You are on page 1of 6

A Solution of Transient Stability Constrained Optimal Power Flow

Junji Kubokawa Tadayoshi Matsuo, Hiroshi Sasaki Yue Yuan


Hiroshima Institute of Tedhnology Hiroshima University Honhai University

Hisanori Itou, Ryuuji Tachi, Takumi Matsubara
Chubu Electric Power Co. Inc.


Abstract
Transient stability constrained optimal power flow (TSCOPF)
has been recognized as a potentially mighty tool for secure and
optimal operation planning of power system since its advent,
especially in nowadays open transmission access environment.
This paper proposes two improvements in TSCOPF proposed
by reference (1). First, in order to reduce the number of
equations of the time steps in TSCOPF, we will introduce
variable time-step width method. Next, we will improve
numerical problems at critical case, where the convergence of
TSCOPF will be difficult at the terminal-point of simulation
period. Computation results on IEEJ WEST10 and IEEJ
WEST30 model system demonstrate the effectiveness of
proposed algorithms.

Introduction
Optimal Power Flow (OPF) has been in existence since 1960s.
The main purpose of an OPF program is to determine the
optimal operation state of a power system by optimizing a
particular objective while satisfying certain specified physical
and operational constraints. Nowadays, power system planners
and operators often use OPF as a powerful assistant tool both
in planning and operating stage. Although a power system
operating point obtained by OPF is economical, it may face
dynamical stability problems. The reason is that conventional
OPF does not include transient stability constraints.
For this reason, Transient Stability Constrained OPF
(TSCOPF) has become an attractive research topic recently.
With TSCOPF, optimal operating state where the power system
remains stable after certain contingencies can be obtained.
Up to now, there exist two basic methods of solving TSCOPF
problem. The first method, represented by reference (2) and (3)
solves transient stability constraints in OPF directly. We call it
direct dynamic OPF method in this paper. The second
method, as in reference, is equivalently converts dynamic
problem into an optimization problem with finite dimensions.
Here we refer to it as indirect dynamic OPF method (4).
Reference (2) proposed a formulation of TSCOPF with a large
number of variables and equations, consequently it is time-
consuming even for a one contingency small scale problem.
The hybrid approach with time -step dynamic simulation and
static optimization was proposed in reference (3). The method
solves large system, however, it faced some difficulties in
updating Lagrange multipliers between the iterations. The
method proposed in reference (4) was tested only on small
systems.
This paper proposes two improvements in TSCOPF which is
based on our previous work (1). In order to reduce the number
of equations of time steps in TSCOPF, we will introduce
variable time -step width method. In the proposed method,
small time step width will be used at near fault period, then
larger step width will be applied to the other period. Next, we
will improve numerical problems at critical case, where the
convergence of TSCOPF will be difficult at the terminal-point
of simulation period. In this case, since some diagonal
elements will become large, we will use it for detecting critical
case.
In order to demonstrate the effectiveness of proposed
algorithms, we applied it to IEEJ WEST10 and IEEJ WEST30
model system. The calculation time is reduced by nearly 70%
with the variable step-width. Also the simulation results
demonstrated that the proposed algorithm has good
convergence at critical case.

Nomenclature
i
rotor angle of ith generator
i
rotor speed of ith generator
0
rated rotor speed of generators
i
M moment of inertia of ith generators
i
D damping constant of ith generator
mi
P mechanic power input of ith generator
ei
P electric power output of ith generator
COI
position of the inertial center
, ,
i i i
a b c fuel cost coefficients of thermal plant i
,
gi ri
P Q active and reactive power injection at bus i
,
li l i
P Q active and reactive power load at bus i
ij ij
G jB + transfer admittance between buses i and j
i
j
i
V e

magnitude and phase of voltage
i
V
&
at bus i
t integration time -step width
max
T maximum integration period
nb number of buses
ng number of active power sources
nr number of reactive power sources
nt number of integration time intervals
G
S set of active power sources
R
S set of reactive power sources
N
S set of buses
T
S set of integration steps
( ) lower limits of variables and quantities
( ) upper limits of variables and quantities
Bulk Power System Dynamics and Control - VI, August 22-27, 2004, Cortina dAmpezzo, Italy 129

Transient stability model used in TSCOPF
Synchronous machine representation
In our study, the classical generator model for transient stability
analysis is adopted. It allows the transient electrical
performance of the machine to be represented by a simple
voltage source of fixed magnit ude E behind an effective
reactance
d
x . This model offers considerable computational
simplicity.

( )
0
0
(1)
i i
i i i i mi ei
G
M D P P
i S



+

&
&
where
( ) ( )
2
1
sin cos
ng
ei i ii i j ij i j i j ij i j
j
i
P E G E E B E E B

1 + +
]


In the above equations,
ij ij ij
Y G jB + is the driving point
admittance (i=j) and the transfer admittance (i?j).
ij
Y have to
be changed only in the case that there is a change in the
configuration of the network because of fault or switch
operation.

Center of inertial (COI)
In describing the transient behavior of the system, it is
convenient to use inertial center as a reference frame. The
generators angles with respect to COI are used to indicate
whether or not the system is stable. For an ng-generator system
with rotor angles
i
and inertia constant
i
M , the position of
COI is defined as:
1 1
(2)
ng ng
COI i i i
i i
M M




Formulation of TSCOPF problem
Objective function
Minimize total fuel cost:
2
1
(3)
2
G
i i gi i gi
i S
F a b P c P

_
+ +

,



Equality constraints
a) Power flow equations
pf
H :
In our study, the polar coordinate form power flow equations
are used:
( )
( )
cos sin 0
(4)
cos sin 0
i j ij ij ij ij li gi
j i
i j ij ij i j ij li ri
j i
V V G B P P
V V G B Q Q

+ +
+



b) Swing equations
s
H :
By the adoption of any implicit integration rule, equation (1)
can be discretized at each time step. The differential swing
equation set(1) can be converted to the following numerically
equivalent algebraic equation set using the trapezoidal rule:
( ) ( )
( )
( )
1 1
0 0
1 0
1 1 0
0
2
(5)
2
0
,
t t t t
i i i i
t t t t
i i i i mi ei
i
t t
i i mi ei
i
G T
t
t
D P P
M
D P P
M
i S t S

1
+
]

+

1
+ +
1
]


where
2
1
{ sin[ ]
cos[ ]}
ng
t t t t
ei i ii i j ij i j
j
i
t t
i j ij i j
P E G E E B
E E G


c) Initial -value equations
iv
H :
In order to obtain the initial values of rotor angle
0
i
and
constant voltage
i
E in the swing equations, the following
initial-value equations are introduced:
( )
( )
0
2 0
sin 0
(6)
cos 0
i gi i gi di gi
gi i gi i gi di ri
EV x P
V EV x Q



+


Inequality constraints
For the sake of convenience, inequality constraints are divided
into two groups G
uc
and G
c
. G
uc
group contains all the
inequality constraints as that in conventional OPF, while G
c

group consists of the transient stability constraints.

a) Inequality constraints G
uc
:
(7)
gi gi gi G
ri ri ri R
i i i N
P P P i S
Q Q Q i S
V V V i S





b) Stability constraints G
c
:
As mentioned, generators angles with respect to COI are used
to indicate whether or not the system is stable:
0 0
(8)
,
i COI
t t
i COI
G T
i S t S









Formulation of TSCOPF
Assume that x is defined as a 1 n vector:


T
control state n
x x x R
1

]


130 Bulk Power System Dynamics and Control - VI, August 22-27, 2004, Cortina dAmpezzo, Italy
Then, a Dynamic TTC problem may be formulated as the
following NLP problem:

minimize ( )
subject to ( ) 0
( )
f x
h x
g g x g


(9)

where
1
( ) [ ( ), , ( )]
T
m
h x h x h x LL ,
1
( ) [ ( ), , ( )]
T
r
g x g x g x LL .

By introducing slack variable vectors ,
r
l u R , system (9) can
be transformed to:

minimize ( )
subject to ( ) 0
( ) 0 ; ( ) 0
( , ) 0
f x
h x
g x g l g x g u
l u

(10)

Define a Lagrangian function associated with (10) as:

( , , ; , , , , ) ( ) ( ) [ ( ) ]
[ ( ) ]
T T
T T T
L x l u y z w z w f x y h x z g x g l
w g x g u z l w u

+
% %
% %
(11)

where
m
y R and , , ,
r
z w z w R % % are Lagrange multipliers.

, z z w w % % .
Based on the perturbed Karush-Kuhn-Tucker (KKT) optimality
conditions, we have the following equations:

( ) ( ) ( )( ) 0
0
0
( ) 0
( ) 0
( ) 0
( , ) 0, 0, 0 & 0
x
l
u
y
z
w
L f x h x y g x z w
L LZe e
L UWe e
L h x
L g x g l
L g x g u
l u y z w

+

+


+

(12)

where , , ,
r r
L U Z W R

are diagonal matrices with the element
, ,
i i i
l u z and
i
w . 0 > is a perturbed factor. [1,...,1]
T r
e R .
By applying Newtons method to the perturbed KKT equations
(11), the correction equation can be expressed as:


2 2 2
1 1
0
0
0
0
0
0
[ ( ) ( ) ( ) ( )]
( ) ( )( )
( )
( )
( )
m r
i i j j j
i j
x
l
u
T
y
T
z
T
w
y h x z w g x f x x
h x y g x z w L
Z l L z L
W u U w L
h x x L
g x x l L
g x x u L


+ +
+ + +
+
+


+

(13)

where
0 0 0 0 0 0
( , , ; , , )
x l u y z w
L L L L L L

are the values at a point of
expansion and denote the residuals of the perturbed KKT
equations.
2 2
( ) , ( )
i
f x h x and
2
( )
j
g x are Hessian matrices
of ( ) , ( )
i
f x h x and ( )
j
g x .

Reduction of Correction Equation
In order to handle inequality constraints efficiently, a reduced
correction equation i s introduced. This reduction method is
very effective for Dynamic TTC problem.
By eliminating ( , , , ) l u z w from (12), we can derive the
following reduced correction equation:

( , ) ( ) ( )
( ) 0
T
x H J x
y J x

1 1 1

1 1 1

] ] ]
g g
(14)

where

2 2 2
1 2
1 1
1
1 1 1 1
0 0
( ) ( ) ( ) ( ) ( )
( ) ( ) ( )
( ) ( )
( , ) ( ) ( )
( )[ ( ) ]
( )
m r
i i j j j
i j
r
j j T
j j
j j j
T
w z
H H H y h x z w g x f x
w z
g x g x
u l
J x h x
h x y f x
g x U WL L ZL U L e
h x


1
+ + +
1
]
+


+

g
g


It is obvious that the reduced correction equation has
eliminated both variable inequality constraints and functional
inequality constraints. The size of (14), which is determined
only by the number of variables and equality constraints, is
much smaller than that of (13).


Algorithm of the Method
Initialization: Set iteration counter 0 k ; define centering
parameter (0,1] and tolerance
6
10

; chose a starting
point for primal variables and dual variables.

Begin 0,1,... k :
Step 1: (Test for Convergence)
Compute complementary gap:
1
( )
r
k m m m m
m
CGAP l z u w

(15)
If the operating point satisfied the convergence criteria
( )
k
CGAP < , then output the optimal solution and stop.
Otherwise, do Step 2 to Step 5.

Step 2: (Compute the Perturbed Factor)
2
k
k
CGAP
r
(16)

Step 3: (Compute the Perturbed Newton Correction)
Solve the reduced correction equation (14) for ( , )
k k
x y , then
the following equations for ( , ; , )
k k k k
l u z w :

Bulk Power System Dynamics and Control - VI, August 22-27, 2004, Cortina dAmpezzo, Italy 131
0
0
1
1
( )
( )
( )
( )
T
k k z
T
k k w
k k
k k
l g x x L
u g x x L
z L Z l LZe e
w U W u UWe e

+
'

+
'

(17)
The perturbed Newton correction
k
v is:
[ , , ; , , ]
T
k k k k k k k
v x l u y z w

Step 4: (Determine the Maximum Step Length)
Perform the ratio test to determine the maximum primal and
dual step lengths that can be taken in the Newton direction:
min min : 0; : 0 ,1
min min : 0; : 0 ,1
( 1,2,..., )
m m
k m m
m
m m
m m
k m m
m
m m
l u
P l u
l u
z w
D z w
z w
m r

_
< <
' ;

,
_
< >
' ;

,

(18)

Form the step length matrix:
0.99 [ ,..., ; ,..., ]
k k k k k
diag P P D D

The scalar 0.99 is a safety factor to ensure that the next point
will satisfy the strict non-negativity conditions imposed on the
slack variables.

Step 5: (Update Variables)
Update
the primal and dual variables by:

1 k k k k
v v v
+
+ (19)

then return to Step 1.

End

Computational implementation
TSCOPF contains a large number of constraints. For a system
with ng-generators and one contingency are assumed to have
the same number of integration time intervals nt, the number of
inequality constraints imposed by considering transient
stability is about ng*nt, and the number of equality constraints
imposed is about twice of this. Because of the dramatically
increasing of the scale of TSCOPF problem compared with
conventional OPF, it is very challenge to develop efficient
implementation approaches.
By examining the framework of the primal-dual Newton IPM
algorithm, it is obvious that major computational burden stems
from solving the correction equations. In considering this
problem, we could set out from two ways. One is to reduce the
size of the correction equations as far as possible; the other is
to make properly arranging of the correction equations so that
sparsity programming technique can be exploited.

Reduce the size of the correction equations
First, it is obvious that if we can reduce the number of equality
constraints, we can certainly reduce the size of the correction
equations. We would like to mention here, in TSCOPF study,
the key point in the swing equation set (1) is how to express the
electric power output of the machine
ei
P . It greatly influences
how to incorporate transient stability constraints into TSCOPF
formulation. In our study, we introduce the formulation in
Section3, which incorporated network equations into swing
equations. Our motive for doing this is to reduce the size of the
correction equations, as the number of equality constraints
decreased considerably if network equations are incorporated.
Secondly, if a large integration time-step width t can be used,
the number of equality constraints can also be decreased.
However, in numerical transient stability solution, the
integration time-step width influences the analyzing accuracy.
Reference (5) reported that for solving the differential
equations with the implicit trapezoidal rule of integration the
results are acceptable with a time -step width as large as 0.01
second. Because of this, in our study, we exploit a time -step
width 0.01s.

The arrangement of the correction equations
Fig.1 illustrates the matrix frame of correction equations of
TSCOPF. Where PG and QR indicate control variables
gi
P ( )
G
i S and
ri
Q ( )
R
i S respectively; PF, SE and IV indicate
the variables corresponding to power flow equations, swing
equations and initial-value equations respectively. This matrix
is a symmetrical matrix and fairly sparse. In order to reduce
both storage and solution time, only the non-zero elements in
upper triangle and diagonal need to be saved.

nr*
nr
4nb*
4nb
ng*
ng
4ng*
4ng
4ng*4ng(t=1,,T
max
)
PG
IV
zero elements
non-zero elements in upper triangle and diagonal
non-zero elements in lower triangle
QR
PF
t=1
SE
t=2
SE
:
t=Tmax-
1 SE
t=Tmax
SE

Fig.1 Matrix frame of correction equations of TSCOPF



Improvements in TSCOPF
This paper proposes two improvements in TSCOPF. First, in
order to reduce the number of differential equations of time
steps in TSCOPF, we will introduce variable time-step width
method. Next, we will improve numerical problems at critical
132 Bulk Power System Dynamics and Control - VI, August 22-27, 2004, Cortina dAmpezzo, Italy

case, where the convergence of TSCOPF will be difficult at the
terminal-point of simulation period.

Change of integration time-step width
As mentioned above, if a large integration time -step width can
be used, the number of equality constraints would decrease.
However, if we use a fixed time-step width with larger than
0.01 seconds, accuracy of the obtained solution would not be
satisfactory.
In order to reduce the number of equations of the integration
parts of TSCOPF, we introduced variable time-step width
method. In the proposed method, smaller time-step (0.01s) was
used at near fault period, then larger time-time step (0.05s)
were applied to the other period.

Numerical problem improvement
After several simulations, we found that there are numerical
problems at critical case, where the convergence of TSCOPF
will be difficult at the terminal-point of simulation period. In
this case, since some diagonal elements will become large, we
will use it for detecting the critical case. In order to avoid the
numerical difficulties at the terminal-point, the simulation
period was extended when the diagonal elements become
larger than specified threshold.

Test results and discussions
Simulation conditions
a) Contingencies: All of the contingencies are three-phase
grounding fault (3LGO). For each contingency, the fault occurs
at 0.1s and is removed 70ms later by the opening of one of the
double line. This is consistent with the simulation condition of
the IEEJ WEST10 and IEEJ WEST30 model system in IEEJ
technical report No.754 (6).

Table 1 System Data
IEEJ WEST10 IEEJ WEST30
node 27 115
branch 26 124
generator 10 30
transformer 10 30
http://www.iee.or.jp/pes/model/english/index.html

b) Stability threshold: For all the machines, we assigned 100
degree and +100 degree as the lower and upper limits of angles
with respect to COI respectively. In real-world power system,
such thresholds could be determined by operating experience.

c)Computer spec: We used a Sun Ultra-10 workstation, which
has a 440MHz Ultra
TM
SPARC
TM
-IIi CPU.

Introduction of variable time-step width method
In order to demonstrate the effectiveness of the variable time-
step width method, we have applied TSCOPF to both the
conventional method (CASE1) and the proposed method
(CASE2).

a)CASE1 : T
max
=2sec
t=0.01sec fixed
The maximum integration period is set to be 2.0s and the
integration time-step width is fixed to be 0.01s for all
simulation period.

b)CASE2 : T
max
=2sec
t=0.01sec 0~0.5sec
t=0.05sec 0.5~2sec
The maximum integration period is set to be 2.0s. Time -step
width 0.01s was used at near fault period, then 0.05s was
applied to the other period. In this case, the total number of
time-step will be reduced 60% compared with CASE1.

Test results of CASE1 and CASE2
Table 2 and 3 shows the simulation result for CASE1 and
CASE2 respectively. The total cost for CASE1 and 2 is almost
the same for all contingencies for both system (Table 2 and 3).
Also, the difference of each optimal generator outputs in
CASE1 and 2 were less than 2% in all simulations. The result
indicates that the almost same operating point can be obtained
even if it introduces variable time -step width method with five
times larger time -step width. Moreover, calculation time has
been reduced about 70% by introducing variable time -step
width method, except contingency h.
CASE2 for contingency h in IEEJ WEST10 was not converged,
also CASE1 for the same contingency has much iteration
compared with other contingencies. In this case, we found that
some diagonal elements of SE part (see Fig. 1) became larger
than other parts.
Fig. 2 illustrates the dynamic response at contingency h for
IEEJ WEST10 (T
max
=2sec). From this figure, a constraint for
generator angle with respect to COI was active at the terminal-
point of simulation period. Since only the elements of SE part
near the terminal-point of simulation period became large, we
thought that this was the cause of numerical problems. In order
to improve the numerical problem, we have extended the
simulation period to avoid active COI constraint at terminal
period.

Improvement in numerical problem at critical case
As mentioned above, the contingency h of WEST10 became a
critical case for COI constraints. In this case, since some
diagonal elements of SE part will become large, we will use it
for detecting the occurrence of numerical difficulties. After
detecting the case, the proposed method will automatically
extend the simulation period.
In order to explain the effectiveness of the proposed method,
we have applied TSCOPF to longer simulation period (CASE3
and 4), which was automatically extended.

CASE3 : T
max
=3sec
t=0.01sec fixed
CASE4 : T
max
=3sec
t=0.01sec 0~0.5sec
t=0.05sec 0.5~3sec



Bulk Power System Dynamics and Control - VI, August 22-27, 2004, Cortina dAmpezzo, Italy 133

Table2 Simulation result for CASE1 and 2 (west10)
cost
Itera
tion
Time(s)
CASE2/
CASE1
case1 225623.7 34 210.11
case2 225608.0 23 63.34
case1 225538.2 29 181.35
case2 225521.7 23 63.94
case1 225568.1 33 216.45
case2 225568.7 27 73.89
case1 225406.2 42 260.49
case2 225685.7 30 80.58
case1 222775.2 45 265.89
case2 222775.0 42 109.29
case1 222775.1 21 122.27
case2 222775.0 20 50.63
case1 222775.0 22 126.64
case2 222775.1 20 52.57
case1 223825.0 57 320.96
case2
h
not converged
g 0.4151
c 0.3414
f 0.4141
d 0.3093
e 0.4110
b 0.3526
Fault
a 0.3015


Table3 Simulation result for CASE1 and 2 (west30)
cost
Itera
tion
Time(m)
CASE2/
CASE1
case1 179124.8 43 112.41
case2 179124.8 27 34.62
case1 179139.7 42 104.43
case2 179143.2 26 32.68
case1 178636.2 33 81.25
case2 178636.2 24 30.04
case1 179374.6 45 110.84
case2 179375.3 32 40.07
case1 178636.2 42 103.69
case2 178636.2 29 36.12
case1 178636.2 36 90.28
case2 178636.2 25 31.44
f 0.3482
d 0.3615
e 0.3483
c 0.3697
Fault
a 0.3080
b 0.3129

-100
-50
0
50
100
0.0 1.0 2.0 3.0
time(sec)
g
e
n
e
r
a
t
o
r

a
n
g
l
e
w
i
t
h

r
e
s
p
e
c
t

t
o
C
O
I

(
d
e
g
r
e
e
)

Fig2. Dynamic response at contingency h for IEEJ WEST10
(T
max
=2sec).

Test results of CASE3 and 4
Fig. 3 illustrates the dynamic response at contingency h for
IEEJ WEST10 (T
max
=3sec). Table 4 and 5 shows the simulation
result for CASE3 and 4 respectively.
As shown in Fig. 3, the COI constraint was not active at
terminal point.
From Table 4, the convergence characteristics were improved
for both CASE3 and 4. The number of iteration was reduced
50% for CASE3. The speed up rate of CASE1/CASE4 for
contingency h is 320.96/83.48=384%.
With our two proposed method, we can solve the TSCOPF
quite faster in numerically difficulty case.

-100
-50
0
50
100
0.0 1.0 2.0 3.0
time(sec)
g
e
n
e
r
a
t
o
r

a
n
g
l
e
w
i
t
h

r
e
s
p
e
c
t

t
o
C
O
I

(
d
e
g
r
e
e
)
4

Fig3. Dynamic response at contingency h for IEEJ WEST10
(T
max
=3sec).

Table4 Simulation result for CASE3 and 4 (west10)
cost
Itera
tion
time(s)
CASE4/
CASE3
case3 223650.6 31 292.76
case4 223688.8 24 83.48
0.2851
Fault
h



Conclusions
In this paper, we first formulated the TSCOPF problem. Then,
the effectiveness of the TSCOPF formulation was
demonstrated on the IEEJ WEST10 and IEEJ WEST30 model
system.
We proposed two improvements in TSCOPF. In order to reduce
the number of equations of time steps in TSCOPF, we
introduced variable time-step width method. Next, we
improved numerical problems at critical case, where the
convergence of TSCOPF will be difficult at the terminal-point
of simulation period.
Simulation results reveal that the proposed method has fast
execution time and good convergence.

References
(1) Yue Yuan, Junji Kubokawa, Hiroshi Sasaki A Study of
Transient Stability Constrained Optimal Power Flow with
Multi-Contingency ,Transactions of The Institute of Electrical
Engineers of Japan, Vol.122-B, No7, pp.798-804, July 2002
(2)D.Gan,R.J.Thomas,R.D.Zimmerman:Stability-Constrained
Optimal Power Flow, IEEE Trans.on PWRS, 15, No2, 535-
540(2000)
(3)E. De Tuglie, M. Dicorato,M. La Scala, P. Scarpellini, A
Static Optimization Approach to Assess Dynamic Available
Transfer Capability, IEEE Trans. On Power Systems, Vol. 15,
No. 3, August 2000
(4)L.Chen, Y.Tada, H.Okamoto, R.tanabe A.Ono,:Optimal
Operation Solutions of Power Systems with Transient Stability
Constraints, IEEE Trans. On Circuits and Systems I:
Fundamental Theory and Applications,48, No3, 327-339(2001)
(5) H.W.Dommel, N.Sato: Fast Transient Stability Solutions,
IEEE Trans. On PAS, 91, N0.4, 1643-1650(1972)
(6)IEEJ technical committee: Standard Models of Power
Systems, IEEJ Technical Report, No.754, November, 1999
http://www.iee.or.jp/pes/model/english/index.html

134 Bulk Power System Dynamics and Control - VI, August 22-27, 2004, Cortina dAmpezzo, Italy

You might also like