Professional Documents
Culture Documents
4, DECEMBER 2015
AbstractReal-time economic dispatch (RTED) is performed fps (ps ) Solar PV generation PDF.
every 515 min with the static snapshot forecast data. During the fl (l) Load PDF (normal distribution).
period between two consecutive schedules, generators participate Gij , Bij Transfer conductance and susceptance between
in managing power imbalance, based on participation factors
(PFs) from previous economic dispatch (ED). In modern power bus i and bus j, respectively.
systems with considerable renewable energy sources that have n Number of buses in the system.
high variability, this conventional approach may not adequately NL Number of loads/demands.
accommodate the economic implication of the said variability. NG Number of conventional thermal generators.
This paper proposes the evaluation of best-fit PFs by taking NW Number of wind generators/farms.
into account the minute-to-minute variability of solar, wind, and
load demand, for a scheduling period. Since best-fit PFs are NS Number of solar PV plants.
evaluated only once, i.e., at the start of scheduling interval, the PGi Scheduled power from ith conventional thermal
dimensionality of optimization problem remains the same as that generator (MW) in interval T .
of conventional approach. The proposed approach is suggested for T 1
PGi Power output of ith conventional thermal gener-
sequential and dynamic variants. Results for two test systems have ator (MW) in previous interval (T 1).
been obtained to verify the benefit of the proposed approach.
PDi Forecasted real power demand in interval T .
Index TermsAutomatic generation control (AGC), economic PWj Power output from jth wind farm (MW).
dispatch (ED), electricity markets, renewable energy sources, un- PS,k Power output from kth solar PV plant (MW).
certainty, variability.
pr,j Rated wind power from jth wind generator.
max
PS,k Maximum power output from kth solar PV plant.
N OMENCLATURE
PDl,av Available power demand from lth load. This
a i , bi , c i Cost coefficients of ith thermal generator. is a random variable, with a range of PDl min
Cp,Dl Penalty cost function for lth load. PDl,av PDlmax
.
Cp,Sk Penalty cost function for kth solar photovoltaic PSk,av Available solar power from the kth solar power
(PV) plant. generator. This is a random variable, with a range
Cp,Wj Penalty cost function for not using all available of 0 PSk,av PSr,k .
power from the jth wind power generator. PWj,av Available wind power from the jth wind power
Cr,Dl Reserve cost function for lth load. generator. This is a random variable, with a range
Cr,Sk Reserve cost function for kth solar PV plant. of 0 PWj,av pr,j .
Cr,Wj Reserve cost function relating to uncertainty of p WEG power output.
wind power. This is effectively a penalty asso- l Uncertain load.
ciated with overestimation of the available wind ps Solar PV power output.
power. QDi Forecasted reactive power demand in inter-
dj Direct cost coefficient of jth wind farm. val T .
up down
ek Direct cost coefficient of kth solar PV plant. RGi , RGi Ramp up and ramp down limits of conventional
fp (p) Wind energy generator (WEG)/wind power thermal generators (MW/hr).
probability density function (PDF). t Subinterval (1 min) index.
T Interval (10 min) index.
VDk Load bus voltage magnitude.
Vi , Vj Voltage magnitudes at bus i and bus j.
Manuscript received November 10, 2013; revised April 6, 2014; accepted i , j Voltage angles at bus i and bus j.
May 11, 2014. Date of publication June 16, 2014; date of current version
November 20, 2015.
S. Surender Reddy is with the Department of Railroad and Electrical
Engineering, Woosong University, Daejeon 300-718, Korea (e-mail: salkuti. I. I NTRODUCTION
surenderreddy@gmail.com).
P. R. Bijwe and A. R. Abhyankar are with the Department of Electrical
Engineering, Indian Institute of Technology Delhi, New Delhi 110 016, India
(e-mail: prbijwe@ee.iitd.ac.in; abhyankar@ee.iitd.ac.in).
Digital Object Identifier 10.1109/JSYST.2014.2325967
I N the past, search for new energy resources was primarily
dictated by the fast depletion and rising cost of the fossil
fuels [1]. Nuclear energy was considered to be a very good
1932-8184 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
SURENDER REDDY et al.: RTED CONSIDERING POWER GENERATION VARIABILITY AND UNCERTAINTY 1441
option. However, some accidents in the last two decades in change significantly within the short amount of time. Intro-
nuclear reactors and the increasing awareness of the alarming ducing dynamic dispatch can better manage the ramp capa-
rate of global warming has led to a massive push for the large- bility of dispatchable resources for the volatility of renewable
scale integration of wind and solar renewable generation in the generations.
grid. The actual cost of this renewable generation is still quite In [10], a real-time SCED algorithm is designed to carry
high, and governments are providing subsidies to make their out real-time dispatch and the expost real-time pricing in com-
absorption economically viable. Due to the highly uncertain pliance with Midwest Independent System Operators energy
and variable nature of these resources and the subsidies, these and ancillary service market business rules. Compared with a
generations are generally considered to be nonschedulable, conventional static SCED, look-ahead dispatch leverages the
barring a few exceptions. near-term forecasts of variable generation, for an overall more
Wind and solar photovoltaic (PV) power plant technolo- cost-effective utilization of generation assets [11], [12]. Strictly
gies are developing rapidly, and receive increasing attention speaking, optimum RTED should consider all the operational
throughout the world. Most of the renewable energy sources costs involved, as accurately as possible, without making the
such as wind and solar power have a maximum generation limit optimization unduly complex, and computationally inefficient
that changes with time (variability), and this limit is not known from the implementation point of view. With the present state
with perfect accuracy (uncertainty), unlike the conventional en- of art, in real time, energy, reserve, and regulation markets
ergy sources [2]. It is well known that wind power is uncertain are cleared simultaneously or sequentially, typically at 5-m
as well as variable. Solar power is less uncertain, but has higher intervals, depending on the individual market structures [8].
variability compared to the wind power. Uncertainty is a serious This real-time dispatch is again sequential in some markets and
issue in planning for longer time frames. However, in real- dynamic in some others. New York ISO (NYISO) performs a
time economic dispatch (RTED), the uncertainty range is much rolling window-type dynamic dispatch, every 5 min, for the
smaller, as the forecast is more accurate as compared with that next 60-min operation [13]. In [14], a high-speed dynamic
in a longer term [3]. North American Electricity Reliability economic load dispatch is proposed by utilizing the short-term
Council (NERC) studies have shown that PV installations can load forecast, which is suitable for power systems with large
change output by 70%, in a time frame of 210 min, many renewable energy penetration suffering from large disturbances
times per day. whose fluctuations might cause infeasibility and whose predic-
The presence of any intermittent source of power increases tion error requires continuous change in operation schedule.
balancing requirement and associated costs [4]. Because of However, the approach proposed in [14] is suitable for verti-
wind and solar variability, there has been widespread interest in cally integrated systems, and the issues of its applications in
quantifying the increase in ancillary services required to inte- the energy market environment have not been studied.
grate wind and solar over various timescales [5]. Various types In the recent wind and solar PV generation integration studies
and details of ancillary services are defined in [6]. Regulation [15][17], the simulations have been fixed to one time resolu-
ancillary is the service required to follow moment-to-moment tion. This makes it difficult to analyze the effects of variability
fluctuations in system load and variable power generation [7]. within the single time-frame. To accommodate the variability
The requirement of regulation service is not locational, and and uncertainty in the scheduling programs, the system opera-
the suppliers are responsible for both regulation up and down tors carry operating reserves to cover energy imbalances. The
services in the whole market [8]. A metric to quantify the effects of solar PV power variability and forecast uncertainty
subhourly variability cost of individual wind plants and show its on electric power grid operation in the Arizona Public Service
use in valuing reductions in wind power variability is developed system is investigated in [18]. Both variability and uncertainty
in [9]. A methodology that allows for studying the impacts of can create challenges for grid operators, when balancing elec-
variability and uncertainty at multiple operational timescales is tricity generation with demand, while obeying power system
developed in [2]. constraints and maintaining the lowest possible costs. Since
Traditionally, real-time security-constrained economic dis- it is often difficult to predict the variations of both load and
patch (SCED) provided by the Independent System Operators renewable generator output accurately, Rosehart [19] proposes
(ISOs)/Regional Transmission Organizations (RTOs) solves en- an optimal power flow algorithm to make optimized results not
ergy dispatch and/or ancillary service clearing for one target only have a high probability to achieve minimized generation
interval. Real-time SCED inputs include demand forecast, net cost, but also robust to the uncertain operating states. In [20], a
scheduled interchange, and forecast of intermittent resource stochastic mixed integer linear optimization scheduling model
outputs, such as wind and solar, for the target interval. However, treating wind power forecasts and load forecasts as uncertain
under single step/interval dispatch, limited ramp capability may parameters is used to study the operational impacts of high
not be effectively utilized to prepare for significant changes wind penetrations on the power system. The model includes a
on demand or intermittent resource outputs in the near future. scenario-generation methodology, treatment of reserves, and a
Thus, ISOs/RTOs are looking into the dynamic/integrated dis- rolling stochastic unit commitment and dispatch driven by up-
patch. With dynamic dispatch, RTOs can use multiple intervals dated forecasts. A dynamic scheduling of operating reserves in
of demand and wind forecast to solve for multiple-interval an electricity market with wind power is proposed in [21]. The
optimized real-time SCED. The coordinated decision making dynamic reserves are integrated into a two-settlement market-
over these multiple intervals makes it a dynamic dispatch. clearing mechanism, where energy and reserves are dispatched
The power outputs from those renewable energy resources can and priced simultaneously.
1442 IEEE SYSTEMS JOURNAL, VOL. 9, NO. 4, DECEMBER 2015
To the best of our knowledge, all the approaches for RTED (T = 1, 2, . . . , Nint ). Again, this 515-min scheduling inter-
consider frozen static snapshots of the power system at (for val T is subdivided into Nsub of 1-min subintervals (t =
example) 5-min intervals. The variable generation cost between 1, 2, . . . , Nsub ). The subinterval duration may be changed as
two consecutive scheduling intervals is ignored in this opti- per forecast availability in a given system.
mization. The motivation in this paper is to accommodate the
variable cost of generation at suitable discrete subintervals (e.g., A. Conventional Sequential RTED
1 min), for which the load and nonschedulable renewable power
generation forecast data are available. The conventional practice has been to perform RTED sep-
During the RTED scheduling interval (e.g., 5 min), the arately/sequentially for each interval T . In this approach, an
regulating generators share the power imbalance as per their optimization is performed for a given snapshot, at the beginning
participation factors (PFs) evaluated at the base-point genera- of a scheduling interval of 515 min. From this optimized solu-
tions. This paper proposes a method that considers variability tion, a set of constant generation PFs [22] are evaluated for use
and uncertainty cost of renewable energy sources between two in automatic generation control (AGC) during the scheduling
consecutive scheduling intervals. In the proposed approach, the interval T . As the power imbalance occurs due to load and
evaluation of these best-fit PFs is based on the total cost renewable power generation uncertainty, and variability, the
of energy over the entire RTED interval, consisting of 1-min generators adjust their generation as per PFs already evaluated,
discrete subintervals. Both sequential and dynamic dispatch in order to restore power balance. The objective is to minimize
models are considered. All the relevant constraints, including the cost of all generations, while supplying the load for a given
generation rate constraint (GRC), are considered in both the snapshot. The mathematical model for conventional sequential
models. The highlight of the proposed approach is that the opti- RTED is presented as follows.
mization is done only once at the start of scheduling interval, Minimize the cost in each scheduling interval T
and the best-fit PFs are evaluated from this optimization.
NG
Therefore, the dimensionality of optimization problem, and CGi (PGi ) (1)
the implementation, remains the same as that of conventional i=1
approach.
subjected to the equality and inequality constraints given next.
II. P ROBLEM F ORMULATION B. Equality and Inequality Constraints
The market practice considered in this paper is such that the 1) Nodal Power Balance Constraints:
wind and solar PV powers are treated to be nonschedulable. In
this market practice, the rule/policy is to absorb all wind and
n
solar PV power into the system, and pay a tariff established by PGi PDi = Vi Vj (Gij cos ij + Bij sin ij ) (2)
j=1
the regulator, which is different from the market-clearing price.
This rule is normally followed in systems trying to increase n
QGi QDi = Vi Vj (Gij sin ij Bij cos ij ) (3)
the share of renewable power in the generation portfolio. Wind j=1
farms and solar plants come into the ISO market as price takers,
and do not directly affect the locational marginal prices. where i = 1, 2, . . . , n; ij = i j . Yij = Gij + jBij is the
In order to bring out the difference between the existing and (i, j) entry of the nodal admittance matrix.
the proposed RTED approaches, we first discuss the former. 2) Generation Limits in Each Interval T: The generator
First, the simpler sequential dispatch model will be considered, operation constraint after including the ramp rate limit of
and then, the dynamic dispatch model will be introduced. For generators can be described as
simplicity in understanding, all cases are first discussed with min T 1 max T 1 up
only variability of loads and then renewable energy sources. max PGi , PGi RGi down
PGi min PGi , PGi +RGi .
One case is finally discussed, where both variability and un- (4)
certainty effects are considered. The maximum penetration of wind power into the system is
The assumptions in the mathematical models are organized given by
as follows.
1) The wind and solar generations are nonschedulable. PW,j PR,j . (5)
Hence, the system operator will have to compulsorily use
whatever generation is available from these sources. The maximum penetration of solar PV into the power system is
2) System load and renewable generation forecasts are as- given by
sumed to be available at every 1-min subinterval, between PS,k PS,k
max
. (6)
the two successive stages of economic dispatch (ED).
3) Generators from the previous stage of ED are assumed to 3) Security Constraints: These include the limits on the
be known, in order to check GRC constraint satisfaction load bus voltage magnitudes and line flow limits. That is
at this stage.
In the mathematical models, the scheduling period of 1 h
min
VDk VDk VDk
max
, k = 1, 2, . . . , ND (7)
(60 min) is divided into Nint of 515-min scheduling intervals |Lij | Lij
max
(8)
SURENDER REDDY et al.: RTED CONSIDERING POWER GENERATION VARIABILITY AND UNCERTAINTY 1443
In the generation evolution from one subinterval to other, The desired output for each thermal generator in subinterval 1
GRCs are also considered, in addition to the equality and is given by
inequality constraints of the previous problem. In view of this, PGi,1 = PGi,0 + PGi,1 . (19)
the generation limits in subinterval t are given by
min Similarly, the desired output for each thermal generator in
max PGi , PGi,t1 RGi,t
down
subinterval 2 is given by
up
PGi,t min PGimax
, PGi,t1 + RGi,t . (10) PGi,2 = PGi,0 + PGi,2 (20)
where
As explained earlier, we intend to simplify the multistage
optimization problem in (9) to a simpler single optimization PGi,2 = PFi [PSolar,2 + PWind,2 + PLoad,2 ]. (21)
problem using best-fit PFs, which remain constant over the
scheduling interval T . It may be noted that the usual operat- This process will be repeated for all subintervals t in the
ing practice of having a fixed set of PFs over the interval is also scheduling interval T . It may be noted that PFs are already
applicable in this proposed approach. The amount of generation defined in terms of base-case generations.
change required (PGi,t ) in each subinterval t is worked out From the aforementioned discussion, it can be seen that
using these PFs (PFi ) and the amount of generation changes the objective function consists of generation costs in all the
due to solar, wind, and load demand variabilities (PSolar,t , subintervals. The generations in each subinterval have the same
PWind,t , and PLoad,t ), respectively, in each subinterval t, set of PFs. Hence, the objective function can be expressed in
1444 IEEE SYSTEMS JOURNAL, VOL. 9, NO. 4, DECEMBER 2015
terms of only the base-point generations. Since, the PFs depend although the generations vary minute to minute, the PFs on
on the optimization involving all subinterval generations, these which these depend, change only once during an interval.
can be considered to be best-fit PFs. Hence, it is expected to provide more economical benefit com-
Although the terms in the objective function are more, the pared with the conventional one.
optimization variables in this approach are same as that in
the conventional approach. This is because in the proposed
approach, the optimization is performed only once at the begin- F. Proposed Sequential RTED Considering Variability
ning of interval T , similar to the conventional approach. The and Uncertainty
generations in each subinterval t (1 min) are obtained using In the real time, the forecast uncertainty is much small;
best-fit PFs. These best-fit PFs are evaluated only once at hence, the variability plays a more prominent role. However,
the beginning of each interval. Hence, the computational effort there is still some forecast uncertainty at the real time, since dis-
in this optimization is almost same as that in the conventional patch signals must be sent out to the generators ahead of time.
approach. Frequent dispatch intervals (typically 515 min) reduces the
This proposed sequential RTED is repeated in every schedul- uncertainty. Here, it is assumed that the system has conventional
ing interval T (e.g., 10 min) over the scheduling period (e.g., thermal generators, wind farms, and solar PV plants. Because
60 min). of the uncertainty of the wind and solar energy and load demand
available at any given time, factors for overestimation and
D. Conventional Dynamic RTED underestimation of available wind and solar energy and load
demand must be included in the model [23]. Therefore, the
Dynamic dispatch has normally been done in daily hydro objective function is formulated as follows.
thermal scheduling or unit commitment. Because of the highly Minimize
variable renewable generation, it is being felt that there is a need
to have dynamic dispatch for even shorter durations like 1 h.
N sub
NG
This is important not just from the economics point of view CGi (PGi )
but also for having smooth transition of schedules every few t=1 i=1
minutes.
N sub
NW
In the hourly (60 min) dynamic dispatch, the scheduling + [Cr,Wj (PWj PWj,av )+Cp,Wj (PWj,av PWj )]
period is divided into Nint of 515-min intervals. The opti- t=1 j=1
mization variables are now generations in all these intervals, to
N sub
NS
minimize the total energy cost over the entire scheduling period. + [Cr,Sk (PSk PSk,av )+Cp,Sk (PSk,av PSk )]
The mathematical model is written as follows: t=1 k=1
Minimize
N sub
NL
N int
NG
T + [Cr,Dl (PDl PDl,av )+Cp,Dl (PDl,av PDl )] . (24)
CGi PGi (22) t=1 l=1
T =1 i=1
The terms in the aforementioned expression are explained as
subjected to equality and inequality constraints in all intervals follows:
over the scheduling period. The first term in the objective function (24) is the fuel cost of
conventional thermal generators, and is given by
E. Proposed Dynamic RTED With Variable Energy Costs
2
CGi (PGi ) = ai + bi PGi + ci PGi . (25)
The difference in this dynamic dispatch and the conventional
one is the same as that in the sequential one, viz., incorporation The second term is the reserve requirement cost, which repre-
of minute-to-minute generation variation cost. As explained sents the cost due to the available wind power being less than
earlier, the PFs for each interval (e.g., 10 min) are evaluated as the scheduled wind power. This cost function helps to deter-
independent variables. These best-fit PFs remain fixed in the mine the deficit power it might produce, from the distribution
respective intervals. The mathematical model for this problem function. This reserve cost function is given by
is written as follows.
Minimize Cr,Wj [PWj PWj,av ] = kr,j [PWj PWj,av ]
N
int N sub
NG
T
PWj
CGi PGi,t (23)
= kr,j (PWj p)fp (p)dp (26)
T =1 t=1 i=1
0
subjected to equality and inequality constraints in all the subin-
tervals and intervals over the scheduling period. where kr,j is the reserve cost coefficient for the jth wind
It is important to note that the dimensionality of this problem generator.
is same as that in the conventional dynamic dispatch, de- The third term is the penalty cost; this is obtained from the
spite additional summation over the intervals. This is because, concept of underestimation of wind power. This cost function
SURENDER REDDY et al.: RTED CONSIDERING POWER GENERATION VARIABILITY AND UNCERTAINTY 1445
helps to determine the excess power it might produce, from the III. W IND AND S OLAR E NERGY S YSTEMS
scheduled value. That is
A. Wind Energy System
Cp,Wj [PWj,av PWj ] = kp,j [PWj,av PWj ] The wind power derived will follow the stochastic nature as
compared to the wind speed [24], [25]. For each wind power
Prj
unit, the power output for a given wind speed input is given by
= kp,j (p PWj )fp (p)dp (27) [26], [27]
PWj
0, for vw,t < vi and vw,t > vo
vw,t vi
where kp,j is the penalty cost coefficient for the jth wind p = pr vr vi , for vi vw,t vr (32)
generator. pr , for vr vw,t vo
The fourth term in the objective function is the reserve
cost for solar PV power generation. This is obtained from the where vw,t is the forecasted wind speed (m/s) at t, vi is the
concept of overestimation of solar PV power. That is cut-in wind speed, vr is the rated wind speed, and vo is the cut-
out wind speed. pr is the rated power output for wind power
Cr,Sk [PSk PSk,av ] = kr,k [PSk PSk,av ] generation.
PSk
= kr,k (PSk ps )fps (ps )dps (28) B. Solar Energy System
min
PSk The solar irradiation to energy conversion function of the PV
generator or power output from PV cell is given by [28]
where kr,k is the reserve cost coefficient for the kth solar PV
power. PSr G2
, for 0 < G < Rc
The fifth term represents the penalty cost for solar PV power, PS (G) = Gstd Rc (33)
PSr G , for G > Rc
which is obtained from the concept of underestimation of solar Gstd
PV power. That is
where it is noted that PV cell temperature is neglected and
Cp,Sk [PSk,av PSk ] = kp,k [PSk,av PSk ] G solar irradiation forecast in W/m2 ;
Gstd solar irradiation in the standard environment set as
max
PSk
1000 W/m2 ;
= kp,k (ps PSk )fps (ps )dps (29) Rc a certain irradiation point set as 150 W/m2 ;
PSk
PSr equivalent rated power output of the PV generator.
The solar PV active power generation can either be controlled
where kp,k is the penalty cost coefficient for the kth solar PV by the power tracking control scheme or to be charged into
power. batteries. Therefore, the maximum penetration of PV to system
The sixth term in the objective function is the reserve cost for is given by
load demand. This is obtained from the concept of overestima-
tion of load demand. That is PS,k PS,k
max
(34)
where kr,k is the reserve cost coefficient for the kth load IV. R ESULTS AND D ISCUSSION
demand.
The seventh term represents the penalty cost for load de- We will now use the objectives described in Section II on
mand, which is obtained from the concept of underestimation modified IEEE 30- and 118-bus test systems to examine the
of load demand. That is effectiveness of the proposed optimum scheduling approach for
the windsolarthermal power system. These test systems are
Cp,Dk [PDk,av PDk ] = kp,k [PDk,av PDk ] used to find the impact of variability with a high penetration of
max
renewable power generation and load demand.
PDk
The prescribed time interval for the real-time dispatch for
= kp,k (l PDk )fl (l)dl (31) energy bids are every 515 min, and optimizes over a 60-min
PDk period. In our study, we have considered real-time dispatch for
every 10 min (each interval time T is 10 min), and optimizing
where kp,k is the penalty cost coefficient for the kth load energy for 60 min (the total scheduling period is 60 min). This
demand. 60-min scheduling period is only for dynamic approach and
1446 IEEE SYSTEMS JOURNAL, VOL. 9, NO. 4, DECEMBER 2015
TABLE I
O PTIMUM G ENERATION S CHEDULES AND O BJECTIVE F UNCTION VALUES FOR M ODEL 1 W ITH IEEE 30-B US S YSTEM
TABLE II
O PTIMUM G ENERATION S CHEDULES AND O BJECTIVE F UNCTION VALUES FOR M ODEL 2 W ITH IEEE 30-B US S YSTEM
TABLE III
O PTIMUM G ENERATION S CHEDULES AND O BJECTIVE F UNCTION VALUES FOR M ODEL 3 W ITH IEEE 30-B US S YSTEM
schedules obtained from first interval as base points. The 2000.1082 $/hr and 34.0175 $/hr, respectively. The second
optimum total (generation and variability) cost obtained in interval uses the generation schedules obtained from the first
interval 2 is 2045.4599 $/hr. This sequential optimization pro- interval as base points, and optimizes (9). The optimum total
cess will be repeated throughout the scheduling period (60 min cost (generation and variability cost) obtained in this interval is
or six intervals). The optimum total cost (sum of all six 1981.3660 $/hr. This sequential optimization process consider-
intervals cost) obtained over the 60-min period using this ing variability cost will be repeated throughout the scheduling
conventional sequential approach (Model 1) is 13195.7165 $/hr. period (60 min, i.e., six intervals). The optimum total cost (sum
2) Model 2 (Pricing Energy and Variability Costs Using of all six intervals cost) obtained over the 60-min period using
Proposed Sequential RTED Approach): Table II shows the the proposed sequential approach is 12 919.2331 $/hr, which is
optimum best-fit generation schedules and energy costs in 2.1% less than the cost obtained from Model 1.
each interval using proposed sequential RTED with variable 3) Model 3 (Pricing Energy Using Conventional Dynamic
energy cost (Model 2). In each interval of 10 min, the best-fit RTED Approach): Table III shows the optimum generation
PFs are evaluated considering minute-to-minute variability of schedules and energy and variability costs in each inter-
solar, wind, and demand. The variability cost for each minute is val for the conventional dynamic/integrated RTED approach
calculated using these best-fit PFs. Here, in each interval, (9) (Model 3). In this conventional dynamic approach, the objective
is optimized sequentially over the scheduling period (60 min). function (22) is optimized over the scheduling period (i.e.,
In the first interval (10 min), (9) is posed as objective 60 min). This objective (sum of all intervals cost) optimizes
function, and the obtained generation and variability costs are the cost of all intervals simultaneously. For the comparison
1448 IEEE SYSTEMS JOURNAL, VOL. 9, NO. 4, DECEMBER 2015
TABLE IV
O PTIMUM G ENERATION S CHEDULES AND O BJECTIVE F UNCTION VALUES FOR M ODEL 4 W ITH IEEE 30-B US S YSTEM
TABLE V
O PTIMUM G ENERATION S CHEDULES AND O BJECTIVE F UNCTION VALUES FOR P ROPOSED S EQUENTIAL
RTED C ONSIDERING VARIABILITY AND U NCERTAINTY W ITH IEEE 30-B US S YSTEM
purpose, the variability cost is calculated after getting the which are evaluated taking into account the expected variations
generation schedules using constant PFs in each interval. The in loads and renewable power generations over the entire in-
optimum generation cost (sum of all six intervals) obtained terval, is proposed. In this sense, the new PFs can be termed
is 12 808.0363 $/hr, and the variability cost is 186.5388 $/hr. as best-fit PFs. Here, other than the method of calculation of
Therefore, the total cost obtained with this conventional dy- PFs, the regulation/AGC modalities remains the same as that in
namic RTED approach (Model 3) is 12994.5751 $/hr. the conventional one.
4) Model 4 (Pricing Energy and Variability Costs Using Table V shows the optimum generation schedules and energy
Proposed Dynamic RTED Approach): Table IV shows the op- costs in each interval using the proposed sequential RTED
timum best-fit generation schedule and energy and variability approach considering variability and uncertainty. In this case,
costs in each interval for the proposed dynamic/integrated the objective function (24) is optimized in each interval T .
RTED approach. In this proposed dynamic approach, (23) is The optimum variability and uncertainty costs obtained with
optimized over the scheduling period. This objective (sum of all this approach are 185.7141 $/hr and 50.3901 $/hr, respectively.
intervals cost) optimizes the energy and variability costs of all From this, it can be seen that, in the real time, uncertainty
intervals (six intervals or 60 min) simultaneously. The optimum has lesser impact on results compared to the variability. The
generation and variability costs obtained in the first interval are total cost (sum of all six intervals) obtained in this case is
2025.0597 $/hr and 33.1087 $/hr, respectively, which are less 13 110.8728 $/hr, which is 1.46% higher compared to the cost
than the costs obtained from the conventional dynamic RTED obtained from Model 2, as given in Table II.
approach. The optimum total (sum of all six intervals) cost
obtained in this approach is 12720.8953 $/hr, which is 2.1%
B. Results on IEEE 118-Bus System
less than the cost obtained from Model 3.
Hence, in the results obtained from the aforementioned mod- The modified IEEE 118-bus 186-branch test system [31]
els for IEEE 30-bus system, it can be seen that the optimum cost consists of 19 generators. For these studies, we have modified
obtained over the scheduling period using Model 4 is 3.6% less this system with wind power injected at buses 92 and 111 and
than the cost obtained from Model 1, 1.5% less than the cost solar PV power injected at buses 31, 46, and 87. The rated
obtained from Model 2, and 2.1% less than the cost obtained capacity for each wind farm is considered as 100 MW, and for
from Model 3. solar PV plant is 80 MW.
5) Simulation Results on Proposed Sequential RTED Con- 1) Model 1 (Pricing Energy Using Conventional Sequen-
sidering Variability and Uncertainty: As explained earlier, a tial RTED Approach): Table VI shows the objective function
methodology of finding out a different set of constant PFs, values in each interval over the scheduling period of 60 min
SURENDER REDDY et al.: RTED CONSIDERING POWER GENERATION VARIABILITY AND UNCERTAINTY 1449
TABLE VI
O BJECTIVE F UNCTION VALUES FOR M ODEL 1 W ITH IEEE 118-B US S YSTEM
TABLE VII
O BJECTIVE F UNCTION VALUES FOR M ODEL 2 W ITH IEEE 118-B US S YSTEM
TABLE VIII
O BJECTIVE F UNCTION VALUES FOR M ODEL 3 W ITH IEEE 118-B US S YSTEM
TABLE IX
O BJECTIVE F UNCTION VALUES FOR M ODEL 4 W ITH IEEE 118-B US S YSTEM
using the conventional sequential RTED approach. Here, each objective function values in each interval over the scheduling
interval is optimized to obtain the energy cost. For the com- period using proposed dynamic RTED with variable costs
parison with the proposed sequential RTED (Model 2), the (Model 4). In this approach, generations will vary minute to
variability cost is calculated using constant PFs obtained from minute based on the PFs. The optimum energy and variability
converged generation schedules in each interval. The optimum costs obtained with this approach are 1 031 804.7163 $/hr and
generation and variability costs obtained over the scheduling 14 746.2947 $/hr, respectively. Therefore, the total cost (sum
period is 1 095 775.2217 $/hr and 16 811.2405 $/hr, respec- of generation and variability costs) obtained in this model is
tively. Therefore, the total cost obtained in this model is 1 046 551.0110 $/hr. Hence, the total cost obtained in this model
1 112 586.4622 $/hr. is 4.46% less than the total cost obtained from Model 3.
2) Model 2 (Pricing Energy and Variability Costs Using From the case studies on modified IEEE 118-bus test sys-
Proposed Sequential RTED Approach): Table VII shows the tem, it can be observed that the optimum cost obtained from
objective function values in each interval using proposed se- Model 4 is 5.9% less than the cost obtained from Model 1, 1.4%
quential RTED with variable energy cost. The optimum energy less than from Model 2, and 4.46% less than from Model 3.
and variability costs obtained over the scheduling period (sun of The case studies show the benefit of proposed RTED approach
all intervals cost) is 1 046 264.4601 $/hr and 15 625.7689 $/hr, over conventional RTED on sequential and dynamic variants
respectively. Hence, the total cost (sum of generation and considering solar, wind, and load variabilities.
variability costs) is 1 061 530.2290 $/hr, which is 4.6% less than
the cost obtained from Model 1.
V. C ONCLUSION
3) Model 3 (Pricing Energy Using Conventional Dynamic
RTED Approach): Table VIII shows the objective function Conventional RTED ignores the impact of renewable gen-
values in each interval over the scheduling period using the eration variability during the scheduling interval. This paper
conventional dynamic RTED approach (Model 3). In this proposes a model, which incorporates the variability cost in
model, for the comparison purpose, the variability costs due the optimization using sequential and dynamic models. The
to solar, wind, and load variability in each interval are calcu- proposed approach uses best-fit PFs to find the optimal
lated using constant PFs obtained from the converged solution. schedules. Here, the variability cost is calculated in every
The obtained optimum energy and variability costs over the minute, using these PFs between two successive stages of ED
scheduling period using this approach are 1 079 318.3591 $/hr executions within the real-time dispatch. The dimensionality
and 16 175.8546 $/hr, respectively. Therefore, the total cost of this approach is same as that in the conventional approach.
obtained in this model is 1 095 494.2137 $/hr. Hence, the amount of effort in optimization does not increase.
4) Model 4 (Pricing Energy And Variability Costs Using Results on IEEE 30- and 118-bus test systems demonstrate the
Proposed Dynamic RTED Approach): Table IX shows the benefit of the proposed approach over the conventional ones.
1450 IEEE SYSTEMS JOURNAL, VOL. 9, NO. 4, DECEMBER 2015