You are on page 1of 13

Available online at www.sciencedirect.

com

Applied Soft Computing 8 (2008) 15491561 www.elsevier.com/locate/asoc

Differential evolution approach for optimal reactive power dispatch


M. Varadarajan *, K.S. Swarup
Department of Electrical Engineering, Indian Institute of Technology, Madras, India Received 6 June 2007; received in revised form 28 October 2007; accepted 4 December 2007 Available online 15 December 2007

Abstract Differential evolution based optimal reactive power dispatch for real power loss minimization in power system is presented in this paper. The proposed methodology determines control variable settings such as generator terminal voltages, tap positions and the number of shunts to be switched, for real power loss minimization in the transmission system. The problem is formulated as a mixed integer nonlinear optimization problem. A generic penalty function method, which does not require any penalty coefcient, is employed for constraint handling. The formulation also checks for the feasibility of the optimal control variable setting from a voltage security point of view by using a voltage collapse proximity indicator. The algorithm is tested on standard IEEE 14, IEEE 30, and IEEE 118-Bus test systems. To show the effectiveness of proposed method the results are compared with Particle Swarm Optimization and a conventional optimization technique Sequential Quadratic Programming. # 2007 Elsevier B.V. All rights reserved.
Keywords: Optimal power ow; Reactive power dispatch; Loss minimization; Differential evolution; Penalty function

1. Introduction Global optimization of non-continuous, non-linear functions arising from real world complex engineering problems, which may have large number of local minima and maxima, is quite challenging. A number of deterministic approaches based on branch and bound and real algebraic geometry are found to be successful in solving these problems to some extend. Of late, stochastic and heuristic optimization techniques, such as evolutionary algorithms (EA), have emerged as efcient tools for global optimization. It has been applied to a number of engineering problems in diverse elds and one such eld is power system optimization. The power system is a complex network used for generating and transmitting electric power. It is expected to operate with consumption of minimal resources giving maximum security and reliability. The optimal power ow (OPF) problem is an important tool to help the operator achieve these goals by providing the optimal settings of all controllable variables. The various objectives of OPF problem are (1) minimization of cost of generation; (2) minimization of transmission losses or optimal reactive power dispatch;
* Corresponding author. E-mail address: varadarajan.me@gmail.com (M. Varadarajan). 1568-4946/$ see front matter # 2007 Elsevier B.V. All rights reserved. doi:10.1016/j.asoc.2007.12.002

(3) minimization of shift in controls; (4) minimization of cost of VAr Investment; (5) maximization of social benet. Because of its signicant inuence on secure and economic operation of power systems, optimal reactive power dispatch has received an ever-increasing interest from electric utilities. In this paper the optimal reactive power dispatch is done, which is a sub-problem of the OPF problem, with an objective to reduce transmission line power losses. It is an effective method to improve voltage level, decrease network losses and maintain the power system running under normal conditions. All controllable variables, such as tap ratio of transformers, output of shunts, reactive power output of generators and static reactive power compensators, are determined which minimizes real power losses or other appropriate objective functions, satisfying a given set of physical and operational constraints. While transformer tap ratios and output of shunts have discrete values, reactive power output of generators, bus voltage magnitudes and angles have, on the other hand, continuous values. Hence the reactive power dispatch optimization is a combinatorial optimization problem has to be formulated as a mixed integer, nonlinear problem. A number of mathematical optimization techniques have been proposed in literature to solve the OPF problem. For decades, conventional optimization techniques such as linear

1550

M. Varadarajan, K.S. Swarup / Applied Soft Computing 8 (2008) 15491561

programming (LP), quadratic programming (QP), gradient method, Newton method, and Interior Point methods have been used for the solving optimal reactive power dispatch problem [14]. LP method requires that the objective function and constraints have linear relationship, which may lead to loss of accuracy. The gradient and Newton methods suffer from the difculty in handling inequality constraints. Conventional methods are not efcient in handling problems with discrete variables. The combinatorial search approaches, branch and bound and cutting plane algorithms, which are usually used to solve the mixed integer programming model, are nonpolynomial and all suffer from the curse of dimensionality making them unsuitable for large scale OPF problems. In recent years global optimization techniques such as genetic algorithms (GA), evolutionary programming (EP), evolutionary strategies (ES), and particle swarm optimization (PSO) have been proposed to solve the OPF problem. Ma and co-workers [5,6] used EP for optimal reactive power dispatch. Lee et al. [7] solved the reactive power dispatch and investment planning problem by using a simple genetic algorithm (SGA) combined with the successive linear programming method. The Benders cut are constructed during the SGA procedure to enhance the robustness and reliability of the algorithm. Yoshida et al. [8] proposed an algorithm for reactive power and voltage control considering voltage security assessment using PSO. Zhao et al. [9] proposed a solution to the reactive power dispatch problem with PSO using multi-agent systems. This paper investigates the applicability of differential evolution (DE) algorithm for reactive power dispatch to minimize real power loss in transmission network. DE is a simple, population based search algorithm, for global optimization. It has demonstrated its robustness and effectiveness in a variety of applications, such as neural network learning and innite impulse response (IIR) lter design [10,11]. DE differs from other EAs in the mutation and recombination phases. Unlike stochastic techniques such as GA and ES, where perturbation occurs in accordance with a random quantity, DE uses weighted differences between solution vectors to perturb the population. DE employs a greedy selection process with implicit elitist features. It has a minimum number of EA control parameters, which can be tuned effectively. The authors in [1214] used DE for Optimal Var Planning. In all the previous works reported in literature, inequality constraints were handled by use of a penalty function approach, i.e., the constraint violation is multiplied by a penalty coefcient or parameter and added to the objective function. Deb [15] proposed a penalty function method without penalty coefcients to overcome the difculty in choosing penalty coefcients for GA based constrained optimization problems. Although a penalty term is added to the objective function to penalize infeasible solutions, the method differs from the way the penalty term is dened in conventional methods and in earlier EA implementations. This penalty parameterless strategy is applicable only to population based approach because it requires the population to be divided into two sets: feasible and infeasible sets. The tness function depends on the

feasible and infeasible solutions. Since in a conventional optimization approach, there is only one member in each iteration, such a penalty parameterless scheme cannot be applied. In this paper, the penalty parameterless scheme is applied for reactive power optimization using DE. The method converges to the optimum solution, successfully meeting all equality and inequality constraints. The validity of the proposed method is tested on standard IEEE systems. Results obtained using PSO and a conventional optimization technique sequential quadratic programming (SQP) are also provided for comparing the performance of the proposed method. 2. Optimal power ow The optimal power ow (OPF) is a static, non-linear, and non-convex optimization problem, which determines a set of optimal variables from the network state, load data and system parameters. Optimal values are computed in order to achieve a certain goal such as generation cost or transmission line power loss minimization subjected to equality and inequality constraints. In general the OPF problem can be presented as min fx; u (2.1) (2.2) (2.3) xmax umax (2.4) (2.5)

s:t: gx; u 0 hx; u xmin umin x u 0

where, fx; u is the objective function that typically includes total generation cost (active power dispatch) or total losses in transmission system (reactive power dispatch). Generally, gx; u represents the loadow equations and hx; u represents transmission line limits and other security limits such as voltage security margin (VSM). The vector of dependent and control variables are denoted by x and u respectively. In general, the dependent vector includes bus voltage angles u, load bus voltage magnitudes V L , slack bus real power generation Pg;slack , and generator reactive power Qg. x u; V L ; Pg;slack ; Qg T (2.6)

The control variable vector consists of real power generation, Pg (except slack bus); generator terminal voltage, V g ; transformer tap ratio, t; and reactive power generation or absorption of shunt capacitor and reactors, Qsh . u Pg ; V g ; t; Qsh T (2.7)

Of the control variable mentioned in Eq. (2.7) Pg and V g are continuous variables, while tap ratio of the tap changing transformer and reactive power output of shunt devices, Qsh , are discrete variables. Loss minimization is usually required when cost minimization is the main goal, with generator active power generation as a control variable. When all control variables are utilized in a cost minimization, a subsequent loss minimization will not yield further improvements. Therefore,

M. Varadarajan, K.S. Swarup / Applied Soft Computing 8 (2008) 15491561

1551

in optimal reactive power dispatch problem, such as loss minimization, active power generation of all generators except slack generator is xed during the optimization procedure. 3. Optimal reactive power dispatch The objective function here is to minimize the active power loss (PLOSS ) in the transmission system. There are two basic approaches to loss minimization, namely the slack bus approach and the summation of losses on individual lines. Sometimes it is desirable to minimize losses in a specic area and hence, the second approach which is more generic, is used in this work. 3.1. Objective function Network losses, either for the whole network or for certain sections of network, are non-separable functions of dependent and independent variables. min PLOSS
Nl X k1

q P2 Q 2 l l VCPIi

Smax l

l 1; . . . ; N l i 1; . . . ; N B

(3.7) (3.8)

VCPIthreshold

The inequality constraints on control (independent) variable limits are given by


min Vg;i

V g;i Qsh;i tk

max Vg;i

i 1; . . . ; N PV i 1; . . . ; N sh

(3.9) (3.10) (3.11)

Qmin sh;i
min tk

Qmax sh;i

max tk

k 1; . . . ; N T

gk tk V i 2 V 2 2tk V i V j cos ui j j

(3.1)

where, gk is the conductance of branch k between buses i and j, N l the number of branches, tk tap ratio of transformer connected in branch k, V i is voltage magnitude at bus i, and ui j is the voltage angle difference between buses i and j. 3.2. Constraints The minimization of the above objective function is subjected to a number of equality and inequality constraints. The equality constraints are real and reactive power balance at each node i.e. load ow equations given by Pi V i
NB X V j Gi j cos ui j Bi j sin ui j 0 j1

where, N G , N PV , N sh and N T are the number of generators, PV buses, shunts and transformers respectively. Pl , Ql and Smax are l real, reactive and maximum apparent power ow in line l. VCPIi the voltage collapse proximity indicator at bus i, V g;i and V L;i are bus voltage magnitude at generator and load bus i, respectively, Qg;i the reactive power generation at bus i, Qsh;i the shunt reactive power at bus i, tk the tap ratio of transformer k min max and Pg;slack the real power generation at slack bus. VL;i , VL;i , min max min max min max min max min Vg;i , Vg;i , Pg;slack , Pg;slack , tk , tk , Qg;i , Qg;i , Qsh;i , and Qmax , are minimum and maximum limits of the corresponding sh;i variables, respectively. 4. Differential evolution Differential evolution is a simple population based, stochastic parallel search evolutionary algorithm for global optimization. DE is capable of handling non-differentiable, non-linear, and multi-modal objective functions. In DE, the population consists of real valued vectors with dimension D that equals the number of design parameters/control variables. The size of the population is adjusted by the parameter N P. The population of a DE algorithm is randomly initialized within the initial parameter bounds. The optimization process is conducted by means of three main operations: mutation, crossover and selection. In each generation, each individual of the current population becomes a target vector. For each target vector, the mutation operation produces a mutant vector, by adding the weighted difference between two randomly chosen vectors to a third vector. The crossover operation generates a new vector, called trial vector, by mixing the parameters of the mutant vector with those of the target vector. If the trial vector obtains a better tness value than the target vector, then trial vector replaces the target vector in the next generation. The evolutionary operators are described below. 4.1. Initialization In DE, a solution or an individual i, at generation G is a multidimensional vector xG xi;1 ; . . . ; xi;D . The population is i initialized by randomly generating individuals as xG xkmin rand0; 1 xkmax xkmin i;k k 2 1; D i 2 1; N P ; (4.1)

(3.2)

for i 1; . . . ; N B 1
NB X V j Gi j sin ui j Bi j cos ui j 0 j1

Qi V i

(3.3)

for i 1; . . . ; N PQ where, N B is the number of buses, N PQ the number of PQ buses, Gi j and Bi j are real and imaginary part of (i, j)th element of bus admittance matrix, Pi and Qi are net real and reactive power injection at bus i. The inequality constraints on security limits (dependent variables) are given by Pmin g;slack
min VL;i

Pg;slack V L;i Qg;i

Pmax g;slack i 1; . . . ; N PQ i 1; . . . ; N G

(3.4) (3.5) (3.6)

max VL;i

Qmin g;i

Qmax g;i

1552

M. Varadarajan, K.S. Swarup / Applied Soft Computing 8 (2008) 15491561

where, N P is the population size, D is the solutions dimension i.e. number of control variables. Each variable k in a solution vector i in the generation G is initialized within its boundaries xkmin and xkmax . 4.2. Mutation DE does not use a predened probability density function to generate perturbing uctuations. It relies upon the population itself to perturb the vector parameter. For every i 2 1; . . . ; N P the weighted difference of two randomly chosen population vectors, xr2 and xr3 , is added to another randomly selected population member, xr1 , to build a mutated vector vi. vi xG SxG xG r1 r2 r3 (4.2)

By using this selection procedure, all individuals of the next generation are as good as, or better than the individuals of the current population. 4.5. Stopping criteria The stopping criteria depends on the type of problem. The iterative procedure can be terminated when any of the following criteria is met, (i) an acceptable solution has been reached, (ii) a state with no further improvement in solution is reached, (iii) control variables has converged to a steady value or (iv) a predened maximum number of iterations have been completed. 5. Constraint handling The most common approach in the EA to handle constraints is to use penalties. The basic approach is to dene the tness values of an individual by extending the domain of the objective function. 5.1. Penalty function based on penalty coefcients In this method of constraint handling, in minimization problems, the tness function Fx is dened as the sum of the objective function f x and a penalty term which depends on the constraint violation hhxi. Fx f x
n X R j hh j xi2 j1

In Eq. (4.2), i; r 1 ; r 2 and r 3 are mutually different indices from the current generation. S is the user dened parameter, called step size, which is typically chosen from the range [0, 2]. If vi is found outside variable limit, it will then be xed to the violated upper or lower limit. 4.3. Crossover The next task after mutation is crossover, to increase the diversity of the perturbed parameter vectors. A trial vector ui is created incorporating the mutated vector vi and the target vector xi : ui uG1 i;k  vi;k xG i;k if randk;i CR or k I rand if randk;i > CR and k 6 I rand (4.3)

(5.1)

where randk;i 2 0; 1 and I rand is chosen randomly from the interval 1; . . . ; D once for each vector to ensure that at least one vector component originates from the mutated vector vi. Eq. (4.3) is applied for every vector component i 2 1; . . . ; N P , k 2 1; . . . ; D. CR is the DE control parameter, called the crossover rate, and is a user dened parameter within range [0, 1]. Trial parameter with randomly chosen index, I rand , is taken from mutant vector to ensure that the trial vector does not duplicate xi . 4.4. Selection DE determines survivors by pairwise comparison i.e. a form of tournament selection. If the trial vector ui has an equal or better objective function value than that of its target vector xi, it replaces the target vector in the next generation. Otherwise, target retains its place in the population for at least one more generation. By comparing each trial vector with the target vector from which it inherits parameters, DE more tightly integrates recombination and selection. All solutions in the population have the same chance of being selected as parents. xG1 i  uG1 i xG i if f uG1 i otherwise f xG i (4.4)

where hi gives the absolute value of the operand if the operand is negative and returns a zero if the operand is positive. The parameter R j is the penalty coefcient of the jth inequality constraint and it is user dened parameter. For reactive power dispatch optimization problem, equality constraints given by (3.2) and (3.3) are met by the load ow solution, while (3.9) (3.11) are enforced during the population coding and (3.8) is considered outside the optimization loop. Hence effectively, the inequality constraints to be handled here are (3.4)(3.7). In penalty function method, this is incorporated by modifying the objective function as given below. X 2 2 F f R1 Pg;slack Plim R2 V i Vilim g;slack
i 2 N PQ

X
i 2 NG

R3 Qgi

2 Qlim gi

X
i 2 Nl

R4 jSl j Smax 2 l

(5.2)

where, R1 , R2 , R3 , and R4 are penalty coefcients associated with real power generation at slack bus, voltage magnitude, reactive power generation, and apparent line ow limit violations respectively. Plim , Vilim , and Qlim can be expressed in g;slack gi general form as 8 max if xi > xmax < xi i xlim xmin if xi < xmin (5.3) i i i : xi otherwise

M. Varadarajan, K.S. Swarup / Applied Soft Computing 8 (2008) 15491561

1553

Since the order of magnitude violation is different for different constraints, it is difcult to nd a unique value for R1, R2 , R3 , and R4 . These can be xed only by trial and error method and problem dependent. 5.2. Effect of penalty coefcients Theoretically, the penalty should be kept just above the limit below which infeasible solutions are optimal. In most problems the exact location of the boundary between the feasible and infeasible regions is unknown and hence minimum penalty rule is not easy to implement in practice. To highlight the inuence of the choice of penalty coefcients on the objective function value, simulation studies were carried out on various test systems using DE algorithm with different values of penalty coefcients as explained above. The results obtained for IEEE 14-bus system [8] are listed in Table 1. The table shows the results for three different cases with same initial population. It can be observed that the PLOSS value is greatly dependent on the choice of penalty coefcients. 5.3. Penalty function based on feasibility In this scheme, employed in this paper, the composite tness function for any x is given as follows  f x if x is feasible Fx (5.4) f max CVx otherwise Here, f max is the objective function value of the worst feasible solution in the population. In situation where none of the solutions in a population are feasible, f max is dened. Hence, such situations are handled by articially inserting the base case solution into the population. CV(x) is the overall constraint violation of solution x. It is calculated as follows. CVx max 0; Pg;slack Pmax ; Pmin Pg;slack g;slack g;slack N PQ X max 0; V i Vimax ; Vimin V i
i1 NG X max 0; Qgi Qmax ; Qmin Qgi gi gi i1 Nl X max jSl j Smax l l1

violations alone. Hence, both the objective function value and constraint violation are not combined in any solution in the population. Thus there is no need to have any penalty coefcient R for this approach. The advantages of this scheme as compared to the usual penalty parameter based scheme are (i) The tedious process of choosing a suitable penalty coefcient R can be avoided, the inappropriate choice of which will affect the nal solution and (ii) there is no need to evaluate the objective function value for individuals with constraint violation, which reduces the computation time. The following criteria are enforced while selecting the individuals for the next generation. (1) Any feasible solution is preferred to any infeasible solution. (2) Among two feasible solutions, the one having better objective function value is preferred. (3) Among two infeasible solutions, the one having smaller constraint violation is preferred. 6. Differential evolution approach to optimal reactive power dispatch The control variables selected for reactive power dispatch problem are: the generator voltages, tap ratio of tap changing transformers and output of shunts. Among the control variables, the generator voltages are continuous, whereas the transformer tap ratios and the outputs of shunts are discrete. But tap ratio of transformers and output of shunts depend upon the tap position and the number of shunts switched. Hence, the generator voltage V g , tap position (integer), and the number of shunts to be switched (integer) are selected as control variables for optimization problem. 6.1. Treatment of control variables In its basic form, DE algorithm can handle only continuous variables. However, reactive power source installations and tap position of tap changing transformers are discrete variables in the reactive power dispatch problem. In this paper, DE has been extended to handle mixed integer variables, by the proper treatment of control variables as explained below. For integer variables the value is rounded off to the nearest integer value of the variable.  for continuous variables xi xi (6.1) b xi c for integer variables The b x c function gives the nearest integer less than or equal to x. A typical individual xi can be represented as
1 N sh xi Vg ; . . . ; Vg PV ; n1 ; . . . ; nN T ; n1 sh ; . . . ; nN sh t Q t Q

(5.5) All feasible solutions have zero constraint violation and all infeasible solutions are evaluated according to their constraint
Table 1 Effect of penalty coefcients on PLOSS Coefcients R1 R2 R3 R4 PLOSS (MW) Case 1 500 1000 100 100 13.42 Case 2 100 2000 200 100 13.29 Case 3 200 3000 300 100 13.31

(6.2)

where nt is the number of tap positions in a tap changing transformer and nQsh is the number of shunt reactive power devices available at a particular bus. Initial generator terminal voltages, which are continuous variables, are generated randomly between upper and lower limits of the voltage specication values. The value is then modied in the search procedure, within the specied limits.

1554

M. Varadarajan, K.S. Swarup / Applied Soft Computing 8 (2008) 15491561

Transformer tap positions, which are integer, are initially generated randomly between the minimum and maximum tap positions. The value is then modied in the search procedure among existing set of integer tap positions. Based on the tap position, the corresponding tap ratio is calculated as follows.
min tk tk nk Dtk

(6.3)

where, nk is the number of tap positions and Dtk is the step size. Using this tap ratio, the corresponding admittance of the transformer is determined for the load ow calculation. Reactive power compensation devices at a bus, which are again integer in nature, are initialized randomly from the integer set generated between 0 to the number of existing equipment at the bus. This value is also modied in the search procedure, always limiting it to be within the integer set of existing shunt devices. 6.2. Termination criteria

Step 3. Evaluate the tness of the individuals according to Eq. (5.4). Step 4. Perform mutation and crossover operation as described in Sections 4.2 and 4.3. Step 5. Select the individuals for the next generation as given in Section 4.4. Step 6. Store the best individual of the current generation. Step 7. Repeat Steps 25 till the termination criteria is met. Step 8. Select the control variable setting corresponding to the overall best individual. Step 9. Determine VCPI at each bus for the selected control variable setting and check whether it is less than threshold value. Step 10. If the solution is acceptable, output the best individual and its objective value. Otherwise, take the settings corresponding to the next best individual and repeat the Step 8. 7. Simulation results

The iterative procedure can be terminated when any of the following criteria is met, i.e., an acceptable solution has been reached, a state with no further improvement in solution is reached (stall generation), control variables has converged to a steady value or a predened number of iterations have been completed. In most of the cases, it is not easy to test whether the obtained solution is the most acceptable one. Also, the lack of further improvement in the solution or convergence of control variables need not necessarily translate to achievement of the global solution. A commonly used approach is to run the iterations to a xed maximum number of generations which is dependent on the problem under consideration. Usually the number of maximum generations is xed by a trial and error process. In this work, combination of maximum number of generations and stall generation limit is used as the termination criteria. 6.3. Algorithm DE is employed to nd the best control variable setting starting from a randomly generated initial population. At the end of each generation, the best individuals, based on the tness value, are stored. The VCPI at the bus k obtained from Eq. (A.1), will vary from zero to one, with zero indicating a voltage stable condition and one indicating a voltage collapse [16]. The VCPI value for the best individual is compared with the threshold value and if the value is less than the threshold value, it indicates a voltage secure condition. The threshold value is xed by conducting off-line study on the system for different operating conditions. Evaluation of the voltage security, independent of the OPF algorithm simplies the optimization procedure. The details of the proposed algorithm is as follows. Step 1. Generate an initial population randomly within the control variable bounds. Step 2. For each individual in the population, run power ow algorithm such as Newton Raphson method, to nd the operating points.

The proposed DE approach for optimal reactive power dispatch algorithm is tested on standard IEEE 14-bus [8], IEEE 30 [17], and IEEE 118 [18] bus test systems. Table 2 gives the details of the test systems. A comparative study with PSO, employing a constriction coefcient [1922], was done to verify the performance of the proposed algorithm. The DE and PSO algorithm was implemented using MATLAB15.3 running on Pentium IV PC. DE and PSO parameters used for the simulation are summarized in Table 3. Number of individuals in a population for each test system is decided by experimentation. To validate and compare the results obtained by DE algorithm, the dispatch problem is also solved by SQP technique, using Matlab Optimization Toolbox [23], assuming all the variables to be of continuous. The results of DE and PSO, which follow, are the best solutions over 30 independent trails. 7.1. IEEE 14-bus system The modied IEEE 14-bus system data and initial operating conditions of the system is given in [8]. For IEEE 14-bus system shown in Fig. 1, there are 14 buses, out of which 5 are generator buses. Bus 1 is the slack bus, 2, 3, 6 and 8 are taken as PV generator buses and the rest are PQ load buses. The network has 20 branches, 17 of which are transmission lines and 3 are tap
Table 2 Description of test systems IEEE 14 No. No. No. No. No. No. No. No. No. of of of of of of of of of buses N B generators N G transformers N T shunts N sh branches N l equality constraints inequality constraints control variables discrete variables 14 5 3 2 20 28 65 10 5 IEEE 30 30 6 4 2 41 60 125 12 6 IEEE 118 118 54 9 12 186 236 566 75 21

M. Varadarajan, K.S. Swarup / Applied Soft Computing 8 (2008) 15491561 Table 3 Simulation parameters DE Population size Max. no. of generations Step size (F) Crossover rate (CR) 30 200 0.6 0.8 PSO Population size Max. no. of generations C1 , C2 vmin vmax x 30 200 2.05 0.4 0.9 0.7298

1555

changing transformers. It is assumed that capacitor compensation is available at buses 9 and 14. Totally, there are nine control variables which consists of four PV generator voltages, three tap changing transformers with 20 discrete steps of 0.01 p.u. each and two shunt compensation capacitor banks with three discrete steps of 0.06 p.u. each. Fig. 2 gives the performance of the optimization technique in terms of PLOSS with DE and PSO for the best run out of 30 trials. It can be observed that PLOSS reduces over the evolutions and converge to a minimum value. From the base case value of 13.49 MW, the PLOSS was reduced to 13.239 MW with DE. The iterative procedure is terminated when there is no change in the result for 40 consecutive generations or when 200 generations are reached, whichever occurs rst. To understand and study how evolution is going on, information about average value, standard deviation and variance at each generation were observed. The Fig. 3 shows this information for the proposed algorithm. To verify the performance of DE, the results are compared with PSO and a conventional optimization technique SQP. Table 4 shows the minimum value of PLOSS in MW obtained by different methods. Between DE and PSO approaches, DE performance is better as it obtained the optimum solution with less number of generations and function evaluations. Since SQP method assumes all variables as continuous, after optimization procedure, load ow program is used to nd the actual PLOSS , with the discrete variables are adjusted to the nearest possible value.

Fig. 2. Performance characteristics of IEEE 14-bus system.

A good optimization results in convergence of all control variables to a steady value. Fig. 4 shows the variation of the continuous control variable, V g , with respect to the number of generations. All generator voltages settle to a steady value by 30 generations. Fig. 5 shows the variation of the discrete control variables tap position and capacitor bank switching. It can be observed that all discrete control variables also converge well before 30 generations. Fig. 6 shows the effect of optimum control variable setting on static voltage security in terms of VCPI. At all buses, the VCPI value is less than the threshold value for this system, which is 0.2065 as obtained from off-line studies. Table 5 gives the details of the control variables and PLOSS obtained with different optimization techniques. In order to verify the robustness of the proposed methodology simulation is carried out for 30 independent runs with different initial population. For each run, the nal solution and cpu time were observed. The important statistical details are listed in Table 6. It can be seen that DE algorithm

Fig. 1. Network diagram of IEEE 14-bus system.

Fig. 3. Statistics of DE in each generation for IEEE 14-bus system.

1556

M. Varadarajan, K.S. Swarup / Applied Soft Computing 8 (2008) 15491561

Table 4 PLOSS before and after optimization for IEEE 14-bus system Compared item PLOSS (MW) No. of iterations No. of function evaluations Base case 13.49 DE 13.239 63 1890 PSO 13.250 80 2400 SQP 13.246 9 316

more robust than PSO and faster. To ensure a near optimum solution for any random trial, the standard deviation for multiple runs should be very low, which is satised better by DE, when compared to PSO. 7.2. IEEE 30-bus system In this section, results obtained for IEEE 30-bus system is presented. The system data giving branch parameters and loads are available in [17]. The network consists of 41 branches, six generator buses and 24 load buses. Four branches 69, 610, 4 12 and 2728 have tap changing transformers with 20 discrete

Fig. 6. VCPI for IEEE 14-bus system.

Fig. 4. Convergence of control variable V g for IEEE 14-bus system.

steps of 0.01 p.u. each. The buses with possible reactive power source installations are 10 and 24. The available reactive powers of capacitor banks are within the interval 0 to 30 MVAr in discrete steps of 1 MVAr. All bus voltages are required to be maintained within the range of 0.951.1 p.u. Voltages of PQ buses 26 (V 26 0:932 p.u.), 29 (V 29 0.940 p.u.) and 30 (V 30 0:928 p.u.) violates the lower limit in the base case. Fig. 7 shows the convergence characteristics for the best solution. It can be seen that PLOSS is reduced to a minimum value of 5.011 MW from the base case loss of 5.66 MW. Fig. 8 shows the information about the average value, standard deviation and variance of the population at each generation. The PLOSS values before and after optimization obtained with various methods are given in Table 7. As in the case of IEEE 14-bus system, all control variables converge to a steady value. Fig. 9 shows the variation of the control variable, V g , with respect to the number of generations. All generator voltages settle to a steady value by 40 generations. Fig. 10 shows the variation of the discrete control variables tap position and capacitor bank switching. It can be observed that all discrete control variables also converges to steady value. Table 8 shows the control variable setting and PLOSS obtained by different methods.
Table 5 Values of control variables (p.u.) and PLOSS before and after optimization for IEEE 14-bus system Variable V g2 V g3 V g6 V g8 T 47 T 49 T 56 QC9 QC14 Base case 1.0450 1.0100 1.0700 1.0900 0.9467 0.9524 0.9091 0.1800 0.1800 13.49 DE 1.0449 1.0416 1.1000 1.1000 1.0600 1.0400 1.1000 0.1800 0.0600 13.239 PSO 1.0443 1.0138 1.1000 1.0882 1.0700 1.0400 1.0000 0.1800 0.0600 13.250 SQP 1.0442 1.0124 1.1000 1.1000 1.0586 1.0634 1.0781 0.1751 0.0632 13.246 PSO [8] 1.0463 1.0165 1.1000 1.1000 0.9400 0.9300 0.9700 0.1800 0.0600 13.32

Fig. 5. Convergence of discrete control variables for IEEE 14-bus system.

PLOSS (MW)

M. Varadarajan, K.S. Swarup / Applied Soft Computing 8 (2008) 15491561 Table 6 Statistical details for IEEE 14- bus system Compared item PLOSS best (MW) PLOSS worst (MW) PLOSS average (MW) Standard deviation Average no. of iterations Average CPU time (s) DE 13.239 13.275 13.250 0.0161 62 8.172 PSO 13.250 13.402 13.352 0.0640 74 9.283 Table 7 PLOSS before and after optimization for IEEE 30-bus system Compared item PLOSS (MW) No. of iterations No. of function Evaluations Base case 5.66 DE 5.011 66 1980 PSO 5.116 70 2100

1557

SQP 5.043 36 2465

The bus voltage prole before and after optimization is shown in Fig. 11. Voltages at all buses, including the buses 26, 29 and 30, are now within the required range of 0.951.1 p.u. Fig. 12 shows the effect of optimum control variable setting on static voltage security in terms of VCPI. At all buses, the VCPI value is less than the threshold value for this system. which is 0.1121 as obtained from off-line studies. Statistical details for IEEE 30-bus system for 30 independent runs are shown in Table 9. It can be observed that

performance and robustness of DE algorithm are better than PSO. DE obtains the optimum value in less number of function evaluations. The standard deviation for 30 runs is very low in the case of DE when compared to PSO, which ensures a near optimum solution for any random trial. 7.3. IEEE 118 bus system In this section performance of DE based optimal reactive power dispatch was evaluated on IEEE 118 bus system with simulation parameters given in Table 3 and the network data

Fig. 7. Performance characteristics for IEEE 30-bus system.

Fig. 9. Convergence of control variable V g for IEEE 30-bus system.

Fig. 8. Statistics of DE in each generation for IEEE 30-bus system.

Fig. 10. Convergence of discrete control variables for IEEE 30-bus system.

1558

M. Varadarajan, K.S. Swarup / Applied Soft Computing 8 (2008) 15491561

Table 8 Values of control variable (p.u.) and PLOSS before and after optimization for IEEE 30-bus system Variable V g1 V g2 V g5 V g8 V g11 V g13 T 69 T 610 T 412 T 2728 QC10 QC24 PLOSS (MW) Base case 1.0500 1.0220 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 1.0000 0.1000 0.1000 5.66 DE 1.0500 1.0446 1.0247 1.0265 1.1000 1.1000 1.0000 1.1000 1.0800 0.9200 0.2600 0.1000 5.011 PSO 1.0500 0.9679 1.0262 1.0267 1.1000 1.1000 0.9700 1.1000 1.0600 0.9200 0.3000 0.1000 5.116 SQP 1.0500 1.0467 1.0386 1.0293 1.0837 1.1000 1.0222 1.0453 1.0686 1.0819 0.2974 0.0999 5.043 PSO [24] 1.0178 1.0246 1.0247 1.0142 1.0172 0.9961 1.0969 0.9251 1.0005 1.0071 0.1537 0.0622 5.092 IPM [24] 1.1000 1.0541 1.1000 1.0335 1.1000 1.0149 0.9933 1.0593 1.0088 0.9971 0.1525 0.0893 5.101

given in [18]. The network consists of 186 branches, 54 generator buses and 64 load buses. Nine branches have tap changing transformers with 20 discrete steps of 0.01 p.u. each. There are 12 reactive power source installations. The available reactive

powers of capacitor banks are within the interval (030) MVAr in discrete steps of 1 MVAr. All bus voltages are required to be maintained within the range of 0.951.1 p.u. Voltages of PQ buses 53 and 118 violates the lower limit in the base case. Fig. 13 shows the real power loss variation against generations. It can be seen that PLOSS is reduced to a minimum value of 128.318 MW from a base case loss of 132.45 MW. Fig. 14 shows evolution of the algorithm in terms of average value, standard deviation and variance of population at each generation. It is also found that all control variables converge to a steady value by the time termination criteria is satised. Table 10 lists the minimum PLOSS obtained by using different methods namely DE, PSO [24] and IPM [24]. Statistical details for IEEE 118 bus system is as follows: best, worst, and average PLOSS obtained for 30 simulations are 128.318, 129.579, and 129.0817 MW, respectively. The standard deviation of PLOSS is 0.345 MW. Average cpu time taken is 42.1556 s with an average of 193 iterations. 7.4. Effect of initial population and population size

Fig. 11. Voltage prole for IEEE 30-bus system before and after optimization.

To study the effect of initial population on the performance of the algorithm, simulation is carried out on test systems with

Fig. 12. VCPI at each bus for IEEE 30-bus system.

Fig. 13. Performance characteristics of IEEE 118-bus system using DE.

M. Varadarajan, K.S. Swarup / Applied Soft Computing 8 (2008) 15491561 Table 9 Statistical details for IEEE 30-bus system Compared item PLOSS best (MW) PLOSS worst (MW) PLOSS average (MW) Standard deviation Average no. of iterations Average CPU time (s) DE 5.011 5.022 5.013 0.0026 66 13.647 PSO 5.116 5.218 5.1254 0.0291 69 16.420

1559

Table 10 PLOSS before and after optimization for IEEE 118-bus system Base case PLOSS (MW) 132.45 DE 128.318 PSO [24] 131.908 IPM [24] 132.110 Fig. 16. Effect of population size on PLOSS for IEEE 30-bus system.

Fig. 14. Statistics of DE at each generation for IEEE 118-bus system.

a constant population size. Fig. 15 shows the effect of initial population on the nal solution for four trials in the case of IEEE 30-bus system. It can be observed that even though the convergence is different for different initial population, the algorithm converges to the optimum solution. It is found that the algorithm is insensitive to the initial population i.e. starting points for the optimization process. It is also essential to study the effect of population size on the optimization procedure. Even though a number of heuristic relations are available for nding the population size, there is no hard and fast rule which can be universally adopted. In most cases, the population size is xed by trial and error method. Fig. 16 shows the effect of population size on the objective value, with the same initial population for IEEE 30-bus system. As the population size increases a faster convergence to the optimum solution can be obtained at a cost of increased computation time. To achieve a compromise between convergence to the optimal solution and reduced computation time, a population size of 30 was used for all the test cases. 8. Conclusion A differential evolution algorithm based OPF for reactive power dispatch and voltage control in power system planning and operation studies is proposed. The problem is formulated as a mixed integer nonlinear optimization problem. Compared to PSO, DE has fewer control parameters (population size, step size and crossover rate). Further, the penalty parameterless technique of handling inequality constraints, effectively eliminates the trial and error method of assigning penalty coefcients and also makes the process system independent. The proposed DE approach has been evaluated on IEEE 14, IEEE 30, and IEEE 118-bus systems and the results were compared with that obtained using PSO and SQP. DE was found to be more robust as it gave minimum standard deviation among the solutions obtained from multiple random trials. In each case, the security of the system was considered, while optimizing the control variables for real power loss minimization.

Fig. 15. Effect of initial population on PLOSS for IEEE 30-bus system.

1560

M. Varadarajan, K.S. Swarup / Applied Soft Computing 8 (2008) 15491561

Appendix A. Voltage collapse proximity indicator Using the voltage magnitude and voltage angle information, voltage collapse proximity indicator at each bus [16] is calculated as follows. PN 0 V 1 m1; 6 k m VCPIk (A.1) Vk 0 and Vm is given by Y km 0 V m PN
j1; 6 k Y k j

where C C 1 C 2 and C > 4. Usually C 1 and C2 are both set to be 2.05 and x set to be 0.729. Suitable selection of inertia weight v provides a balance between global and local explorations. In general, the inertia weight v is usually be set as decreasing linearly from vmax to vmin , according to the following equation v vmax vmax vmin iter itermax (B.4)

Vm

(A.2)

where, Vk voltage phasor at bus k, Vm voltage phasor at bus m, Y km admittance between buses k and m, N number of buses. The VCPI at the bus k obtained from Eq. (A.1), will vary from 0 to 1, with zero indicating a voltage stable condition and one indicating a voltage collapse. Appendix B. Particle swarm optimization PSO was developed through simulation of simplied social methods and is basically simulation of the social behavior of a ock of birds in two-dimension space. Let x and v represent a particle position and its corresponding velocity in a search space. The best previous position of a particle is recorded and represented as pBest. The index of the best particle among all the particles in the group is represented as gBest [19]. Constriction function is used to ensure the convergence of PSO [20]. The modied velocity of each particle can be calculated using the current velocity and the distance from pBest and gBest as vk1 xvi vk C 1 rand pBest xk C 2 rand i i i gBest xk i (B.1) From the above equation, a certain velocity that gradually gets close to pBest and gBest can be calculated. The particle velocity is limited by some maximum value vmax . This parameter determines the tness with which regions are to be searched between the present position and the target position and enhances the local exploration of the problem space. The current position can be modied by the following equation. xk1 xk vk1 i i i (B.2)

where, itermax is the maximum number of generations and iter is the current generation. Empirical studies have shown that PSO performs well when vmax 0:9 and vmin 0:4 [25,21]. A general strategy for setting v, C1 , and C2 , to guarantee the convergence of the particles is given in [22].

References
[1] N. Deeb, S.M. Shahidepour, Linear reactive power optimization in a large power network using the decomposition approach, IEEE Trans. Power Syst. 5 (2) (1990) 428435. [2] S. Granville, Optimal reactive dispatch through interior point methods, IEEE Trans. Power Syst. 9 (1) (1994) 136146. [3] J.A. Momoh, S.X. Guo, E.C. Ogbuobiri, R. Adapa, The quadratic interior point method solving power system optimization problems, IEEE Trans. Power Syst. 9 (3) (1994) 13271336. [4] A.M. Chebbo, M.R. Irving, Combined active and reactive despatch Part 1: Problem formulation and solution algorithm, IEE Proc. Gener. Transm. Distrib. 142 (4) (1995) 393400. [5] Q.H. Wu, J.T. Ma, Power system optimal reactive power dispatch using evolutionary programming, IEEE Trans. Power Syst. 10 (3) (1995) 12431249. [6] L.L. Lai, J.T. Ma, Application of evolutionary programming to reactive power planning Comparison with nonlinear programming approach, IEEE Trans. Power Syst. 12 (1) (1997) 198206. [7] K.Y. Lee, X. Bai, Y.M. Park, Optimization method for reactive power planning by using a modied simple genetic algorithm, IEEE Trans. Power Syst. 10 (4) (1995) 18431850. [8] H. Yoshida, Y. Fukuyama, K. Kawata, S. Takayama, Y. Nakanishi, A particle swarm optimization for reactive power and voltage control considering voltage security assessment, IEEE Trans. Power Syst. 15 (4) (2001) 12321239. [9] B. Zhao, C.X. Guo, Y.J. Cao, A multi-agent based particle swarm optimization approach for reactive power dispatch, IEEE Trans. Power Syst. 20 (2) (2005) 10701078. [10] R. Storn, K. Price, Differential evolution A simple and efcient adaptive scheme for global optimization over continuous spaces, Technical report TR-95-012, ICSI, March 1995. [11] R. Storn, K. Price, Differential evolution A simple and efcient heuristic for global optimization over continuous spaces, J. Global Optim. 11 (1997) 341359. [12] J.-P. Chiou, C.-F. Chang, C.-T. Su, Capacitor placement in large-scale distribution systems using variable scaling hybrid differential evolution, Int. J. Electr. Power Energy Syst. 28 (10) (2006) 739745. [13] C.H. Liang, C.Y. Chung, K.P. Wong, X.Z. Duan, Parallel optimal reactive power ow based on cooperative co-evolutionary differential evolution and power system decomposition, IEEE Trans. Power Syst. 22 (1) (2007) 249257. [14] C.-F. Chang, J.-J. Wong, J.-P. Chiou, C.-T. Su, Robust searching hybrid differential evolution method for optimal reactive power planning in large-scale distribution systems, Electr. Power Syst. Res. 77 (56) (2007) 430437. [15] K. Deb, An efcient constraint handling method for genetic algorithms, Comput. Methods Appl. Mechan. Eng. 186 (2000) 311338.

where vk is the current velocity of particle i at iteration k, vk1 is i i the modied velocity of particle i, rand is the uniformly distributed random number between 0 and 1, xk is the current i position of particle i at iteration k, xk1 is the modied position i of particle i, vi is the inertia weight factor of particle i, x is the constriction factor, and C1 ; C2 are acceleration constant. The constriction factor x is a function of C 1 and C 2 as given below x 2 p j2 C C 2 4C j (B.3)

M. Varadarajan, K.S. Swarup / Applied Soft Computing 8 (2008) 15491561 [16] V. Balamourougan, T.S. Sidhu, M.S. Sachdev, Technique for online prediction of voltage collapse, IEE Proc. Gener. Transm. Distrib. 151 (4) (2004). [17] Q.H. Wu, Y.J. Cao, J.Y. Wen, Optimal reactive power dispatch using an adaptive genetic algorithm, Electr. Power Energy Syst. 20 (8) (1998) 563569. [18] Power System Test Case Archive, 2006, December [Online], Available: http://www.ee.washington.edu/research/pstca/pf118/pg_tca118bus.htm. [19] J. Kennedy, R. Eberhart, A particle swarm optimization, in: Proceedings of IEEE International Conference on Neural Networks, vol. 4, Australia, (1995), pp. 19421948. [20] Math stuff about PSO, 2006, December [Online], Available: http://clerc. maurice.free.fr/pso/index.htm.

1561

[21] Y. Shi, R.C. Eberhart, Empirical study of particle swarm optmization, in: Proceedings of 1999 Congress on Evolutionary Computation, USA, (1999), pp. 19451950. [22] M. Clerc, The swarm and queen: towards a deterministic ans adaptive PSO, in: Proceedings of 1999 Congress on Evolutionary Computation, USA, (1999), pp. 19511957. [23] Matlab Optimization Toolbox 2, Users Guide, January 1999. [24] J.G. Vlachogiannis, K.Y. Lee, A comparative study of particle swarm optimization for optimal steady state performance of power systems, IEEE Trans. Power Syst. 21 (4) (2006) 17181728. [25] Y. Shi, R.C. Eberhart, A modied particle swarm optimizer, in: Proceedings of 1998 IEEE International Conference on Evolutionary Computation, Alaska, (1998), pp. 6973.

You might also like