90

Int. J. Power and Energy Conversion, Vol. 1, No. 1, 2009

Multiobjective optimal VAR dispatch considering control variable adjustment costs M.A. Abido
Electrical Engineering Department, King Fahd University of Petroleum and Minerals, KFUPM P.O. Box # 183, Dhahran, 31261, Saudi Arabia Fax: +966-3-860-3535 E-mail: [email protected]
Abstract: In this paper, strength Pareto evolutionary algorithm (SPEA) is applied to solve the multiobjective optimal reactive power (VAR) dispatch problem. The optimal VAR dispatch problem is formulated as a non-linear constrained multiobjective optimisation problem where the real power loss and the control variable adjustment costs are to be optimised simultaneously. The proposed approach handles the problem as a true multiobjective optimisation problem. A hierarchical clustering algorithm is imposed to provide the decision-maker with a representative and manageable Pareto-optimal set. Moreover, fuzzy set theory is employed to extract the best compromise solution over the trade-off curve. The results demonstrate the capabilities of the proposed approach to generate true and well-distributed Pareto-optimal solutions of the multiobjective VAR dispatch problem in one single run. In addition, the effectiveness of the proposed approach and its potential to solve the multiobjective VAR dispatch problem are confirmed. Keywords: optimal VAR dispatch; evolutionary algorithms; multiobjective optimisation. Reference to this paper should be made as follows: Abido, M.A. (2009) Multiobjective optimal VAR dispatch considering control variable adjustment costs, Int. J. Power and Energy Conversion, Vol. 1, No. 1, pp.90104 Biographical notes: M.A. Abido is currently a Professor at King Fahd University of Petroleum & Minerals, Dhahran, Saudi Arabia. His research interests are power system planning, operation, and optimisation techniques applied to power systems. He is the recipient of KFUPM Excellence in Research Award 2002 and 2007, the First Prize Paper Award of the Industrial Automation and Control Committee of the IEEE Industry Applications Society, 2003, Abdel-Hamid Shoman Prize for Young Arab Researchers in Engineering Sciences, 2005, and the Best Applied Research Award of 15th GCC-CIGRE Conference, Abu-Dhabi, UAE, 2006. He has published more than 120 papers in reputable journals and international conferences.

Introduction

In the past two decades, the problem of reactive power control for improving the economy and security of power system operation has received much attention. Generally, the system performance can be enhanced by reallocating reactive power generations in
Copyright 2009 Inderscience Enterprises Ltd.

Multiobjective optimal VAR dispatch

91

the system. This can be achieved by adjusting transformer taps, generator voltages and switchable VAR sources. In addition, the system losses can be minimised via the redistribution of reactive power in the system. Several methods to solve the optimal reactive power dispatch problem have been proposed in the literature. Generally, there are three approaches to solve this complex problem. The first approach employs non-linear programming technique (Mansour and Abdel-Rahman, 1984). However, non-linear programming based procedures have many drawbacks, such as insecure convergence properties, long execution time and algorithmic complexity. The second approach uses sensitivity analysis and gradient-based optimisation algorithms by linearising the objective function and the system constraints around an operating point (Mamandur and Chenoweth, 1981). Obviously, this problem is a highly complex optimisation problem and may have several local minima. In this case, the gradient-based methods are susceptible to be trapped in local minima and the solution obtained will not be the optimal one. Moreover, calculation of the sensitivity factors is a time consuming process and inefficient for large-scale system applications. The third approach utilises the heuristic methods to search for the optimal solution in the problem space (Iba, 1994; Lai and Ma, 1997; Hsiao and Chiang, 2000; Gan et al., 1996). It has been theoretically proved that these methods converge to the optimal solution with probability one provided that certain conditions are satisfied. These heuristic methods have been applied to solve the optimal VAR dispatch problem with impressive success. Recently, the VAR dispatch problem is formulated as a multiobjective optimisation problem. However, the problem is not treated as a true multiobjective problem (Grudinin, 1998; Hsaio et al., 1994). It was converted to a single objective problem by linear combination of different objectives as a weighted sum (Grudinin, 1998). Unfortunately, this requires multiple runs as many times as the number of desired Pareto-optimal solutions. To avoid this difficulty, the -constraint method for multiobjective optimisation was presented in Hsaio et al. (1994). The most obvious weaknesses of this approach are that it is time-consuming and tends to find weakly non-dominated solutions. On the contrary, the studies on evolutionary algorithms, over the past few years, have shown that these methods can be efficiently used to eliminate most of the difficulties of classical methods (Fonseca and Fleming, 1995; Coello, 1999; Zitzler and Thiele, 1998). Since they use a population of solutions in their search, multiple Pareto-optimal solutions can, in principle, be found in one single run. Recently, some preliminary successful results of optimal VAR dispatch have been reported (Abido and Bakhashwain, 2005) where minimisation of voltage differences has been considered as an objective. This could be detrimental since it utilises scarce reactive resources for no stability or operational benefit. In this paper, the strength Pareto evolutionary algorithm (SPEA) based approach is proposed for solving the multiobjective VAR dispatch optimisation problem. The problem is formulated as a non-linear constrained multiobjective optimisation problem where the real power loss and control variable adjustment costs objectives are optimised simultaneously. A hierarchical clustering technique is implemented to provide the power system operator with a representative and manageable Pareto-optimal set. Moreover, a fuzzy-based mechanism is employed to extract the best compromise solution over the trade-off curve. The effectiveness and potential of the proposed approach to solve the multiobjective VAR dispatch problem are demonstrated.

92

M.A. Abido

Problem formulation

The optimal VAR dispatch problem is to optimise the steady state performance of a power system in terms of one or more objective functions while satisfying several equality and inequality constraints. Generally, the problem can be formulated as follows.

2.1 Objective functions

Real power loss (PL): This objective is to minimise the real power loss in transmission lines that can be expressed as:
J1 = PL =

k =1

nl

g k Vi 2 + V j2 2ViV j cos( i j )

(1)

where nl is the number of transmission lines; gk is the conductance of the kth line; and Vi i and V j j are the voltages at the end buses i and j of the kth line respectively. Control variable adjustment: The proposed objective is to minimise the control variable adjustments that can be expressed as:
J2 =

k =1

NC

wk

0 uk uk max min uk uk

(2)

0 where NC is the number of control variables; uk and uk are the current and initial settings of the kth control variable respectively; and wk is a weighting factor to reflect max min the relative cost of the kth control variable. uk and uk are the maximum and minimum limits of the kth control variable.

2.2 Problem constraints

Equality constraints: These constraints represent typical load flow equations as follows:
PGi PDi Vi

V [G
j j =1 NB j j =1

NB

ij

cos( i j ) + Bij sin( i j )] = 0

(3)

QGi QDi Vi

V [G

ij

sin( i j ) Bij cos( i j )] = 0

(4)

where i = 1,NB: NB is the number of buses; PG and QG are the generator real and reactive power respectively; PD and QD are the load real and reactive power respectively; and Gij and Bij are the transfer conductance and susceptance between bus i and bus j respectively.

Multiobjective optimal VAR dispatch

93

Inequality constraints: These constraints represent the system operating constraints as follows:

Generation constraints: Generator voltages VG and reactive power outputs QG are restricted by their lower and upper limits as follows:
min max VGi VGi VGi , i = 1,..., NG

(5)
(6)

min max QGi QGi QGi , i = 1,..., NG

where NG is number of generators. Transformer constraints: Transformer tap T settings are bounded as follows:
Ti min Ti Ti max , i = 1,..., NT

(7)

where NT is the number of transformers. Switchable VAR sources constraints: Switchable VAR compensations QC are restricted by their limits as follows:
min max Qci Qci Qci , i = 1,..., NC

(8)

where NC is the number of switchable VAR sources. Load bus voltage: These include the constraints of voltages at load buses VL as follows:
min VLi VLi VLmax , i = 1,..., NL i

(9)

2.3 Formulation
Aggregating the objectives and constraints, the problem can be mathematically formulated as a non-linear constrained multiobjective optimisation problem as follows:
Minimise [ J1, J 2 ]

(10)

Subject to:
g (x,u) = 0 h(x,u) 0

(11) (12)

where: x is the vector of dependent variables consisting of load bus voltages VL, generator reactive power outputs QG and transmission line loadings Sl. Hence, x can be expressed as: xT = [VL1 ... VLNL , QG1 ... QGNG ] u (13)

is the vector of control variables consisting of generator voltages VG, transformer tap settings T and shunt compensations Qc. Hence, u can be expressed as:
uT = [VG1 ... VGNG , T1... TNT , Qc1...QcNC ]

(14)

94 g h

M.A. Abido is the equality constraints. is the inequality constraints.

The proposed approach

3.1 Overview
Recently, the studies on evolutionary algorithms have shown that these algorithms can be efficiently used to eliminate most of the difficulties of classical methods that can be summarised as: an algorithm has to be applied many times to find multiple Pareto-optimal solutions most algorithms demand some knowledge about the problem being solved some algorithms are sensitive to the shape of the Pareto-optimal front the spread of Pareto-optimal solutions depends on the efficiency of the single objective optimiser.

In general, the goal of a multiobjective optimisation algorithm is not only to guide the search towards the Pareto-optimal front, but also to maintain the population diversity in the set of the non-dominated solutions.

3.2 Strength Pareto evolutionary algorithm

The proposed SPEA based approach has the following features (Zitzler and Thiele, 1998): it stores externally those individuals that represent a non-dominated front among all solutions considered so far it uses the concept of Pareto dominance in order to assign scalar fitness values to individuals it performs clustering to reduce the number of individuals externally stored without destroying the characteristics of the trade-off front.

Generally, the algorithm can be described in the following steps: Step 1 (initialisation) Generate an initial population and create an empty external Pareto-optimal set. Step 2 (external set updating) The external Pareto-optimal set is updated as follows: a search the population for the non-dominated individuals and copy them to the external Pareto set b search the external Pareto set for the non-dominated individuals and remove all dominated solutions from the set c if the number of the individuals externally stored in the Pareto set exceeds the prespecified maximum size, reduce the set by clustering.

Multiobjective optimal VAR dispatch

95

Step 3 (fitness assignment) Calculate the fitness values of individuals in both external Pareto set and the population as follows: a Assign a real value s [0,1) called strength for each individual in the Pareto optimal set. The strength of an individual is proportional to the number of individuals covered by it. The strength of a Pareto solution is at the same time its fitness. b The fitness of each individual in the population is the sum of the strengths of all external Pareto solutions by which it is covered. In order to guarantee that Pareto solutions are most likely to be produced, a small positive number is added to the resulting value. Step 4 (selection) Combine the population and the external set individuals. Select two individuals at random and compare their fitness. Select the better one and copy it to the mating pool. Step 5 (crossover and mutation) Perform the crossover and mutation operations according to their probabilities to generate the new population. Step 6 (termination) Check for stopping criteria. If any one is satisfied then stop else copy new population to old population and go to Step 2. In this study, the search will be stopped if the generation counter exceeds its maximum number.

3.3 Reducing Pareto set by clustering

In some problems, the Pareto optimal set can be extremely large. In this case, reducing the set of non-dominated solutions without destroying the characteristics of the trade-off front is desirable from the decision-makers point of view. An average linkage based hierarchical clustering algorithm (Morse, 1980) is employed to reduce the Pareto set to manageable size. It works iteratively by joining the adjacent clusters until the required number of groups is obtained.

3.4 Best compromise solution

Upon having the Pareto-optimal set of non-dominated solution, the proposed approach presents one solution to the decision-maker as the best compromise solution. Due to imprecise nature of the decision-makers judgment, the ith objective function Fi is represented by a membership function i defined as (Dhillon et al., 1993):
1 F max F i i i = max Fi Fi min 0 Fi Fi min Fi min < Fi < Fi max Fi Fi max

(15)

where Fi min and Fi max are the minimum and maximum value of the ith objective function among all non-dominated solutions, respectively.

96

M.A. Abido

Implementation of the proposed approach

4.1 The computational flow

Figure 1

Proposed SPEA based approach

Start

Intialize population and Pareto-optimal set, set gen=0

Current Population

Determination of Nondominated Solutions Extend Pareto Set Yes Size<Max. Size

No

Extend Population

Reduce Pareto Set by Clustering

Selection

No

Crossover and Mutation

Is feasible population filled ?

Yes No Select the best compromise solution using fuzzy set theory Is gen < max gen? Yes Updated Pareto Set

Stop

gen = gen + 1

In this study, the basic SPEA has been developed in order to make it suitable for solving real-world non-linear constrained optimisation problems. The following modifications have been incorporated in the basic algorithm:

Multiobjective optimal VAR dispatch a

97

A procedure is imposed to check the feasibility of the initial population individuals and the generated children through GA operations. This ensures the feasibility of Pareto-optimal non-dominated solutions. A procedure for updating the Pareto-optimal set is developed. In every generation, the non-dominated solutions in the first front are combined with the existing Pareto-optimal set. The augmented set is processed to extract its non-dominated solutions that represent the updated Pareto-optimal set. A fuzzy-based mechanism is employed to extract the best compromise solution over the trade-off curve and assist the decision-maker to adjust the VAR sources efficiently.

The computational flow of the proposed SPEA based approach is shown in Figure 1.

4.2 Settings of the proposed approach

The techniques used in this study were developed and implemented using FORTRAN language. On all optimisation runs, the population size and the maximum number of generations were selected as 200 and 500 respectively. The maximum size of the Pareto-optimal set was set as 30 solutions. If the number of the non-dominated Pareto optimal solutions exceeds this bound, the clustering technique is called. Crossover and mutation probabilities were selected as 0.9 and 0.01 respectively in all optimisation runs.

Results and discussions

In this study, the proposed approach was tested on the standard IEEE 30-bus 6-generator test system in order to investigate its effectiveness. The single-line diagram of the IEEE test system is shown in Figure 2 and the detailed data are given in Alsac and Stott (1974). The system has six generators at buses 1, 2, 5, 8, 11 and 13 and four transformers with off-nominal tap ratio in lines 69, 610, 412 and 2728. The lower voltage magnitude limits at all buses are 0.95 pu and the upper limits are 1.1 pu for generator buses 2, 5, 8, 11 and 13 and 1.05 pu for the remaining buses including the reference bus 1. The lower and upper limits of the transformer tapings are 0.9 pu and 1.1 pu respectively. The initial settings of the control variables and the initial values of objective functions are given in Table 1. The number of control variables in this case is 10. At first, the J1 and J2 objectives are optimised individually and the best results are given in Table 1. The convergence of both objectives with individual optimisation is shown in Figures 3 and 4. The problem was handled as a multiobjective optimisation problem where both objectives were optimised simultaneously with the proposed approach. The diversity of the Pareto optimal set over the trade-off surface is shown in Figure 5. It is worth mentioning that the Pareto optimal set has 30 non-dominated solutions. Out of them, two non-dominated solutions that represent the best J1 and J2 are given in Table 2. The proposed approach has been implemented to IEEE 118-bus system with 186 lines. This system contains 54 generators and nine transformers. Therefore, the total number of control variables in this case is 63. The control variable limits are given in Table 3. At first, the J1 and J2 objectives are optimised individually and the objective

98

M.A. Abido

convergence are shown in Figures 6 and 7. The problem was handled as a multiobjective optimisation problem where both objectives were optimised simultaneously with the proposed approach. The diversity of the Pareto optimal set over the trade-off surface is shown in Figure 8.
Figure 2

Single-line diagram of IEEE 30-bus system

29 28

27 30 26 25

23

24

15

18 17 21

19 20

14

16

22

13

12 11

10 9

6 7

Multiobjective optimal VAR dispatch

Table 1

99

The best solutions for individual optimisisation Initial* Best J1 1.050 1.041 1.018 1.017 1.084 1.079 1.002 0.951 0.990 0.940 0.0512 1.5186 Best J2 1.0500 1.0450 1.0100 1.0100 1.0500 1.0500 0.9780 0.9690 0.9320 0.9680 0.0538 0.0000

VG1 VG2 VG5 VG8 VG11 VG13 T69 T610 T412 T2728 J1 J2

1.050 1.045 1.010 1.010 1.050 1.050 0.978 0.969 0.932 0.968 0.0538 0.0000 Source: *Alsac and Stott (1974)

Figure 3

Convergence of power loss minimisation with IEEE 30-bus system

0.054

0.053

Objective Function P L

0.052

0.051

0.050 0 200 400 600 800 1000

Generations

100
Figure 4

M.A. Abido
Convergence of control variable adjustment minimisation with IEEE 30-bus system
1.6

1.2

Objective Function C U

0.8

0.4

0.0 0 200 400

Generations

600

800

1000

Figure 5

Pareto-optimal front of the proposed approach for the 30-bus test system
1.0

0.8

0.6

J2
0.4 0.2 0.0 0.0510 0.0515 0.0520 0.0525 0.0530 0.0535

J1

Multiobjective optimal VAR dispatch

Table 2

101

The results of the proposed approach Best J1 Best J2 1.050 1.044 1.010 1.013 1.050 1.051 0.978 0.969 0.932 0.968 0.0533 0.0388 Best compromise 1.050 1.044 1.020 1.022 1.051 1.062 0.978 0.969 0.960 0.958 0.0519 0.4280

VG1 VG2 VG5 VG8 VG11 VG13 T69 T610 T412 T2728 J1 J2
Table 3

1.050 1.044 1.021 1.023 1.064 1.083 0.978 0.969 1.002 0.949 0.0512 0.9225

Control variable limits of IEEE 118-bus system Min 0.95 0.90 Max 1.1 1.1 Number 54 9

Control variable Generator voltages Transformer taps

Figure 6

Convergence of power loss minimisation with IEEE 118-bus system

1.32

Objective Function J 1

1.28

1.24

1.20 0 100 200 300 400 500

Generations

102
Figure 7

M.A. Abido
Convergence of control variable adjustment minimisation with IEEE 118-bus system
16.0

12.0

Objective Function J 2

8.0

4.0

0.0 0 200 400

Generations

600

800

1000

Figure 8

Pareto-optimal front of the proposed approach for the 118-bus test system
1.0

0.8

J2

0.6

0.4

0.2 0.124 0.126 0.128 0.130 0.132 0.134

J1

Multiobjective optimal VAR dispatch

103

Conclusions

In this paper, multiobjective VAR dispatch optimisation problem considering control variable adjustment costs is solved using the SPEA. The problem has been formulated as multiobjective optimisation problem with real power loss and control variable adjustment costs objectives. A hierarchical clustering technique is implemented to provide the operator with a representative and manageable Pareto-optimal set without destroying the characteristics of the trade-off front. Moreover, a fuzzy-based mechanism is employed to extract the best compromise solution over the trade-off curve. The results show that the proposed approach is efficient for solving multiobjective VAR dispatch problem where multiple Pareto optimal solutions can be found in one simulation run. In addition, the non-dominated solutions obtained are well distributed and have satisfactory diversity characteristics.

Acknowledgements
The author acknowledges the support of King Fahd University of Petroleum and Minerals.

References
Abido, M.A. and Bakhashwain, J.M. (2005) Optimal VAR dispatch using a multiobjective evolutionary algorithm, International Journal of Electrical Power & Energy Systems, January, Vol. 27, No. 1, pp.1320. Alsac, O. and Stott, B. (1974) Optimal load flow with steady state security, IEEE Trans. on Power Apparatus and Systems, Vol. PAS-93, pp.745751. Coello, C.A.C. (1999) A comprehensive survey of evolutionary-based multiobjective optimization techniques, Knowledge and Information Systems, Vol. 1, No. 3, pp.269308. Dhillon, J.S., Parti, S.C. and Kothari, D.P. (1993) Stochastic economic emission load dispatch, Electric Power Systems Research, Vol. 26, pp.186197. Fonseca, C.M. and Fleming, P.J. (1995) An overview of evolutionary algorithms in multiobjective optimization, Evolutionary Computation, Vol. 3, No. 1, pp.116. Gan, D., Qu, Z. and Cai, H. (1996) Large-scale VAR optimization and planning by Tabu search, Electric Power Systems Research, Vol. 39, pp.195204. Grudinin, N. (1998) Reactive power optimization using successive quadratic programming method, IEEE Trans. on Power Systems, Vol. 13, No. 4, pp.12191225. Hsaio, Y.T., Chaing, H.D., Liu, C.C. and Chen, Y.L. (1994) A computer package for optimal multi-objective VAR planning in large scale power systems, IEEE Trans. on Power Systems, Vol. 9, No. 2, pp.668676. Hsiao, Y.T. and Chiang, H.D. (2000) Applying network window schema and a simulated annealing technique to optimal VAR planning in large-scale power systems, Electric Power Systems Research, Vol. 22, pp.18. Iba, K. (1994) Reactive power optimization by genetic algorithm, IEEE Trans. on PWRS, Vol. 9, No. 2, pp.685692. Lai, L.L. and Ma, J.T. (1997) Application of evolutionary programming to reactive power planning-comparison with non-linear programming approach, IEEE Trans. on Power Systems, Vol. 12, No. 1, pp.198206.

104

M.A. Abido

Mamandur, K.R.C. and Chenoweth, R.D. (1981) Optimal control of reactive power flow for improvements in voltage profiles and for real power loss minimization, IEEE Trans. on Power Apparatus and Systems, Vol. PAS-100, No. 7, pp.31853193. Mansour, M.O. and Abdel-Rahman, T.M. (1984) Non-linear VAR optimization using decomposition and coordination, IEEE Trans. on Power Apparatus and Systems, Vol. PAS-103, No. 2, pp.246255. Morse, N. (1980) Reducing the size of non-dominated set: pruning by clustering, Computers and Operations Research, Vol. 7, Nos. 12, pp.5566. Zitzler, E. and Thiele, L. (1998) An Evolutionary Algorithm for Multiobjective Optimization: The Strength Pareto Approach, Swiss Federal Institute of Technology, TIK-Report.