Вы находитесь на странице: 1из 4

DRPT2008 6-9 April 2008 Nanjing China

Comparative Study of System-Wide


Reliability-Constrained Generation Expansion
Problem
Panida Jirutitijaroen, Member, IEEE, and Chanan Singh, Fellow , IEEE

y gi ( ) Generation at bus i in state


AbstractThis paper investigates system-wide reliability yli ( ) Load curtailment at bus i in state
constrained composite system generation expansion problem.
Using stochastic programming framework, the problem is i ( ) Voltage angle at bus i in state
formulated as a two-stage recourse model. Reliability indexes are
system expected unserved energy cost in the objective function II. INTRODUCTION
and system expected power loss in the constraint. The objective of
this study is to determine if a solution from system-wide cost
minimization problem can lead to equally improved reliability in
all individual buses. Expected unserved energy indexes of each
M ost of the optimization techniques proposed for
generation expansion problem [1-4] [6-7] [10] [14]
considers reliability in terms of the overall system. The optimal
bus before and after the optimal planning are calculated and solution yields system reliability maximization, which does not
analyzed. The comparison is done on a 24-bus IEEE-RTS. guarantee fair improvement of reliability to customers at each
bus. When a customer is charged equally the expansion cost,
Index TermsComposite-Reliability System, Power System
Optimization, Reliability, Stochastic Programming, Two-stage each may not receive the same reliability level. It is interesting
Recourse Model. to see if the system-wide reliability maximization can lead to
fair reliability enhancement to all customers. If not, the
1. NOMENCLATURE problem should be reformulated in order to take into account
the distribution of customer reliability at each bus.
A. Indices This paper analyzes the reliability distribution of each bus
I {1,2,,n} Set of system buses resulting from the optimal planning. The optimal planning is
i,j System bus performed using stochastic programming with sample-average
System state (scenario), approximation technique. The problem is formulated as a
State space (all possible scenarios) two-stage recourse model. The first stage variables are the
number of additional units at each bus where the second stage
B. Parameters
variables are network flows. The overall objective is to
cig Cost of an additional generation unit at bus i ($) minimize expansion cost in the first stage and at the same time
cil ( ) Cost of load loss coefficient at bus i in state ($/MW) to minimize operation and expected unserved energy in the
c ( )
o
Generation operation cost at bus i in state ($/MW) second stage. Reliability indexes used in this problem are
i

Additional generation capacity at bus i (MW) system expected unserved energy in the objective function and
M ig
system expected power loss in the constraint.
g i ( ) Capacity of generation at bus i in state (MW) The entire system state space is too large to enumerate and
tij ( ) Tie line capacity between bus i and j in state (MW) directly apply the L-shaped algorithm [13]. To overcome this
li ( ) Load at bus i in state (MW) dimensionality problem, a sampling technique is employed to
bij Tie-line susceptances between bus i and bus j reduce the number of system states. Using Monte Carlo
sampling, the objective function of the second stage, called
B Augmented node susceptance matrix
sample-average approximation (SAA) [11] of the actual
C. Decision variables expected value, is defined by these samples. This
xig Number of additional generators at bus i, integer approximation makes it possible to solve the problem with the
deterministic equivalent model. The solution from
sample-based approach naturally provides statistical inference
This work was supported by Power System Engineering Research Center of lower and upper bounds of the optimal objective value. The
and NSF Grant No. ECS0406794.
C. Singh is with the Department of Electrical and Computer Engineering,
optimal solution is found with sufficient sample size when
Texas A&M University, College Station, TX 77843 USA (e-mail: multiple SAA problems give the same solution.
singh@ece.tamu.edu). The paper is organized as follows. Problem formulation is
P. Jirutitijaroen is with the Department of Electrical and Computer introduced along with the solution technique. The comparative
Engineering, National University of Singapore, Singapore 117576 (e-mail:
elejp@nus.edu.sg ). study is implemented on a 24-bus IEEE-RTS [12]. Concluding

978-7-900714-13-8/08/ 2008DRPT 675


DRPT2008 6-9 April 2008 Nanjing China

remarks are given in the last section. E~ [yil ( )] (12)


where is the upper limit of expected power loss
III. PROBLEM FORMULATION With this reliability constraint in addition to a budget
The expansion problem is formulated with explicit constraint, the problem may be infeasible. Instead of directly
reliability consideration. Uncertainties in generation, imposing the reliability constraint, the objective function is
transmission line capacity, and load are included in the model. modified as follows.
Reliability indexes are system expected unserved energy cost, Min z = cig xig + E~ [ f ( x, ~ )] + P (E~ [ yli ( )] ) (13)
and system expected power loss. Since reliability is one of the iI

constituents, a reliability model needs to be incorporated into where P is a penalty if the expected power loss violates the
the problem formulation. This underlying reliability evaluation limit .
model needs a flow model for evaluating system states To simplify the problem, operation cost in the second stage
regarding their loss of load status. A commonly accepted objective function is neglected. It should be noted that the cost
approach in composite reliability evaluation is to use a DC load of load loss coefficient depends on system states. Mean
flow model. duration of each stage can be calculated by the reciprocal of
The objective is to minimize the expansion cost and equivalent transition rate from that state to others. State mean
maximizing system reliability. An optimization approach duration is presented in (14). Equivalent transition rate of all
based on mixed integer stochastic programming is proposed for components can be calculated using the recursive formula in
the solution of generation expansion planning problem in [2]. [15] when constructing probability distribution function.
The problem is formulated as a two-stage recourse model 1 (14)
D = m
where the first stage decision variables are the number of
l

generators to be installed. The second stage decision variables


g + + g + t + + t + l
iI
i
iI
i
i , jI
ij
i , jI
ij
k =1
k

i j i j
are the actual flows in the network. The failure probability of
where
additional generators can be taken into account by using their
D Mean duration of state (hours)
effective capacities [17].
g + Equivalent transition rate of generation in area i from a
A. A Reliability-Constrained Two-Stage Recourse Model i
capacity of state to higher capacity (per hour)
The first stage problem is shown in the following. g Equivalent transition rate of generation in area i from a
Min z = cig xig + E~ [ f (x, ~ )]
i
(3) capacity of state to lower capacity (per hour)
iI t + Equivalent transition rate of transmission line from area
c (4)
ij
g
s.t. i xig Budget i to area j from a capacity of state to higher capacity
iI
(per hour)
xig 0, integer (5) t Equivalent transition rate of transmission line from area
ij
where constraints (4) is a budget constraint. Constraint (5) is an i to area j from a capacity of state to lower capacity
integer requirement for the number of additional generators. (per hour)
The function in (3) is the second stage objective value of l Equivalent transition rate of area load from state to
k
minimizing operation cost and expected unserved energy cost other load states (per hour)
under a realization of . Using DC load flow model, the ml Total number of area load states
formulation of the second stage problem is presented in the Customer damage function used in this paper is taken from
following. Let
B= B
ij[ ]
, [9]. The function was estimated from electric utility cost survey
f (x , ) = Min [cil ( )y li ( ) + cio ( ) y gi ( )] (6) in the US. For residential loads, interruption cost in dollars per
iI kW-h can be described, as a function of outage duration, by
s.t. y gi ( ) g i ( ) + M ig xig ; i I (7) (15).
c l (D ) = e 0.2503+ 0.2211D 0.0098 D (15)
bij ( i ( ) j ( )) tij ( ); i, j I , i j
2

(8)
yli ( ) li ( ); i I (9) B. Solution Method
Bij j ( ) + y gi ( ) + yli ( ) = li ( ); i I
jJ
(10) The expected cost of load loss can be approximated by
means of sampling. This technique is called sampled-average
y gi ( ), yli ( ) 0 ; i I , i ( ) unrestricted (11)
approximation (SAA). Let 1 , 2 ,, N be N realizations of
where, constraints (7), (8), and (9) are maximum capacity flow
random vector for all uncertainties in the model, the expected
in the network under uncertainty in generation, tie line, and
cost of load loss can be replaced by
load respectively. Constraint (10) constitutes conservation of (16)
1 N
flow in network and (11) presents variable restrictions in the fN (x ) = f (x, k ) .
model. N k =1

In addition to expected unserved energy cost in the objective By the nature of sampling, a solution obtained from this
function, system expected power loss is included in the sample-based approach does not necessarily guarantee
constraint such that it remains equal to or less than a specified optimality in the original problem. The optimal sample-based
value, solutions, when obtained with different sample sets, rather
provide statistical inference of a confidence interval of the

676
DRPT2008 6-9 April 2008 Nanjing China

actual optimal solution [5] [8]. Details of this analysis can be TABLE I
GENERATION AND PEAK LOAD
found in [1-2].
The optimal solution can be extracted when a unique Bus Generation Peak Load
solution is obtained from solving several SAA problems with (MW) 10% higher (MW)
different samples of a given size, N. In theory, optimality 101 192 118.8
should be attained with sufficiently large N. However, it may 102 192 106.7
103 - 198
be possible that each sample yields different solutions for small 104 - 81.4
sample size. If an identical solution is found from solving SAA 105 - 78.1
problems with these samples, it can be concluded that 106 - 149.6
optimality is verified. 107 300 137.5
108 - 188.1
109 - 192.5
IV. TEST SYSTEM AND RESULTS 110 - 214.5
The test system is 24-bus IEEE-RTS. Generators and 113 591 291.5
114 - 213.4
transmission lines parameters can be found in [12]. In this 115 215 348.7
study, system load is grouped into 20 clusters using clustering 116 155 110
algorithm [16]. Generation and peak load data at each bus are 118 400 366.3
shown in Table I. Five buses are chosen as possible locations 119 - 199.1
120 - 140.8
for expansion planning. Unit capacity and cost of additional 121 400 -
units are shown in Table II. Budget is assumed to be 100 122 300 -
million dollars. With original load, the expected power loss () 123 660 -
is 0.07 and this is used as the specified limit in the optimization.
TABLE II
System load is increased by 10% to represent demand growth,
ADDITIONAL GENERATION PARAMETERS
which gives expected unserved energy before unit addition of
1200. All reliability indexes are found from Monte Carlo Bus Unit Capacity (MW) Cost ($m)
simulation with coefficient of variation of 0.05. Loss sharing 101 20 20
policy is implemented in this study. 102 20 20
107 100 100
115 12 12
122 50 50

The optimal solution, found in [2], is to install 5 units at bus


102 which yields system expected power loss of 0.08. Expected
unserved energy of each area before and after optimal planning
are shown in Table III.

TABLE III
EXPECTED UNSERVED ENERGY BEFORE AND AFTER OPTIMAL PLANNING
(MW-H PER YEAR)

Bus Before After


Optimal Planning Optimal Planning
103 41.50 17.91
104 47.23 27.76
107 13.59 5.86
108 2.50 0.31
109 1022.06 591.24
110 9.23 5.08
113 0.00 0.88
114 6.39 6.25
115 33.27 21.51
116 3.95 2.27
118 20.48 17.55
119 0.00 0.61
120 0.00 1.04

It can be seen from Table III that the system-wide reliability


maximization planning does not necessarily yield equally
distributed reliability level in each bus. However, it does tend
to improve the reliability at almost all buses, especially those
with high level of unserved energy. In some cases, bus 113,
119, and 120 sacrifices their reliability for the system since the
reliability level of these buses are worse after the optimal

677
DRPT2008 6-9 April 2008 Nanjing China

planning. The EUE of these buses is, however, very small to [2] P. Jirutitijaroen and C. Singh, Composite-System Generation Adequacy
Planning Using Stochastic Programming with Sample-Average
start with. EUE percentage reduction is shown in Table IV. Approximation, submitted to Proc. of the 2008 IEEE PES Power
Systems Computation Conference, Glasgow, Scotland.
TABLE IV [3] P. Jirutitijaroen and C. Singh, Multi-Area Generation Adequacy
EXPECTED UNSERVED ENERGY REDUCTION Planning Using Stochastic Programming, Proc. of the 2006 IEEE PES
Power Systems Conference and Exposition, Atlanta, Georgia, Oct. 2006
Bus EUE Reduction Percentage [4] P. Jirutitijaroen and C. Singh, Reliability and Cost Trade-Off in
(MW-H PER YEAR) EUE Reduction (%) Multi-Area Power System Generation Expansion Using Dynamic
103 23.58 56.82 Programming and Global Decomposition, IEEE Trans. Power Syst., vol.
104 19.47 41.22 21, no. 3, pp. 1432-1441, Aug. 2006
107 7.73 56.88 [5] J. T. Linderoth, A. Shapiro, and S. J. Wright, ``The Empirical Behavior of
108 2.19 87.60 Sampling Methods for Stochastic Programming'', Annals of Operations
Research, vol. 142, no. 1, pp. 215-241(27) , Feb. 2006.
109 430.81 42.15
[6] S. Kannan, S. M. R. Slochanal and N. P. Padhy, Application and
110 4.15 44.96
Comparison of Metaheuristic Techniques to Generation Expansion
113 -0.88 -
Planning Problem, IEEE Trans. Power Syst., vol. 20, no. 1, pp. 466-475,
114 0.14 2.19 Feb. 2005.
115 11.75 35.32 [7] N. S. Rau and F. Zeng, Adequacy and Responsibility of Locational
116 1.69 42.78 Generation and Transmission Optimization Procedure, IEEE trans. on
118 2.93 14.31 Power systems, vol. 19, no. 4, pp. 2093-2101, Nov. 2004.
119 -0.61 - [8] B. Verweij, S. Ahmed, A. J. Kleywegt, G. Nemhauser, and A. Shapiro,
120 -1.04 - The Sample Average Approximation Method Applied to Stochastic
Routing Problems: A Computational Study, Computational
Optimization and Applications, 24:289-333, 2003.
The percentage EUE reductions of each bus range from 2% [9] L. Lawton, M. Sullivan, K. V. Liere, A. Katz, and J. Eto, "A framework
to 87%. Most of the buses with large EUE reductions, for and review of customer outage costs: Integration and analysis of electric
example, bus 103-110 are in the neighboring area with the utility outage cost surveys" (Nov. 1, 2003). Lawrence Berkeley National
additional units in bus 102. Other distance buses from bus 102 Laboratory. Paper LBNL-54365.
http://repositories.cdlib.org/lbnl/LBNL-54365
seem to have smaller reliability improvement. The results [10] H. T. Firmo and L. F. L. Legey, Generation Expansion Planning: An
indicate that though the reliability improvement may not be Iterative Genetic Algorithm Approach, IEEE Trans. Power Syst., vol.
proportionately distributed across buses, but most of the buses 17, no. 3, pp. 901-906, Aug. 2002.
[11] W. K. Mak, D. P. Morton, and R. K. Wood, Monte Carlo bounding
do experience improvement, especially the most unreliable techniques for determining solution quality in stochastic programs,
ones experiencing a considerable gain. Operations Research Letters, 24:47-56, 1999.
[12] IEEE APM Subcommittee, The IEEE Reliability Test System-1996,
IEEE Trans. Power Syst., vol. 14, no. 3, pp. 1010-1020, Aug. 1999.
V. CONCLUSION [13] J. R. Birge and F. Louveaux, Introduction to Stochastic Programming,
A comparative study on customer reliability level at each 1st Edition, Duxbury Press, Belmont, CA, 1997.
[14] J. Zhu and M. Chow, A Review of Emerging Techniques on Generation
bus before and after optimal planning is conducted. The Expansion Planning, IEEE Trans. Power Syst., vol. 12, no. 4, pp.
planning problem uses stochastic programming approach with 1722-1728, Nov. 1997.
sample average approximation technique. The problem is [15] J. Mitra and C. Singh, Pruning and Simulation for Determination of
Frequency and Duration Indices of Composite Power Systems, IEEE
formulated as a two-stage recourse model with the objective to Trans. Power Syst., vol. 12, no. 1, pp. 471-479, Feb. 1997.
minimize expansion cost in the first stage and reliability cost in [16] A. Lago-Gonzales and C. Singh, The Extended Decomposition -
the second stage. Reliability is included in terms of expected Simulation Approach for Multi-Area Reliability Calculations, IEEE
Trans. Power Syst., vol. 5, no.3, pp. 864-869, Aug. 1990.
unserved energy in the objective function and expected power [17] L. L. Garver, Effective Load Carrying Capability of Generating Units,
loss in the constraint. The objective function of the problem is IEEE Trans. Power Appar. Syst., vol. PAS-85, no. 8, pp. 910-919, Aug.
approximated by the sample-average using Monte Carlo 1966.
sampling. Generation expansion planning is implemented to
the 24-bus IEEE-RTS.
Results show that system-wide reliability optimization may
not equally improve reliability level at each bus but most buses
experience improvement in reliability, especially those
suffering the most. It is likely that buses within close distance
to the additional generators benefit more from the optimal
planning. Perhaps that is the reason for the placement of the
additional generators in those locations. This information is
useful for the Independent System Operators to overlook the
electric market and design appropriate mechanism in order to
promote fair pricing for reliability.

VIREFERENCES
[1] P. Jirutitijaroen and C. Singh, Reliability Constrained Multi-Area
Adequacy Planning Using Stochastic Programming with
Sample-Average Approximations, IEEE Trans. Power Syst., to be
published.

678

Вам также может понравиться