Electrical Power and Energy Systems 25 (2003) 97–105

www.elsevier.com/locate/ijepes

A niched Pareto genetic algorithm for multiobjective

environmental/economic dispatch
M.A. Abido*
Department of Electrical Engineering, King Fahd University of Petroleum and Minerals, KFUPM Box 183, Dhahran 31261, Saudi Arabia
Received 2 July 2001; revised 13 February 2002; accepted 2 April 2002

Abstract
A niched Pareto genetic algorithm (NPGA) based approach to solve the multiobjective environmental/economic dispatch (EED) problem
is presented in this paper. The EED problem is formulated as a non-linear constrained multiobjective optimization problem. The proposed
NPGA based approach handles the problem as a multiobjective problem with competing and non-commensurable cost and emission
objectives. One of the main advantages of the proposed approach is that there is no restriction on the number of optimized objectives. The
proposed approach has a diversity-preserving mechanism to overcome the premature convergence problem. A hierarchical clustering
algorithm is developed and imposed to provide the decision maker with a representative and manageable Pareto-optimal set. In addition,
fuzzy set theory is employed to extract the best compromise solution. Several optimization runs of the proposed approach are carried out on
the standard IEEE 30-bus test system. The results demonstrate the capabilities of the proposed approach to generate well-distributed Pareto-
optimal non-dominated solutions of the multiobjective EED problem in one single run. The comparison with the classical methods
demonstrates the superiority of the proposed approach and confirms its potential to solve the multiobjective EED problem. q 2002 Elsevier
Science Ltd. All rights reserved.
Keywords: Economic dispatch; Environmental impact; Multiobjective optimization; Evolutionary algorithms

1. Introduction options require installation of new equipment and/or

modification of the existing ones that involve considerable
The basic objective of economic dispatch (ED) of capital outlay and, hence, they can be considered as long-
electric power generation is to schedule the committed term options. The emission dispatching option is an
generating unit outputs so as to meet the load demand at attractive short-term alternative in which the emission in
minimum operating cost while satisfying all unit and system addition to the fuel cost objective are to be minimized. Thus,
equality and inequality constraints. This makes the ED the ED problem can be handled as a multiobjective
problem a large-scale highly non-linear constrained optim- optimization problem with non-commensurable and contra-
ization problem. In addition, the increasing public aware- dictory objectives. In recent years, this option has received
ness of the environmental protection and the passage of the much attention [4 – 11] since it requires only small
Clean Air Act Amendments of 1990 have forced the utilities modification of the basic ED to include emissions.
to modify their design or operational strategies to reduce Different techniques have been reported in the literature
pollution and atmospheric emissions of the thermal power pertaining to environmental/economic dispatch (EED)
plants. problem. In Refs. [4,5] the problem has been reduced to a
Several strategies to reduce the atmospheric emissions single objective problem by treating the emission as a
have been proposed and discussed [1 – 3]. These include constraint. This formulation, however, has a severe
installation of pollutant cleaning equipment, switching to difficulty in getting the trade-off relations between cost
low emission fuels, replacement of the aged fuel-burners and emission. Alternatively, minimizing the emission has
with cleaner ones, and emission dispatching. The first three been handled as another objective in addition to the cost. A
linear programming based optimization procedures in which
* Tel.: þ966-3-860-4379; fax: þ 966-3-860-3535. the objectives are considered one at a time was presented in
E-mail address: [email protected] (M.A. Abido). Ref. [6]. However, many mathematical assumptions have to
0142-0615/03/$ - see front matter q 2002 Elsevier Science Ltd. All rights reserved.
PII: S 0 1 4 2 - 0 6 1 5 ( 0 2 ) 0 0 0 2 7 - 3
98 M.A. Abido / Electrical Power and Energy Systems 25 (2003) 97–105

be given to simplify the problem. Furthermore, this utions. A hierarchical clustering technique is implemented
approach does not give any information regarding the to provide the power system operator with a representative
trade-offs involved. and manageable Pareto-optimal set. A fuzzy-based mech-
In other research direction, the multiobjective EED anism is employed to extract the best compromise solution.
problem was converted to a single objective problem by The potential of the proposed approach to handle the
linear combination of different objectives as a weighted sum multiobjective EED problem is investigated. Several runs
[7 –10]. The important aspect of this weighted sum method are carried out on a standard test system and the results are
is that a set of non-inferior (or Pareto-optimal) solutions can compared to the classical techniques. The effectiveness of
be obtained by varying the weights. Unfortunately, this the proposed approach to solve the multiobjective EED
requires multiple runs as many times as the number of problem is demonstrated.
desired Pareto-optimal solutions. Furthermore, this method
cannot be used in problems having a non-convex Pareto-
optimal front. To avoid this difficulty, the 1-constraint 2. Problem formulation
method was presented in Refs. [11 –13]. This method
optimizes the most preferred objective and considers the The EED problem is to minimize two competing
other objectives as constraints bounded by some allowable objective functions, fuel cost and emission, while satisfying
levels 1. The most obvious weaknesses of this approach are several equality and inequality constraints. Generally the
that it is time-consuming and tends to find weakly non- problem is formulated as follows.
dominated solutions.
The recent direction is to handle both objectives 2.1. Problem objectives
simultaneously as competing objectives. A fuzzy multi-
objective optimization technique for EED problem was Minimization of fuel cost. The generator cost curves are
proposed [14]. However, the solutions produced are sub- represented by quadratic functions with sine components to
optimal and the algorithm does not provide a systematic represent the valve loading effects. The total $/h fuel cost
framework for directing the search towards Pareto-optimal F(PG) can be expressed as
front. An evolutionary algorithm based approach evaluating  h  i
X
N
 
the economic impacts of environmental dispatching and fuel FðPG Þ ¼ ai þ bi PGi þ ci P2Gi þ di sin ei Pmin
Gi 2 PGi 
switching was presented in Ref. [15]. However, some non- i¼1
dominated solutions may be lost during the search process ð1Þ
while some dominated solutions may be misclassified as
non-dominated ones due to the selection process adopted. A where N is the number of generators, ai, bi, ci, di, and ei are
fuzzy satisfaction-maximizing decision approach was the cost coefficients of the ith generator, and PGi is the real
successfully applied to solve the biobjective EED problem power output of the ith generator. PG is the vector of real
[16]. However, extension of the approach to include more power outputs of generators and defined as
h iT
objectives is a very involved question. A multiobjective
PG ¼ PG1 ; PG2 ; …; PGN ð2Þ
stochastic search technique (MOSST) for the multiobjective
EED problem was presented in Ref. [17]. However, the Minimization of emission. The total ton/h emission E(PG)
technique is computationally involved and time-consuming. of atmospheric pollutants such as sulphur oxides SOx and
In addition, there is no effort to avoid the search bias to some nitrogen oxides NOx caused by fossil-fueled thermal units
regions in the problem space that may result in premature can be expressed as
convergence. This degrades the Pareto-optimal front and
more efforts should be done to preserve the diversity of the X
N    
EðPG Þ ¼ 1022 ai þ bi PGi þ gi P2Gi þ zi exp li PGi ð3Þ
non-dominated solutions. i¼1
Recently, the studies on evolutionary algorithms have
shown that these methods can be efficiently used to where ai, bi, gi, zi, and li are coefficients of the ith generator
eliminate most of the difficulties of classical methods emission characteristics.
[18 –21]. Since they use a population of solutions in their
search, multiple Pareto-optimal solutions can, in principle, 2.2. Problem constraints
be found in one single run. A non-dominated sorting genetic
algorithm was presented for EED problem [22]. However, Generation capacity constraint. For stable operation,
the technique is computationally involved due to ranking of real power output of each generator is restricted by lower
all population members into different fronts. and upper limits as follows:
In this paper, a niched Pareto genetic algorithm (NPGA) Pmin max
i ¼ 1; …; N ð4Þ
Gi # PGi # PGi ;
based approach is proposed to solve the EED optimization
problem. The proposed approach has a diversity-preserving Power balance constraint. The total power generation
mechanism to find widely different Pareto-optimal sol- must cover the total demand PD and the real power loss in
 M.A. Abido / Electrical Power and Energy Systems 25 (2003) 97–105 99

transmission lines Ploss. Hence, x 1 dominates x 2 if the following two conditions are
satisfied:
X
N
PGi 2 PD 2 Ploss ¼ 0 ð5Þ
i¼1
;i [ {1; 2; …; Nobj } : fi ðx1 Þ # fi ðx2 Þ ð12Þ

Security constraints. For secure operation, the trans- ’j [ {1; 2; …; Nobj } : fj ðx1 Þ , fj ðx2 Þ ð13Þ
mission line loading Sl is restricted by its upper limit as
If any of the above condition is violated, the solution x 1
Sli # Smax
li ; i ¼ 1; …; nl ð6Þ does not dominate the solution x 2. If x 1 dominates the
where nl is the number of transmission lines. solution x 2, x 1 is called the non-dominated solution. The
solutions that are non-dominated within the entire search
2.3. Problem statement space are denoted as Pareto-optimal and constitute the
Pareto-optimal set or Pareto-optimal front.
Aggregating the objectives and constraints, the problem
can be mathematically formulated as a non-linear con- 4. The proposed approach
strained multiobjective optimization problem as follows.
Recently, the studies on evolutionary algorithms have
Minimize ½FðPG Þ; EðPG Þ ð7Þ
PG shown that these algorithms can be efficiently used to
eliminate most of the difficulties of classical methods such
Subject to as multiple runs and sensitivity to the shape of the Pareto-
gðPG Þ ¼ 0 ð8Þ optimal front. In general, the goal of a multiobjective
optimization algorithm is not only guide the search towards
hðPG Þ # 0 ð9Þ
the Pareto-optimal front but also maintain population
where g and h are the equality and inequality constraints, diversity in the set of the non-dominated solutions.
respectively. Unfortunately, a simple GA tends to converge towards a
single solution due to selection pressure, selection noise,
and operator disruption [23].
3. Principles of multiobjective optimization To overcome these difficulties, the NPGA based
approach is proposed in this work. The elements of the
Many real-world problems involve simultaneous optim- proposed approach can be described as follows.
ization of several objective functions. Generally, these 4.1. Niched Pareto genetic algorithm
functions are non-commensurable and often competing and
conflicting objectives. Multiobjective optimization with Horn et al. [24] proposed a tournament selection scheme
such conflicting objective functions gives rise to a set of based on Pareto dominance principles. Unlike the technique
optimal solutions, instead of one optimal solution. The presented in Ref. [22], only two individuals are randomly
reason for the optimality of many solutions is that no one selected for tournament. To find the winner solution, a
can be considered to be better than any other with respect to comparison set that contains a number of other individuals
all objective functions. These optimal solutions are known in the population is randomly selected. Then, the dominance
as Pareto-optimal solutions. of both candidates with respect to the comparison set is
A general multiobjective optimization problem consists tested. If one candidate only dominates the comparison set,
of a number of objectives to be optimized simultaneously he is selected as the winner. Otherwise, implement sharing
and is associated with a number of equality and inequality procedure to specify the winner candidate. Generally, the
constraints. It can be formulated as follows tournament selection is carried out as follows.
Minimize fi ðxÞ i ¼ 1; …; Nobj ð10Þ Pareto domination tournaments. Consider a set of N
x
( population members, each having Nobj objective function
gj ðxÞ ¼ 0 j ¼ 1; …; M values. The following procedure can be used to find the non-
Subject to : ð11Þ dominated set of solutions:
hk ðxÞ # 0 k ¼ 1; …; K
where fi is the ith objective functions, x is a decision vector Step 1: Begin with i ¼ 1:
that represents a solution, Nobj is the number of objectives. Step 2: Pick randomly two candidates for selection x 1
M and K are the numbers of equality and inequality and x 2.
constraints, respectively. Step 3: Pick randomly a comparison set of individuals
For a multiobjective optimization problem, any two from the population.
solutions x 1 and x 2 can have one of two possibilities: one Step 4: Compare each candidate, x 1 and x 2, against each
dominates the other or none dominates the other. In a individual in the comparison set for domination using the
minimization problem, without loss of generality, a solution conditions for domination given in Eqs. (12) and (13).
100 M.A. Abido / Electrical Power and Energy Systems 25 (2003) 97–105

Step 5: If one candidate is dominated by the comparison represented by a real number within its lower limit ai and
set while the other is not, then select the later for upper limit bi, i.e. xi [ ½ai ; bi : The RCGA crossover and
reproduction and go to Step 7, else proceed to Step 6. mutation operators are described as follows.
Step 6: If neither or both candidates are dominated by the Crossover. A blend crossover operator (BLX-a ) has
comparison set, then use sharing to choose the winner. been employed in this study. This operator starts by
Step 7: If i ¼ N is reached, stop selection procedure, else choosing randomly a number from the interval ½xi 2 aðyi 2
set i ¼ i þ 1 and go to Step 2. xi Þ; yi þ aðyi 2 xi Þ; where xi and yi are the ith parameter
values of the parent solutions and xi , yi : To ensure the
Sharing procedure. To prevent the genetic drift problem, balance between exploitation and exploration of the search
a form of sharing should be carried out when there is no space, a ¼ 0:5 is selected. This operator can be depicted as
preference between two candidates. This form of sharing shown in Fig. 1.
maintains the genetic diversity along the population fronts Mutation. The non-uniform mutation operator has been
and allows the GA to develop a reasonable representation of employed in this study. In this operator, the new value x0i of
the Pareto-optimal front. Generally, the basic idea behind the parameter xi after mutation at generation t is given as
sharing is: the more individuals are located in the (
neighborhood of a certain individual, the more its fitness 0
xi þ Dðt; bi 2 xi Þ if t ¼ 0
xi ¼ ð17Þ
value is degraded [23,24]. xi 2 Dðt; xi 2 ai Þ if t ¼ 1
The sharing procedure is performed in the following way
for the candidate i: and
 b

Step 1: Begin with j ¼ 1: Dðt; yÞ ¼ y 1 2 rð12t=gmax Þ ð18Þ
Step 2: Compute a normalized Euclidean distance
measure with another individual j in the current where t is a binary random number, r is a random number
population, as follows r [ ½0; 1; gmax is the maximum number of generations, and
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
!
b is a positive constant chosen arbitrarily. In this study,
u Nobj
uX Ji 2 Jj 2
t k
b ¼ 5 was selected. This operator gives a value x0i [ ½ai ; bi 
k
dij ¼ u l
ð14Þ such that the probability of returning a value close to xi
k¼1 Jk 2 Jk
increases as the algorithm advances. This makes uniform
where Nobj is the number of problem objectives. The search in the initial stages where t is small and vary locally
parameters Jku and Jkl are the upper and lower values of at the later stages.
the kth objective function Jk.
Step 3: This distance dij is compared with a prespecified 4.3. Reducing Pareto set by clustering
niche radius sshare and the following sharing function
value is computed as: In some problems, the Pareto-optimal set can be
8   extremely large or even contain an infinite number of
>
<1 2 dij 2 solutions. In this case, reducing the set of non-dominated
; if dij # sshare
Shðdij Þ ¼ sshare ð15Þ solutions without destroying the characteristics of the trade-
>
:
0; otherwise off front is desirable from the decision maker’s point of
view. An average linkage based hierarchical clustering
algorithm [26] is employed to reduce the Pareto set to
Step 4: Set j ¼ j þ 1: If j # N; go to Step 2, else calculate manageable size. It works iteratively by joining the adjacent
niche count for the candidate i as follows: clusters until the required number of groups is obtained. It
X
N can be described as: given a set P whose size exceeds the
mi ¼ Shðdij Þ ð16Þ maximum allowable size N, it is required to form a subset
j¼1 P p with the size N. The algorithm is illustrated in the
following steps.
Step 5: Repeat the above steps for the second candidate.
Step 6: Compare m1 and m2. If m1 , m2 ; then choose the
first candidate, else choose the second candidate.

4.2. Real-coded genetic algorithm (RCGA)

Due to difficulties of binary representation when dealing

with continuous search space with large dimension, the
proposed approach has been implemented using real-coded
genetic algorithm (RCGA) [25]. A decision variable xi is Fig. 1. Blend crossover operator (BLX-a ).
 M.A. Abido / Electrical Power and Energy Systems 25 (2003) 97–105 101

Step 1: Initialize cluster set C; each individual i [ P order to make it suitable for solving real-world non-linear
constitutes a distinct cluster. constrained optimization problems. The following modifi-
Step 2: If number of clusters # N, then go to Step 5, else cations have been incorporated in the basic algorithm.
go to Step 3.
Step 3: Calculate the distance of all possible pairs of
(a) To satisfy the problem constraints, a procedure is
clusters. The distance dc of two clusters c1 and c2 [ C is
imposed to check the feasibility of the initial
given as the average distance between pairs of individ-
population individuals and the generated children
uals across the two clusters
through GA operations. This ensures the feasibility of
1 X Pareto-optimal solutions.
dc ¼ dði ; i Þ ð19Þ
n1 n2 i1 [c1 ;i2 [c2 1 2 (b) A procedure for updating the Pareto-optimal set is
developed. In every generation, the non-dominated
where n1 and n2 are number of individuals in clusters c1 solutions in the first front are combined with the
and c2, respectively. The function d reflects the distance existing Pareto-optimal set. The augmented set is
in the objective space between individuals i1 and i2. processed to extract its non-dominated solutions that
Step 4: Determine two clusters with minimal distance dc. represent the updated Pareto-optimal set.
Combine these clusters into a larger one. Go to Step 2. (c) A hierarchical clustering procedure based on the
Step 5: Find the centroid of each cluster. Select the average linkage method is incorporated to provide
nearest individual in this cluster to the centroid as a the decision maker with a representative and manage-
representative individual and remove all other individ- able Pareto-optimal set without destroying the charac-
uals from the cluster. teristics of the trade-off front.
Step 6: Compute the reduced non-dominated set P p by (d) A fuzzy-based mechanism is employed to extract the
uniting the representatives of the clusters. best compromise solution over the trade-off curve and
assist the decision maker to adjust the generation levels
4.4. Best compromise solution efficiently.

Upon having the Pareto-optimal set of non-dominated 5.2. Settings of the proposed approach
solution, the proposed approach presents one solution to the
decision maker as the best compromise solutions. Due to The techniques used in this study were developed and
imprecise nature of the decision maker’s judgment, the ith implemented on 133 MHz PC using FORTRAN language.
objective function is represented by a membership function On all optimization runs, the population size and the
mi defined as [8] maximum number of generations were selected as 200 and
8
> 1 Fi # Fimin 500, respectively. The maximum size of the Pareto-optimal
>
>
>
< F max 2 F set was chosen as 50 solutions. If the number of the non-
i i
mi ¼ Fimin , Fi , Fimax ð20Þ dominated Pareto-optimal solutions exceeds this bound, the
>
> F max
2 F min
clustering technique is called. Crossover and mutation
>
>
i i
:
0 Fi $ Fimax probabilities were selected as 0.9 and 0.01, respectively, in
all optimization runs.
For each non-dominated solution k, the normalized
membership function m k is calculated as
X
Nobj 6. Results and discussions
mki
mk ¼ i¼1
ð21Þ In this study, the standard IEEE 30-bus 6-generator test
X X
M N obj
system is considered to investigate the effectiveness of the
mki proposed approach. The single-line diagram of this system
k¼1 i¼1
is shown in Fig. 2 and the detailed data are given in Refs. [6,
where M is the number of non-dominated solutions. The 11]. The values of fuel cost and emission coefficients are
best compromise solution is the one having the maximum given in Table 1.
value of m k. Initially, the effect of comparison set size on the
proposed approach performance is investigated. The effec-
tiveness of the proposed approach to produce a representa-
5. Implementation of the proposed approach tive Pareto-optimal front is examined for different sizes
starting from a size of 5 to 75% of the population size. Fig. 3
5.1. The basic modifications shows the Pareto-optimal front with different comparison
set sizes. It is clear that the performance is degraded with the
In this study, the basic NPGA has been developed in increase of comparison set size. It was observed that 10%
102 M.A. Abido / Electrical Power and Energy Systems 25 (2003) 97–105

Fig. 3. Effect of the comparison set size on Pareto-optimal front.

nation of cost and emission objectives as follows

Minimize wFðPG Þ þ ð1 2 wÞlEðPG Þ ð22Þ
PG

where l is a scaling factor which was selected as 3000 in

this study and w is a weighting factor. To generate 50 non-
dominated solutions, the algorithm was applied 50 times
with varying w as a random number w ¼ rand½0; 1: The
Pareto-optimal front of RCGA is shown in Fig. 5. Applying
Fig. 2. Single-line diagram of IEEE 30-bus test system. the proposed NPGA based approach, the distribution of the
non-dominated solutions in Pareto-optimal front is shown in
size gives a satisfactory performance of the proposed Fig. 6. It is clear that the solutions are diverse and well
approach. distributed over the trade-off curve. Comparing Figs. 5 and
To demonstrate the potential of the proposed approach 6, it can be concluded that, the non-dominated solutions of
for different problem complexities and trade-off surfaces, the proposed approach not only have better diversity
two different cases have been considered as follows. characteristics but also were obtained in a single run.
Case (a). For comparison purposes with the reported The run time per generation of the single objective
results, the system is considered as lossless and the security approach to produce only one solution was 14.22 s while
constrain is released. At first, fuel cost and emission are that of the proposed approach to produce 50 solutions was
optimized individually to get the extreme points of the 14.46 s. It is quiet evident that the proposed approach run
trade-off surface. Convergence of fuel cost and emission time to generate the entire Pareto set is only 1.7% more than
objectives are shown in Fig. 4. The best results of cost and that of the aggregation method to generate only one
emission when optimized individually are given in Table 2. solution. This demonstrates that the proposed approach is
For completeness, the RCGA was applied to find the much faster and more efficient than the classical techniques
Pareto-optimal solutions where the problem was treated as a in handling the multiobjective optimization problems.
single objective optimization problem by linear combi- The results of the proposed approach were compared to

Table 1
Generator cost and emission coefficients

G1 G2 G3 G4 G5 G6

Cost a 10 10 20 10 20 10
b 200 150 180 100 180 150
c 100 120 40 60 40 100
Emission a 4.091 2.543 4.258 5.426 4.258 6.131
b 25.554 26.047 25.094 23.550 25.094 25.555
g 6.490 5.638 4.586 3.380 4.586 5.151
z 2.0 £ 1024 5.0 £ 1024 1.0 £ 1026 2.0 £ 1023 1.0 £ 1026 1.0 £ 1025
l 2.857 3.333 8.000 2.000 8.000 6.667
 M.A. Abido / Electrical Power and Energy Systems 25 (2003) 97–105 103

Fig. 4. Convergence of cost and emission objective functions of case (a).

Fig. 6. Pareto-optimal front of the proposed approach in case (a).

results of cost and emission when optimized individually are

given in Table 2. The values of the best cost and the best
emission objectives with the proposed approach are given in
Tables 3 and 4, respectively. The distribution of the non-
dominated solutions of RCGA when applied for 50 times is
shown in Fig. 8. The distribution of the non-dominated
solutions of the proposed approach is shown in Fig. 9. It can
be seen that the proposed approach preserves the diversity
of the non-dominated solutions over the trade-off front and
produce the non-dominated solutions in one single run.
Best compromise solution. The membership functions
given in Eqs. (21) and (22) are used to evaluate each
member of the Pareto-optimal set. Then, the best compromise
solution that has the maximum value of membership function
Fig. 5. Pareto-optimal front of objective aggregation in case (a). can be extracted. This procedure is applied in both cases and
those reported using linear programming (LP) [6] and the best compromise solutions are given in Table 5.
MOSST [17]. The comparison is given in Tables 3 and 4. It is
quite evident that the proposed approach gives better results.
Case (b). In this case, the power loss has been taken into 7. Conclusion
account. Convergence of fuel cost and emission objectives
when optimized individually are shown in Fig. 7. The best In this paper, an approach based on the NPGA has been

Table 2
The best solutions for cost and emission optimized individually

Case (a) Case (b)

Best cost Best emission Best cost Best emission

PG1 0.10954 0.40584 0.11516 0.41007

PG2 0.29967 0.45915 0.30552 0.46308
PG3 0.52447 0.53797 0.59724 0.54349
PG4 1.01601 0.38300 0.98088 0.38950
PG5 0.52469 0.53791 0.51421 0.54386
PG6 0.35963 0.51012 0.35417 0.51501
Fuel cost ($/h) 600.114 638.260 607.777 645.222
Emission (ton/h) 0.22214 0.19420 0.21985 0.19418
104 M.A. Abido / Electrical Power and Energy Systems 25 (2003) 97–105

Table 3
Test results of best fuel cost of the proposed approach

LP [6] MOSST [17] Proposed

Case (a) Case (b)

PG1 0.1500 0.1125 0.1080 0.1245

PG2 0.3000 0.3020 0.3284 0.2792
PG3 0.5500 0.5311 0.5386 0.6284
PG4 1.0500 1.0208 1.0067 1.0264
PG5 0.4600 0.5311 0.4949 0.4693
PG6 0.3500 0.3625 0.3574 0.3993
Best cost 606.314 605.889 600.259 608.147
Corresp. emission 0.22330 0.22220 0.22116 0.22364
Fig. 7. Convergence of cost and emission objective functions of case (b).

Table 4
Test results of best emission of the proposed approach

LP [6] MOSST [17] Proposed

Case (a) Case (b)

PG1 0.400 0.4095 0.4002 0.3923

PG2 0.4500 0.4626 0.4474 0.4700
PG3 0.5500 0.5426 0.5166 0.5565
PG4 0.4000 0.3884 0.3688 0.3695
PG5 0.5500 0.5427 0.5751 0.5599
PG6 0.5000 0.5142 0.5259 0.5163
Best emission 0.19424 0.19418 0.19433 0.19424
Corresp. cost 639.600 644.112 639.182 645.984

Table 5
Best compromise solutions of the proposed approach

Case (a) Case (b)

Fig. 8. Pareto-optimal front of objective aggregation in case (b).
PG1 0.2696 0.2227
PG2 0.3673 0.3787
PG3 0.5594 0.5560
PG4 0.6496 0.7147
PG5 0.5396 0.5500
PG6 0.4486 0.4424
Cost 612.127 615.097
Emission 0.19941 0.20207

presented and applied to environmental/economic power

dispatch optimization problem. The problem has been
formulated as multiobjective optimization problem with
competing fuel cost and environmental impact objectives.
The proposed approach has a diversity-preserving mechan-
ism to find widely different Pareto-optimal solutions. 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 Fig. 9. Pareto-optimal front of the proposed approach in case (b).
 M.A. Abido / Electrical Power and Energy Systems 25 (2003) 97–105 105

is efficient for solving multiobjective optimization where [11] Yokoyama R, Bae SH, Morita T, Sasaki H. Multiobjective generation
multiple Pareto-optimal solutions can be found in one dispatch based on probability security criteria. IEEE Trans Power Syst
1988;3(1):317–24.
simulation run. In addition, the non-dominated solutions in
[12] Abou El-Ela AA, Abido MA. Optimal operation strategy for reactive
the obtained Pareto-optimal set are well distributed and power control. Modell, Simulation Control, Part A, AMSE 1992;
have satisfactory diversity characteristics. The most import- 41(3):19– 40.
ant aspect of the proposed approach is that any number of [13] Hsiao YT, Chiang HD, Liu CC, Chen YL. A computer package for
objectives can be considered. optimal multi-objective VAR planning in large scale power systems.
IEEE Trans Power Syst 1994;9(2):668– 76.
[14] Srinivasan D, Chang CS, Liew AC. Multiobjective generation
schedule using fuzzy optimal search technique. IEE Proc—Gener
Acknowledgments Transm Distrib 1994;141(3):231–41.
[15] Srinivasan D, Tettamanzi A. An evolutionary algorithm for evaluation
The author acknowledges the support of King Fahd of emission compliance options in view of the clean air act
University of Petroleum & Minerals. amendments. IEEE Trans Power Syst 1997;12(1):152–8.
[16] Huang CM, Yang HT, Huang CL. Bi-objective power dispatch using
fuzzy satisfaction-maximizing decision approach. IEEE Trans Power
Syst 1997;12(4):1715– 21.
References [17] Das DB, Patvardhan C. New multi-objective stochastic search
technique for economic load dispatch. IEE Proc—Gener Transm
[1] El-Keib AA, Ma H, Hart JL. Economic dispatch in view of the clean Distrib 1998;145(6):747–52.
air act of 1990. IEEE Trans Power Syst 1994;9(2):972–8. [18] Fonseca CM, Fleming PJ. An overview of evolutionary algorithms in
[2] Talaq JH, El-Hawary F, El-Hawary ME. A summary of environmen- multiobjective optimization. Evol Comput 1995;3(1):1 –16.
tal/economic dispatch algorithms. IEEE Trans Power Syst 1994;9(3): [19] Coello CAC, Christiansen AD. MOSES: a multiobjective optimiz-
1508–16. ation tool for engineering design. Engng Optim 1999; 31(3):337– 68.
[3] Helsin JS, Hobbs BF. A multiobjective production costing model for [20] Goldberg DE. Genetic algorithms for search, optimization, and
analyzing emission dispatching and fuel switching. IEEE Trans Power machine learning. Reading, MA: Addison-Wesley; 1989.
Syst 1989;4(3):836–42. [21] Coello CAC. A comprehensive survey of evolutionary-based multi-
[4] Brodesky SF, Hahn RW. Assessing the influence of power pools on objective optimization techniques. Knowledge Inform Syst 1999;1(3):
emission constrained economic dispatch. IEEE Trans Power Syst 269–308.
1986;1(1):57 –62. [22] Abido MA. A new multiobjective evolutionary algorithm for
[5] Granelli GP, Montagna M, Pasini GL, Marannino P. Emission environmental/economic power dispatch. Proceedings of IEEE PES
constrained dynamic dispatch. Electr Power Syst Res 1992;24:56–64.
Summer Meeting, Vancouver, Canada; 15 –19 July; 2001.
[6] Farag A, Al-Baiyat S, Cheng TC. Economic load dispatch multi-
[23] Mahfoud SW. Niching methods for genetic algorithms. PhD Thesis.
objective optimization procedures using linear programming tech-
University of Illinois at Urbana– Champaign; 1995.
niques. IEEE Trans Power Syst 1995;10(2):731–8.
[24] Horn J, Nafpliotis N, Goldberg DE. A niched Pareto genetic algorithm
[7] Zahavi J, Eisenberg L. Economic–environmental power dispatch.
for multiobjective optimization. Proceedings of the First IEEE
IEEE Trans Syst, Man, Cybernet 1985;SMC-5(5):485–9.
Conference on Evolutionary Computation, IEEE World Congress on
[8] Dhillon JS, Parti SC, Kothari DP. Stochastic economic emission load
Computational Intelligence, vol. 1, Piscataway, NJ: IEEE Service
dispatch. Electr Power Syst Res 1993;26:179–86.
Center; 1994, p. 67–72.
[9] Xu JX, Chang CS, Wang XW. Constrained multiobjective global
optimization of longitudinal interconnected power system by genetic [25] Herrera F, Lozano M, Verdegay JL. Tackling real-coded genetic
algorithm. IEE Proc—Gener Transm Distrib 1996;143(5):435–46. algorithms: operators and tools for behavioral analysis. Artif Intell
[10] Chang CS, Wong KP, Fan B. Security-constrained multiobjective Rev 1998;12(4):265–319.
generation dispatch using bicriterion global optimization. IEE Proc— [26] Morse JN. Reducing the size of nondominated set: pruning by
Gener Transm Distrib 1995;142(4):406–14. clustering. Comput Oper Res 1980;7(1–2):55–66.