the
rated exhaust temperature, which has been corrected to take
into account the ambient temperature. The error produced is
fed into the inlet guide vane temperature controller, to produce
a signal to control the IGVs. The IGV actuator is represented
by a first order lag.
4) Over-Firing
On some CCGTs over-firing is permitted. The transient
temperature is allowed to rise above the rated exhaust
temperature for a short period of time. Even though over-
firing
is not generally used on the CCGTs on the Irish system, the
model has the capability if required. In the case of over-firing,
the rated exhaust temperature used in the comparison with the
measured exhaust temperature at the input to the temperature
controller is raised temporarily.
D. HRSG and Steam Turbine Model
The purpose of this CCGT model is to study the primary
response of the CCGT (approximately the first twenty
seconds) following a frequency event. Therefore the dynamics
of the HRSG and the steam turbine will have little impact on
the overall response of the unit due to the long time constants
associated with the processes. The steam turbine will not be
able to provide any primary response of its own due to the fact
that it is normally operated in sliding pressure mode.
A simplified HRSG model has been included in the CCGT
model. The steam production of the HRSG primarily depends
on the exhaust gas temperature and the flow of the exhaust
gases and therefore these are the inputs. The resultant steam
flow then passes through two time lags, which take into
account the heat capacitance and the much longer boiler
storage time constants. The output from the steam turbine is
then simply a scaled form of the steam flow.

Fig. 2. Outline of Heat Recovery Steam Generator and steam turbine
model
E. Parameter Values
The parameter values used in the CCGT models are largely
based on standard industrial values and the values used in
previous work in this area. However some tuning of the model
to represent individual units on the Irish system has taken
place with the limited data available at the moment. More
detailed monitoring of CCGT responses is currently taking
place and more thorough tuning and validation of the
developed model is proposed once this data is available.
IV. SYSTEM MODEL
The system model being used in this study is a frequency
model of the island electricity system of Ireland. This
emergency reserve model of the Irish system [2] reproduces
the primary frequency response of the Irish system following
a system event. It is a single busbar model, consisting of
dynamic models for each generating unit and also a single
measurement based dynamic load model. The inertias of the
individual units are lumped together as a single generating
inertia. This is achieved by summing the kinetic energies of
the individual units to obtain a single generation kinetic
energy value. The single kinetic energy value is then
incorporated in the feedback loop to calculate the system
frequency. This simplification is possible since it is assumed
that the frequency at all points on the system is the same at all
times and thus any change in kinetic energy will be felt
uniformly through out the network.
The generating capacity of the system largely consists of
thermal generators, with gas turbines, hydro generators,
combined cycle gas turbines and a single pumped storage unit.
In this study the effects of bringing increasing proportions of
combined cycle gas turbines onto the system are being
examined. Since thermal units are the oldest generating units
on the system, in this study these units are being replaced pro
rata with combined cycle gas turbines to study the effects of
bringing increasing proportions of CCGTs onto the system.
V. RESULTS
The software used for these simulations is Matlab Simulink.
Initially 3300MW is generated and consumed on the system.
Base loaded thermal units generate 1200MW and the other
2100MW is generated by a mixture of thermal units, gas
turbines [7][8] and a single pumped storage unit, all at an
191
initial operating point of eighty percent of capacity. Ambient
Temperature is assumed to be 15C. The inertias of the
individual units are lumped together into one inertial constant,
as described in Section III, and all units on the system,
including CCGTs, have a 4% droop. Combined cycle gas
turbine units, each of generating capacity of 300MW, are then
introduced onto the system. These base loaded CCGTs replace
base loaded thermal plant one at a time, until the entire base
loaded portion of the generation consists of base loaded
CCGTs. It is assumed for simplification that the inertias of the
CCGTs are identical to the inertias of the large thermal units
that they are replacing.
At time t=5s, an increase of 10% occurs in the load on the
system. This is as equivalent to losing 330MW of generation,
except the inertia of the unit is not lost. It is important to note
that the size of the largest generating unit on the Irish system
is now 400MW, so this is not the worst-case scenario. The
frequency response of the system is shown in Fig. 3 and Fig.
4. Note: For clarity, responses in Fig. 3, 4 and 5 are shown for
the extreme cases of (a) no base loaded CCGTS and (b) four
base loaded CCGTs. The responses for the intermediate cases
of one, two and three base loaded CCGTs lie between these
responses.

Fig. 3. System frequency response following 10% step load increase, with
(a) No CCGTs, (b) 1 CCGT, (c) 2 CCGTs, (d) 3 CCGTs and (b) 4 CCGTs.

Fig. 4. System frequency response following 10% step load increase (close
up of Fig. 3), with (a) No CCGTs, (b) 1 CCGT, (c) 2 CCGTs, (d) 3 CCGTs
and (b) 4 CCGTs

It can be seen from Fig. 3 and Fig. 4 above that when there
are no base loaded combined cycle gas turbine units, the
frequency nadir is reached at 0.542Hz. As base loaded
thermal units are replaced by base loaded CCGTs, it can be
seen that the frequency nadir is lower as the proportion of
CCGTs is increased. When four CCGT units are online
(1200MW) the frequency drop is 0.657 Hz. Thus by
replacing base loaded thermal units with base loaded CCGTs,
there is a 20% increase in frequency drop. It can also plainly
be seen that the rate at which the frequency falls is influenced
by the amount of CCGT units online, with the rate increasing
with higher amounts of CCGT generation.

Fig. 5. A 300MW thermal units response following 10% step load
increase, with (a) No CCGTs, (b) 1 CCGT, (c) 2 CCGTs, (d) 3 CCGTs and
(b) 4 CCGTs

In Fig. 5 above the power output of a single 300MW
thermal unit free to respond to any frequency change is
presented. It can be seen that a peak of 34.9MW is reached,
which corresponds to the increase in power output of the unit,
when there are no CCGTs on the system. When the base
loaded portion of the generation consists entirely of CCGTs, it
can be seen that this peak is 13.9% higher, at 40.9MW.
However, not only are the governing units responding with
higher power output with increasing amounts of CCGTs
online, the rate of increase is marginally faster too.

Fig. 6. A simulated example of the power output change of a 300MW base
loaded CCGT generated by the model, when (a) No de-rating is applied and
(b) the CCGT is de-rated by 1%. The frequency is also included.

The deviation in power output with changing frequency of a
base loaded CCGT as generated by the model is illustrated in
192
Fig. 6. It can be seen that the active power drop follows the
decline in frequency, and also that as long as the frequency
remains below nominal frequency, the active power will
remain below its rated value.
It was clearly illustrated in Fig. 3 and 4 that as the number
of base loaded CCGTs on the system increase, they have a
detrimental impact on the system frequency. Fig 7 reveals that
this negative effect does not grow linearly with increasing
numbers of base loaded CCGTs, but actually grows more
rapidly.

Fig. 7. The lowering of the frequency nadir with increasing proportions of
base loaded CCGTs on the system

VI. DISCUSSION
From the results presented in Section V, it can be seen that
as the proportion of base loaded combined cycle gas turbine
units is increased, replacing alternative base loaded units pro
rata, the frequency response of the system is adversely
affected. The frequency nadir deteriorates by 0.115Hz or 20%
when just over a third of the system is made up of base loaded
CCGTs. This deterioration in frequency is due to the inability
of base loaded CCGTs to maintain continuous active power
output when frequency falls without exceeding temperature
limits. A simulated example of this characteristic is illustrated
in Fig. 6.
As the number of base loaded CCGTs increases, the effect
on the frequency is exacerbated. However this worsening of
frequency response does not appear to be linear, but in fact
deteriorates more quickly. The reason for this is that as
increasing numbers of base loaded CCGTs are brought online,
a lower minimum frequency is achieved, which in turn causes
the power drop of each individual CCGT to increase further
still.
There are two possible means of preventing or minimising
the drop in active power with frequency of the base loaded
CCGT. The first is to allow over-firing for as long as the
frequency remains below nominal frequency in order to
sustain a steady output. However over-firing is not normally
used by CCGTs on the Irish system. The second option
available is to de-rate the CCGT unit. Although this method
will reduce the characteristic described above, it may not be
the best solution from an economic point of view. This is due
to the fact that CCGTs are generally designed to operate at
base load and this is where efficiency is greatest. Fig. 6
illustrates the case of a base-loaded 300MW CCGT both
before and after de-rating of 1% has been applied.
For other units operating online, at a load point below base
load, the consequences of any increase in the rate and extent
of frequency fall from nominal point are important, since they
would be required to respond to a greater extent and also at a
faster rate. This would increase fuel costs and also wear and
tear on the machine.
When minimising frequency deviation is of such vital
importance to the security of a power system, any increase in
the frequency drop is critical and needs serious consideration
[13]. From the point of view of the system operator, anything
that increases the frequency drop is very significant.
In the context of studying frequency control on the Irish
power system other concurrent developments need to be
considered. As well as growing numbers of CCGTs, there are
an increasing number of wind turbines being installed on the
system. These wind turbines do not generally have any
frequency control ability, and may not have any inertial
response to the changes in frequency. Currently the Moyle
interconnector does not provide frequency control and it also
has no inertia. So while the influence of these factors on the
overall system response to a contingency may not be
individually significant, the collective effect will be much
more substantial and could raise serious security concerns for
the system operator. In a borderline situation, on the threshold
of tripping customers, the effect of any further drop in
frequency is crucial. The development of ancillary services
markets will also have an impact on frequency control. The
strategy used by Independent Power Producers (IPP) when
operating their CCGTs is inherently related to the way in
which the markets are operated. IPP will not part load their
plant unless sufficient incentive is provided.
The loss of the HVDC interconnector may become the
largest single contingency on the system and may result in a
need for additional reserve. With the importance of adequate
reserve capacity, as the proportion of CCGTs on the system
grows, they may have to operate at less economical, part load
conditions in order to provide this reserve.
VII. APPENDIX
F1: T
x
= {1/[1+0.005*(15-T
a
)]}*{[T
r
-453*(N
2
+4.21N+4.42)
*0.82*(1-W
f
)]+722*(1-N)+1.94*(MaxIGV-IGV)}

F2: Torque = 1.3*(W
f
-0.23)+0.5*(1-N)

F3: W
x
= N*[519/(T
a(F)
+460)]*(L
igv
)
0.257

VIII. ACKNOWLEDGMENT
and interactions of A. Egan, ESB, T. Wilson, Viridian, J.
OSullivan, ESB National Grid, D. Barry, ESB National Grid,
N. King, ESB National Grid, P. Hayes, Synergen

193
IX. REFERENCES
[1] Hannett, L.N. and Feltes, J.W., Testing and model validation for
combined-cycle power plants, IEEE Power Engineering Society Winter
Meeting 2001, Vol. 2, pp.664-670.
[2] O'Sullivan, J. and O'Malley, M. J., "Identification and validation of
simulation", IEEE Transactions on Power Systems, Vol. 11, pp. 851 -
857, 1996.
[3] A. Ross, Interim Report on the Performance of large CCGT plants
under Non-Standard Frequency Situations PB Power, [Online]
www.gsp.co.nz/library/gsc_june2601/appendix/appendix6_situations.do
c
[4] O'Sullivan, J. and O'Malley, M. J., "A new methodology for the
provision of reserve in an isolated power system", IEEE Transactions on
Power Systems, Vol. 14, pp. 519 524, 1999.
[5] Flynn, M., Sheridan, P., Dillon, J. and O'Malley, M. J., "Reliability and
reserve in competitive electricity market scheduling", IEEE Transactions
on Power Systems, Vol. 16, pp. 78 87, 2001.
[6] ESB National Grid, Grid Code, Version 1.1, October 2002 [Online]
www.eirgrid.com
[7] W.I. Rowen, Simplified Mathematical Representations of Heavy-Duty
Gas Turbines, ASME, vol. 105(1) 1983 (Journal of Engineering for
Power, Series A, October 1983), pp. 865-869.
[8] W.I. Rowen, Simplified Mathematical Representations of Single Shaft
Gas Turbines in Mechanical Drive Service, Turbomachinery
[9] A. Bagnasco, B. Delfino, G.B. Denegri and S. Massucco, Management
and Dynamic Performances of Combined Cycle Power Plants During
Parallel and Islanding Operation, IEEE Trans. On Energy Conversion,
vol. 13, no. 2, June 1998, pp. 194-201.
plants in power system studies, IEEE Trans. on Power Systems, vol. 9,
no. 3, August 1994, pp. 1698-1708.
[11] Q. Zhang and P.L. So, Dynamic Modelling of a Combined Cycle Plant
for Power System Stability Studies, IEEE Power Engineering Society
Winter Meeting, 2000,Volume:2, 2000 pp. 1538 -1543.
[12] C. Goulding, Development of a Modular Combined Cycle Simulation
Model, M. Eng. Sc. Dissertation, Dept. Mech. Eng., University College
Dublin, 2001.
[13] O'Sullivan, J. and O'Malley, M. J. "Economic dispatch of a small utility
with a frequency based reserve policy", IEEE Transactions on Power
Systems, Vol. 11, pp. 1648 - 1653, 1996.
X. BIOGRAPHIES
Gillian Lalor received a B.E. degree in Mechanical engineering from
University College Dublin in 2001. She is currently conducting research for a
Ph. D. in the University College Dublin.

Mark OMalley received B.E. and Ph. D. degrees from University College
Dublin in 1983 and 1987, respectively. He is currently a Professor University
College Dublin with research interests in power systems, control theory and
biomedical engineering. He is a senior member of the IEEE.

Appendix F
Dynamic Frequency Control with
Increasing Wind Generation
195

Dynamic Frequency Control with
Increasing Wind Generation
Gillian Lalor, Julia Ritchie, Shane Rourke, Damian Flynn and Mark J. OMalley

Abstract Frequency control is essential for the secure and
stable operation of a power system. With wind penetration
increasing rapidly in many power systems, ensuring continuous
power system security is vital. The frequency response to a
disturbance on the all Ireland system is simulated for a range of
installed wind capacities under different system conditions. The
purpose of this study is to assess the effects of increased wind
generation on system frequency, and the security of the system
following such disturbances.
Index Terms Frequency control, power generation control,
power system security, wind.
I. INTRODUCTION
A
s concerns about the impact of conventional electricity
generating technologies on the environment continue to
grow, there is a trend towards deploying renewable sources of
energy. Future targets for renewable energy generation have
already been outlined in many countries. In the Republic of
Ireland the EU target for renewable energy is 13.2% of
electrical energy production by 2010, while the corresponding
target for Northern Ireland is 10% [1]. Adoption of wind as a
source of energy is well established and technological
advances are continually increasing the flexibility and
efficiency of wind generation. It is expected that up to 94% of
the proposed renewable energy target in Ireland will be
achieved using wind generation [2].
The electricity system on the island of Ireland is a small
50 Hz system, with a current peak load in the region of
6,000 MW. The system comprises two AC interconnected
power systems, operated by Northern Ireland Electricity (NIE)
and ESB National Grid (ESBNG). There is a DC
interconnection between Northern Ireland and Scotland,
which has a capacity of 500 MW. Due to the relatively small
size of the system, frequency excursions are often sizeable and
at present a frequency event is defined as a deviation of
frequency below 49.5 Hz [3]. System inertia is vital in
determining the rate of fall of frequency following such an
event the lower the inertia, the faster the system frequency
will fall. Consequently, any reduction in inertial response
could be critical following a significant frequency event on
the system.

Board (ESB), ESB National Grid, Commission for Energy Regulation (CER),
Cylon Controls and Enterprise Ireland.
Gillian Lalor, Shane Rourke, and Mark J. OMalley are with Electricity
Research Centre, Department of Electronic and Electrical Engineering,
University College Dublin, Belfield, Ireland. (e-mail: gill.lalor@ee.ucd.ie,
shane.rourke@ee.ucd.ie, mark.omalley@ucd.ie).
Julia Ritchie and Damian Flynn are with the School of Electrical and
Electronic Engineering, The Queens University of Belfast, N. Ireland (e-
mail: j.a.ritchie@qub.ac.uk, d.flynn@qub.ac.uk )
In general, wind turbine generators (WTGs) may be divided
into two principal categories: fixed speed and variable speed.
Fixed speed WTGs employ an induction generator to convert
mechanical energy, generated by the wind turning the wind
turbine rotor, into electrical energy. Frequency control and
voltage support from such machines is difficult, and they do
not provide these ancillary services as a general rule.
However, due to the coupling between rotor speed and system
frequency through the gearbox, the fixed speed WTG
provides an inertial response when the system frequency falls.
The magnitude of this response depends on the stored energy
of the rotor, in addition to the rate of change of frequency.
There are two classes of variable speed wind turbines. The
first is the direct drive wind turbine, which uses a synchronous
generator to convert mechanical energy into electrical energy.
A power electronics converter connects the stator to the grid,
enabling variable speed operation. The second, and more
common implementation is the doubly fed induction generator
(DFIG). Again, attachment of a power electronics converter,
this time to the rotor windings, allows variable speed
operation. Variable speed WTGs offer improved ancillary
service capabilities over fixed speed machines. One drawback
of the effective decoupling of the rotor from the power system
is that variable speed WTGs do not provide any inherent
inertial response to changing system frequency. Recent
studies, however, indicate that, with the addition of a
supplementary control loop, it is possible for DFIGs to
contribute to the system inertia. Since it is thought that no
significant effect on wind turbine aerodynamics will result, it
is possible that future installations of DFIGs will include this
technology [4].
The availability of an inertial response from the wind
turbines is a key factor in determining the effects that
increasing wind generation will have on the system. At times
of reasonably high wind speeds, or even lower wind speeds
with large numbers of installed wind turbines, wind
generation may displace conventional generation on the
system. At present, there is an installed capacity approaching
270 MW of wind generation on the all Ireland electricity
system, with connection agreements in the Republic of Ireland
for a further 600 MW. In addition, 1000 MW is also currently
planned in the near future on the combined Irish system. If
196
wind turbines provide no inertial response, the effect on
system inertia would be increasingly detrimental as the
proportion of wind increases. As a result, to ensure system
reliability, wind could need to be curtailed. An alternative to
curtailing wind would be the reduction in size of the largest
infeed.
Due to the small size of many wind farms in relation to
conventional plant, a large proportion of wind generation is
connected as embedded generation at the distribution level in
many power systems. However, larger wind farms, in
particular offshore developments, may be connected to the
transmission system. In the event of a fault on the network,
leading to a potential islanding condition, loss-of-mains
protection is a requirement for all embedded generation. Wind
turbines currently installed on the all Ireland system apply
either rate of change of frequency (ROCOF) protection or
vector shift protection. Depending on the sensitivity of the
protection equipment settings, however, sympathetic tripping
of the wind turbines can occur during frequency events on the
system. Clearly this poses serious security issues when the
system is already attempting to recover from the initial event,
especially if large amounts of wind generation are online.
For the above reasons, and recognition of further issues
such as load regulation and fault ride through capability, the
Transmission System Operator (TSO) in the Republic of
Ireland has stated that the proposed levels of wind penetration
pose an increased risk in terms of the stability, security and
reliability of the all Ireland system. An immediate halt on the
increase of wind generation connected to the system has
recently been requested until its impact has been thoroughly
assessed [5].
The objective of this work is to study the short-term
dynamic effects of increasing wind penetration on system
frequency control and the reliability of the system.
II. SYSTEM MODEL
The electricity system model developed for this study
represents an emergency reserve model of the Ireland
electricity system. This model of the ESBNG and NIE systems
reproduces the primary frequency response following a
system event. It is a single busbar model, consisting of
dynamic models for each generating unit and a representative
dynamic load model. The simulated response is valid for
15-20 s following an event.
The conventional generating capacity of the combined
system consists of a combination of thermal generators, gas
turbines, hydroelectric generators, combined cycle gas
turbines and a single pumped storage station. There is also a
HVDC link between Northern Ireland and Scotland. The
system is assumed to be in steady state at nominal frequency
prior to an event on the system and the change in power
output from each unit due to frequency disturbance is
calculated in the model. It is possible therefore to neglect units
that provide no reserve, with the exception of their
contribution to the overall system inertial response.
The inertias of the individual units are combined together to
form a single inertia term. This is achieved by summing the
kinetic energies of the individual units along with the load to
obtain a single kinetic energy value, which is then
incorporated in the feedback loop to obtain the system
frequency. This simplification is possible by assuming that the
frequency at all points on the system is the same, and thus any
change in kinetic energy will be felt uniformly throughout the
network. The system frequency is calculated by integrating the
power imbalance between the load and generation, while
taking into account the system inertia.
An initial model of the ESBNG system [6, 7],
representative of the small isolated electricity system of the
Electricity Supply Board in the Republic of Ireland as it was
in 1996, consisted of large thermal generators modelled using
simple boiler and steam turbine models [8]. Validation was
performed using system data from tests carried out and
comparison with previous frequency events that occurred on
the system. The pumped storage station at Turlough Hill, with
its different operating modes, was also incorporated.
Hydroelectric generating units were assumed to provide little
or no reserve, due to their small size and the belief that the
governors on these units usually operated at the fully open
position when online. As such, they were neglected. A simple
ramp based model was used to model the primary response of
the gas turbine units. The system load was represented using a
single model, accounting for the frequency sensitivity of the
load. This load model was validated using measurement-based
techniques with data from system tests.
A similar dynamic model of the NIE system, as it was circa
1986, developed in order to examine the system reliability and
response of units to frequency events on the system was also
available [9]. The system generation consisted of a
combination of reheat and non-reheat thermal generators,
which responded through governor action to any change in
frequency.
The two models described above have been combined and
updated to provide as accurate a reproduction of the entire
Irish electricity system in 2003 as possible with the available
knowledge and data. Since the development of the original
models, new generating plant that has been introduced onto
the system has been incorporated into the updated model and
older decommissioned plant has been removed. The
interconnector between Northern Ireland and the Republic of
Ireland has returned to service and a DC interconnector now
exists between Northern Ireland and Scotland.
A major addition to the model is the inclusion of a
combined cycle gas turbine (CCGT) representation [10],
which is based on existing gas turbine models [11, 12] with
the addition of a heat recovery steam generator (HRSG) and
steam turbine [13, 14, 15]. The steam turbine may be
neglected in this case, apart from its contribution to the system
inertia, by assuming that its response is negligible up to 20
seconds following a frequency event.
The developed frequency model of the all Ireland system is
being used to study the impact of bringing new and diverse
generation, such as wind, onto the Irish system. It is also
currently being used in a study of the effects of combined
cycle gas turbines on the dynamic response of the system
following a frequency event [10].
197
III. MODELLING WIND GENERATION
The majority of the installed WTGs on the Ireland system at
present are of fixed speed technology. It is expected, however,
that nearly all future installations will be of variable speed
design. As previously outlined, the current technology for
variable speed wind turbines means that these units make no
contribution to the overall system inertia.
Two different wind models are used in this paper. One
model contributes to the system inertia and the other provides
no inertia.
It has been assumed in this study that wind turbines provide
no reserve in response to changes in the system frequency.
Although a mix of ROCOF and vector shift protection
equipment is being used on the all Ireland system at present,
their principle of operation is similar and, therefore, the model
is equipped with ROCOF protection.
IV. TEST SYSTEM
The system model has been modified for the purpose of this
study and represents the predicted systems for 2006 and 2009,
including the additional generating capacity expected to come
online. The model also takes into account the anticipated
decommissioning of several units [2, 16]. These years are
chosen for convenience, representing planning years in the
most recent ESBNG forecast statement [2]. For each year, two
extreme cases are considered: winter peak (predicted peak
load for the year), and summer night valley (lowest annual
load). The load and dispatch for each case were established
based on available historical data, and forecasts from ESBNG
[2]. In NIE the loads were assumed to be a scaled
representation, based on historical relationships, of those on
the ESBNG system.
The primary reserve requirement for the all Ireland
electricity system is 80% of the largest infeed, which would be
400MW in the case of the winter peak load. The primary
reserve requirement is divided between ESBNG and NIE,
such that ESBNG provides 210 MW and NIE are responsible
for 110 MW. During summer night valley, the largest infeed is
limited to 320 MW, and the primary operating reserve
requirement is reduced accordingly. During the summer night
valley, the pumped storage station at Turlough Hill is
pumping and provides a large proportion of the primary
operating reserve.
For each year, the predicted level of wind capacity is
considered, in conjunction with a base case of no wind
generation. These projections are calculated to meet the
national indicative targets set out by the EU for the
contribution of electricity from renewable energy sources to
gross electricity consumption by 2010 as listed earlier [1].
The system is assumed to be in steady state prior to a
disturbance on the system. At 5 seconds, the largest infeed
(400 MW or 320 MW) is tripped. For different levels of wind
penetration, the frequency nadir and the maximum rate of
change of frequency are calculated. The increase in primary
operating reserve (reserve at frequency nadir) required to
offset any increased drop in the system frequency is also
calculated if the nadir is lower than in the base case with no
wind. It is important to note, however, that as the system
inertia is reduced, an increase in primary reserve can reduce
the severity of the frequency excursion, but will have no
beneficial effect on the rate at which the frequency falls
immediately subsequent to the event.
V. RESULTS
A selection of results from the 2009 scenario described in
section IV is presented. Projected wind capacity on the 2009
system to meet the EU targets is 1,562 MW, apportioned as
1,200 MW on the ESBNG system, and 362 MW on the NIE
system. For both the winter peak and summer night valley
scenarios, all 1,562 MW of wind capacity is online, but the
power produced by the wind turbines in each case may vary
from 10 to 100%. For each scenario, a base case with no wind
connected to the system is also included.
A. No inertial response from wind generation
Making the initial assumption that WTGs provide no
inertial response, then as the amount of wind generation on
the system increases (replacing conventional plant capable of
providing an inertial response) the total system inertia will
clearly decrease.
In Fig. 1, the impact of increasing the power produced by
wind turbines on the system frequency during the 2009 winter
peak load is shown. For the base case, when no power is being
generated by wind, the frequency drops to a nadir of 49.41
Hz. As the power generated by wind increases to 100% of
installed wind capacity, it can be seen that the frequency falls
further. So, for example, at 50% of capability, the nadir is
49.39 Hz, while at 100%, the minimum frequency has fallen
to 49.36 Hz.

Fig. 1. System frequency following a 400 MW generation trip during the
2009 winter peak, with wind turbines providing no inertia: (a) No wind, (b)
wind generating at 50% of installed wind capacity, and (c) wind at 100%.

The effect of reducing the system inertia due to increased
wind generation is also apparent on examination of the initial
rate at which the frequency falls following the generation trip.
As the system inertia is reduced, the maximum rate of change
of frequency increases from -0.20 Hz/s for no wind
generation, to -0.24 Hz/s when wind is generating at 100%
output.
The effect of a 320 MW generation trip on the system
frequency during the summer night valley for the 2009 system
is illustrated in Fig. 2. The installed wind capacity remains
198
at 1,562 MW and again the wind is assumed to provide no
inertia.
Due to the lower system demand during the summer night
valley, the proposed level of installed wind in 2009 constitutes
a much higher proportion of generation than during the winter
peak. Consequently, the effect on the system frequency nadir
and the rate at which the frequency falls is more evident.

2009 summer night valley, with wind turbines providing no inertia: (a) No
wind, (b) wind generating at 50% of installed wind capacity, and (c) wind at
100%.

The frequency now falls to 49.42 Hz when no wind is
generating. When wind is generating at 50% of installed
capacity, the frequency nadir falls to 49.38 Hz. The minimum
frequency of 49.33 Hz, which is 15% lower than in the base
case, occurs when the power from installed wind capacity
reaches 100%. This is a significant increase in the frequency
deviation and may well cause serious concerns for the security
of the system load shedding on the Irish system begins at a
frequency of 49.3 Hz.
The maximum rate of change of frequency for the 2009
summer night valley scenario varies between a minimum
value of -0.34 Hz/s with no wind generation, to a maximum of
-0.57 Hz/s when all the wind capacity is generating maximum
output. As will be discussed later, typical ROCOF settings on
the Irish system range between -0.4 and -0.5 Hz/s.
B. Inertial response provided by wind generation
Alternatively, it may be assumed that all wind turbines
contribute to the system inertia - an inertial constant of 3.5 s as
suggested in [17] is used. The results in the previous
subsection suggest that problems with increased wind
generation are likely to be discovered during periods of low,
rather than high, demand. Consequently, the 2009 summer
night valley scenario, with 1,562 MW of wind capacity, is
illustrated in Fig. 3.
Now, when wind generation is connected to the system, the
individual inertias of all the wind turbines contribute to the
system inertia. Consequently, at low wind speeds, when wind
turbines are operating at low power outputs, the system inertia
increases above that for the base case, i.e. wind power
displaces only a small proportion of conventional generation
with its associated inertia, and yet contributes the full inertial
response of the wind capacity. Therefore, when wind is
producing 25% of the rated output, the system frequency nadir
rises to 49.44 Hz, which is 3.5% above the base case of no
wind. The frequency falls at a maximum rate of -0.3 Hz/s,
0.04 Hz/s slower than the base case. At a wind power
contribution of 75% of installed wind capacity, the nadir has
dropped below the base case minimum frequency, and reached
49.41 Hz. The maximum rate of frequency change has also
increased by 8.8% to -0.37 Hz/s.

2009 summer night valley, with wind turbines providing an inertial response:
(a) No wind, (b) wind generating at 25% of installed wind capacity, (c) wind
at 75%, and (d) wind at 100%.

C. Scenario comparison and discussion
The results of Fig. 1-3 demonstrate that increased wind
penetration will impact on the frequency nadir and the initial
rate of change of frequency. The difference between the two
extremes of all, or none, of the turbines providing an inertial
response is represented in Fig. 4.

Fig. 4. Increased drop in frequency nadir as wind generation increases: (a)
Wind with no inertia and (b) 1,562 MW wind capacity connected providing
inertia.

In the case of wind turbines providing no inertia, any wind
brought onto the system has a detrimental effect on the
frequency nadir, and this effect will be proportional to the
amount of wind power being generated. However, if the wind
capacity is assumed to provide an inertial response, the system
frequency benefits from the additional inertial contribution at
low wind capacity factors. This benefit persists until the wind
power produced by the installed wind capacity reaches 58.3%
(910 MW). For increasing amounts of wind generation
199
beyond this level, the frequency nadir deteriorates. The
minimum frequency reached, however, at 100% wind
generation is significantly less than when wind contributes to
system inertia, with a drop in the frequency nadir of 0.027 Hz
(5% below the base case), as compared with 0.087 Hz drop
(15% below the base case) without inertial response.
A comparison of the initial rate of change of frequency with
increasing wind power generated for the scenarios with, and
without, an inertial response is shown in Fig. 5. When wind
contributes to the system inertia, the maximum rate at which
frequency falls is reduced below the base case for wind power
generation levels below 910 MW. As such, for a wind
capacity factor of up to 58.3%, the system response to a
frequency event is at least equivalent to, if not better than,
when no wind generation is online, both in terms of the
minimum frequency reached and the maximum rate at which
the frequency falls.
Fig. 5. Increase in rate of change of frequency as wind generation increases:
(a) Wind with no inertia, (b) 1,562 MW wind capacity connected providing
inertia.

If a ROCOF protection setting of -0.4 Hz/s is assumed for
WTGs in the system model, it can be seen that with just
567 MW of wind generation (36.3% of the installed wind
capacity) with no inertial response, the maximum rate of
change of frequency reaches this limit. Any increase in wind
generation beyond this level will cause all the wind generation
to be tripped and the frequency to drop below 47 Hz the
minimum frequency for which all generation must remain
synchronised to the system [3]. Such a scenario is represented
in Fig. 6, with wind generation set at 40% of its capacity. This
scenario does not recognize any load shedding that would, of
course, occur.
The levels of wind generation considered are extremely
probable, implying that wind may need to be curtailed during
periods of low demand. If the ROCOF protection setting was
increased to -0.5 Hz/s, a wind capacity factor of up to 78%
can be tolerated before the ROCOF protection would be
triggered in the summer night valley scenario. In contrast, if
the wind turbines are assumed to contribute to the system
inertia, an initial rate of change of frequency of -0.4 Hz/s is
only reached for wind generating at 95% (1484 MW) of
capacity. On this basis, the requirement for wind curtailment
would be almost negligible.

2009 summer night valley, with wind turbines providing no inertia and
ROCOF protection set at 0.4 Hz/s: (a) No wind, (b) wind generating at 40%
of capacity.

At each level of wind generation, it is proposed that the
frequency nadir should not be less than the base case
(assuming no wind generation). This objective can be
achieved by increasing the primary reserve on the system at
each wind penetration level, as illustrated in Fig. 7.

Fig. 7. Increase in primary reserve requirement as wind generation increases:
(a) Wind with no inertia, (b) 1562 MW wind capacity connected providing
inertia.

Considering first the case where wind provides no inertial
contribution, the primary reserve required to return the
minimum frequency to the base case increases with wind
penetration levels, up to a maximum of 12% when all
1,562 MW of wind is generating at rated output. If wind is
instead providing an inertial response, no additional reserve is
required up to a capacity factor of 58.3% (from Fig. 4), with
the requirement increasing up to a maximum of 4.5% at
maximum power production.
VI. CONCLUSIONS
This study examines the impact on frequency control of
increasing wind generation on the all Ireland electricity
system. In particular, a likely representation of the generation
system (conventional plant and wind turbines) for the 2009
system is considered.
200
Two extreme scenarios are examined, whereby all, or none,
of the wind turbines provide an inertial response following a
frequency transient. In practice, the truth will lie somewhere
in between, whereby wind farms will be composed of a
mixture of fixed speed and variable speed machines (some of
which may incorporate an auxiliary inertial feedback loop),
along with any future technologies. If wind provides no
inertia, times of low system demand may be problematic. If
the installed wind capacity can contribute to system inertia,
the adverse effect on system frequency is considerably
mitigated. Although additional primary reserve will limit the
observed frequency deviation, it will not impact on the initial
rate of change of frequency. Additional inertia, such as
flywheel technology, is required to reduce the initial rate of
change of frequency. As a result, at high levels of wind
generation, either the largest infeed needs to be reduced, wind
generation needs to be curtailed, or wind turbines should be
required to provide an inertial response.
VII. ACKNOWLEDGEMENT
and interactions with colleagues in ESB, J. OSullivan, D.
Barry (ESB National Grid), R. Doherty, E. Denny and A.
Keane (ERC).
VIII. REFERENCES
[1] Directive 2001/77/EC of the European Parliament and the Council of 27
September 2001 on the promotion of electricity produced from
renewable energy sources in the internal electricity market, Official
Journal of the European Communities, October 2001, L 283, pp. 33-40.
[2] ESB National Grid, Forecast Statement 2003-2009,
[3] ESB National Grid, Grid Code, http://www.eirgrid.com/.
[4] Ekanayake, J.; Holdsworth, L.; Jenkins, N., Control of DFIG wind
turbines, Power Engineering Journal, Vol. 17, No. 1, Feb 2003,
pp. 28-32.
[5] ESB National Grid Report, Interim policy on wind connections,
CER/03/282, http://www.cer.ie/.
[6] O'Sullivan, J.; O'Malley, M. J., Identification and validation of dynamic
global load model parameters for use in power system simulation, IEEE
Transactions on Power Systems, Vol. 11, 1996, pp. 851-857.
[7] O'Sullivan, J.; Power, M.; Flynn, M.; O'Malley, M., Modelling of
frequency control in an island system, Proceedings of IEEE Power
Engineering Society 1999 Winter Meeting, Vol. 1, pp. 574-579.
[8] IEEE Working Group on Prime Mover and Energy Supply Models for
System Dynamic Performance, Dynamic models for fossil fueled steam
units in power system studies, IEEE Transactions on Power Systems,
Vol. 6, No. 2, 1991, pp. 753-76.
[9] Thompson, J. G.; Fox, B., Adaptive load shedding for isolated power
systems, IEE Proceedings on Generation, Transmission and
[10] Lalor, G.; OMalley, M., Frequency Control on an Island Power System
with Increasing Proportions of Combined Cycle Gas Turbines,
presented at IEEE Powertech Conference, Bologna, June 2003.
[11] Rowen, W.I., Simplified mathematical representations of heavy-duty
gas turbines, ASME, Vol. 105(1), (Journal of Engineering for Power,
Series A, October 1983), pp. 865-869.
[12] Rowen, W.I., Simplified mathematical representations of single shaft
gas turbines in mechanical drive service, Turbomachinery
[13] Bagnasco, A.; Delfino, B.; Denegri, G.B.; Massucco, S., Management
and dynamic performances of combined cycle power plants during
parallel and islanding operation, IEEE Transactions On Energy
Conversion, Vol. 13, No. 2, June 1998, pp. 194-201.
Vol. 9, No. 3, August 1994, pp. 1698-1708.
[15] Zhang, Q; So, P.L., Dynamic modelling of a combined cycle plant for
power system stability studies, Proceedings of IEEE Power
Engineering Society Winter Meeting, 2000, pp. 1538-1543.
[16] ESB National Grid, Generation adequacy report 2003-2009,
[17] Ekanayake, J.B.; Holdsworth, L.; Wu, X; Jenkins, N., Dynamic
modelling of doubly fed induction generator wind turbines. IEEE
Transactions on Power Systems, Vol. 18, No. 2, 2003, pp. 803-809.
IX. BIOGRAPHIES
Gillian Lalor received a B.E. degree in Mechanical Engineering from
University College Dublin in 2001. She is currently conducting research for a
Ph.D. at University College Dublin, with interests in power system modeling
and control.

Julia Ritchie received a B.Eng degree in Electrical and Electronic
Engineering from The Queens University of Belfast in 2000. She is currently
studying for a Ph.D. in the area of intelligent control and modelling of power
plants and power systems at The Queens University, Belfast.

Shane Rourke received his B.E. degree from the National University of
Ireland in 1999. He has recently completed a M. Eng. Sc. in the Department
of Electronic and Electrical Engineering, University College Dublin. His
research interests are in power systems economics, operations and control.

Damian Flynn is a lecturer in Power Engineering at The Queen's University
of Belfast. His research interests involve an investigation of the effects of
embedded generation sources, especially renewables, on the operation of
power systems. He is also interested in advanced modelling and control
techniques applied to power plant. He is a member of the IEEE
Mark OMalley received B.E. and Ph.D. degrees from University College
Dublin in 1983 and 1987, respectively. He is currently a Professor in
University College Dublin with research interests in power systems, control
theory and biomedical engineering. He is a senior member of the IEEE.

Frequency Control On An Island Power System

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Frequency Control On An Island Power System

Uploaded by

Copyright:

Available Formats

Frequency Control on an Island Power

You might also like