A New Multi-objective Particle Swarm Optimization

for Economic Environmental Dispatch

Hasnae BILIL*, Rachid ELLAIA and Mohamed MAAROUFI
Mohammadia School of Engineers
Mohammed V University –Agdal, Rabat, Morocco
{bilil,ellaia,maaroufi}@emi.ac.ma

Abstract—This paper investigates a new approach of computation The proposed method has already been implemented to
using particle swarm in order to resolve economic environmental calculate the optimal multiobjectif economic environmental
dispatch problem. This approach is called accelerated dispatch of IEEE 30-bus. The results of algorithm application
multiobjective particle swarm optimization (AMOPSO) which will be presented in the paper.
incorporates vector function as objective function and uses
matrix computation and updates solutions set, in each iteration, II. PROBLEM FORMULATION
for developing the Pareto front unlike the existing multi-objective
The resolution of economic environmental dispatch
algorithms which use an external archive. We apply this
approach to resolve the problem which treats fuel cost, NOx
problem consists in optimizing the steady state performance of
emissions and active power losses as competing objectives. three objective functions that represent the fuel cost, the
quantity of NOx emissions and the real power losses while
Keywords: Optimal economic environmental dispatch; particle satisfying several equality and inequality constraints. Then the
swarm optimization; multi-objective optimization; matrix general problem can be formulated as follows:
computation.
A. Objective functions
I. INTRODUCTION Consumed fuel cost: The classical economic dispatch
Power system operators and planners have to ensure energy problem of finding the optimal combination of power
supply with the minimum fuel cost. However, with the generation, which minimizes the total fuel cost while satisfying
increasing awareness of the environmental pollutions caused the total required demand can be mathematically expressed as
by fossil fuel emissions, the traditional economic dispatch follows [8]-[11]:
becomes insufficient. The objective of the optimal economic
( )
m m
environmental dispatch is to minimize three objective F1 = Ct = ∑ Ci ( Pgi ) = ∑ ai +bi Pgi + ci Pgi2 ,(1)
functions: fuel cost, NOx emissions and real power losses i =1 i =1
while satisfying operational constraints.
where Ct: total fuel cost ($/h), Ci: is fuel cost of generator i, ai,
Particle swarm optimization (PSO), originated by Kennedy bi and ci: fuel cost coefficients of generator i, Pgi: power
and Eberhart in 1995, was inspired by the choreography of a generated (p.u) by generator i, and m: number of generators.
bird flock. Then, PSO becomes well-known as an efficient
optimization algorithm used, in first time, for solving single NOx emission: The amount of NOx emission is given as a
objective problems and improved for solving multi-criteria function of generator output, that is, the sum of a quadratic and
problems. In fact, there are many modifications of PSO and a exponential function [8, 9]:
lot of work for adapting PSO to problems with multiple
( )
m
objectives [1, 2]. For example, the study presented in [3] F2 = E NOx = ∑ α i + βi Pgi +γ i Pgi2 + exp(λi Pgi ) , (2)
proposes multi-objective PSO with time variant inertia and i =1
acceleration coefficients where inertia weight and PSO
algorithm parameters expressions depend to iteration number. where αi, βi, γi and λi : are emission coefficients of generator i.
Other study developed in [4] proposes multi-objective PSO Active power losses: the transmission network losses can be
with dynamic population size. Others studies presented in [5, 6, presented using B-coefficients [9]:
7] which use hybridization technique.
m m
Unlike all these studies which use inertia weight to develop F3 = Ploss = ∑ ∑ Pgi Bij Pgj , (3)
Pareto front and an external archive to save non-dominated i =1 i =1
solution, we modify the PSO algorithm for causing it to use
matrix computation, then it incorporates function vector as B. Problem constraints
objective function and uses Pareto dominance for selecting best Power balance constraint: The total power generated must
solutions and updating Pareto set. supply the total load demand and the transmission losses:

978-1-4673-4766-2/12/$31.00 ©2012 IEEE

 m the non dominated solutions but it updates all the best solutions
∑
i =1
Pgi − PD − Ploss = 0 (4) at each iteration. AMOPSO approach is based on the
improvement of the population of the best positions, denoted
Xmobest, which converges toward the Pareto optimal set. The
where PD: total load demand (p.u.), and Ploss: transmission population in developing Xpart benefit from the experiences of
losses (p.u.). the entire developed population.
Maximum and minimum limits of power generation: The We use the matrix representation of these two populations
power generated Pgi by each generator is restricted by upper Xmobest and Xpart.
and lower limits:
Let N be the population size. Then:
Pgimin ≤ Pi ≤ Pgimax . (5)
Xmobest = ( Xmobest1,Xmobest 2 ,...,Xmobest N ) (3)
III. MULTI-OBJECTIVE OPTIMIZATION
We are interested in solving problems of the type: Xpart = (Xpart1,Xpart 2 ,...,Xpart N ) (7)

Minimize: F (x ) = ⎡⎣f 1 (x ), f 2 (x ),..., f Nobj (x ) ⎤⎦ (6) Also, the calculation of the evaluation of these two populations
will use a matrix based. We will apply the vector function
F=[f1 ,f2 ,…,f Nobj]T on the matrix of population. Then:
Subject to: g j ( x ) ≤ 0; j = 1,..., M
F ( Xpart ) = ( F ( Xpart1 ),F ( Xpart 2 ),...,F ( Xpart N ) ) (8)
[
Where x = x 1 , x 2 , ..., x n ] is a dimensional vector, each
T
x i = ( i = 1,..., n , n ≥ 1) can be real-valued, integer-valued or F ( Xparti ) = ( f1( Xparti ), f 2( Xparti ),..., f Nobj ( Xparti )) (9)
boolean-valued. f i ( x ) = ( i = 1, ..., Nobj , Nobj ≥ 1) , In our approach, the population in developing doesn’t
g j ( x ), ( j =1,...,M ,M ≥0 ) are linear or nonlinear arbitrary follow a single best position but it moves toward each non
dominated position. Fig. 1 shows the two objective space of the
functions. case of two-objective problem.
The family of optimal solutions of this MO problem is Then, the expression of the population velocity, at iteration
composed of all those potential solutions such that the t, is as follows:
components of the corresponding objective vectors cannot be
simultaneously improved. This is known as the concept of V ( t ) = ω ⋅ V ( t − 1)
Pareto optimality. In a minimization problem, Pareto

( )
dominance and Pareto optimality are defined as follows [11]: N (10)
+ ∑ ρk AN ( Xmobestk (t − 1)) − Xpart (t − 1)
Definition 1(Pareto dominance): A given vector k =1

[
x = x 1 , x 2 , ..., x n ] is said to dominate y = [ y1 , y2 , ..., yn ] if X part (t ) = X part (t −1) + V (t ) (11)
and only if ∀i ∈ {1, 2, ..., n} , xi ≤ yi and
where A N ( X ) = ( X , ..., X ) . ω and ρ k are the acceleration
∃i ∈ {1, 2,..., n} , xi < yi . 


Nelement

Definition 2(Pareto optimality): For a general MO

*
problem, a given solution x ∈ F (where F is the feasible (Xmobestk – Xparti)
solution space) is Pareto optimal if and only if there is no Vi(t+1)
x ∈ F that dominate x * .
Objective function f2

F(Xparti(t)
IV. ACCELERATED MULTI-OBJECTIVE PARTICLE SWARM
OPTIMIZATION
In this paper, we develop a novel approach in order to
better explore the search space then improve final result and
speed up the convergence of the algorithm.
Unlike the existing studies presented in [3, 5, 6, 7]which
apply the PSO algorithm for solving multiobjective problems,
F(Xparti(t+1 ω*Vi(t
the AMOPSO approach doesn’t use an external archive to save
Objective function f1
: Population in developingXpart
: Developed population Xmobest
Figure 1 : Two objective space of the case of two-objective
coefficients.
AMOPSO algorithm does not use an external archive for Start
saving best solutions but improves best positions set
(developed population Xmobest) at each iteration. Then each
particle in motion improves the position of the entire developed Initialize populations Xpart and Xmobest
population if it has a better position. Fig. 2 shows the steps of by using initialization process
developed population updating at iteration t.
A. Flow chart of AMOPSO for ORPD
Fig. 3 shows the flow chart of Accelerated Multi-objective Evaluate objective functions
Particle Swarm Optimization applying to economic
environmental power dispatch.
Compare F(Xpart) and F(Xmobest)

Update Xmobest and F(Xmobest)

X part = ⎡⎣ X part 1 X part 2 ... X part i ... X part N ⎤⎦


N particles
If Yes
Xparti ≺ Xmobesti Is stop
Then
creteriasat Stop
isfied ?
Xmobesti ← Xparti

Xmobest = ⎡⎣Xmobest1 Xmobest 2 ... Xmobesti ... Xmobest N ⎤⎦ No


Nparticles
Update population Xpart
We apply permutations of Xpart columns then each particle updates
the entire developed population simultaneously

X p art ← ⎡⎣ X p art 2 ... X p art i + 1 ... X p art N X part 1 ⎤⎦

Figure 3 : Flow chart of AMOPSO for Optimal Economic /
 Environmental Power Dispatch
N p articles

X part = ⎡⎣ X part 1 X part 2 ... X part i ... X part N ⎤⎦ V. SIMULATION AND RESULTS

N particles The described approach is applied to the used network
If
IEEE test network, including 30 buses, 6 thermal generators
Xparti ≺ Xmobesti (Fig. 4) [8]-[11].
Then
There are 41 transmission lines and the system demand is
Xmobesti ← Xparti 283.4 MW. The information of the system, fuel cost and
emission function coefficients could be obtained from [9].
Xmobest = ⎡⎣ Xmobest1 Xmobest 2 ... Xmobest i ... Xmobest N ⎤⎦

Nparticles We apply our Accelerated MOPSO approach to resolve
optimal power dispatch for one, two and three objectives which
are cited before.
… A. Mono-objective optimization:
Minimal consumed fuel cost:
X part ← ⎡⎣ X part N X part 1 ... X part N −i ... X part N −1 ⎤⎦
 Table 1 gives minimal consumed cost fuel with corresponding
Nparticles
NOx emissions and active losses.
X part = ⎡⎣ X part 1 X part 2 ... X part i ... X part N ⎤⎦
 Table 1 : Minimal fuel cost
N particles
If
Minimal fuel Generated power (MW)
Xparti ≺ Xmobesti cost($)
Then 587.9 Pg1 Pg2 Pg3 Pg4 Pg5 Pg6

Xmobesti ← Xparti Corresponding

Emissions
Xmobest = ⎡⎣ Xmobest1 Xmobest 2 ... Xmobest i ... Xmobest N ⎤⎦ (kg/h)
 271 11.7 30.7 45.6 102.9 52.9 33.7
Nparticles
Corresponding
Figure2 : Steps of developed population updating at iteration t. losses (MW)
57
 We note that the results obtained by Accelerated PSO are
better than those obtained by classical PSO.
B. Bi-objective optimization:
Fig. 4, 5 and 6showPareto front for bi-objective functions
of fuel cost/ NOx emissions, compensation fuel cost/ Active
losses and Emissions/ Active loss using accelerated MOPSO.
Tables 5, 6 and 7 give principal points from these Pareto
front.
Table 4 : Pareto front of fuel cost/ NOx emissions

fuel Emissi Generated power (MW)

cost ons
($/h) (kg/h) Pg1 Pg2 Pg3 Pg4 Pg5 Pg6

640 213 45.4 54.6 54.9 22.7 56.8 55.2

606 224 35.4 44.5 48.5 57.2 54.7 45.6
592 244 22.5 32.9 53.0 79.9 55.3 49.5

Table 5 : Pareto front of fuel cost/ Active losses

fuel Active Generated power (MW)

cost losses
($/h) (MW) Pg1 Pg2 Pg3 Pg4 Pg5 Pg6

620 1.87 9.0 27.3 92.6 63.7 84.8 5.2

596 10.1 22.5 27.3 54.6 109.7 56.3 19.6
588 44 13.8 32.8 46.9 100.2 56.0 27.7

Table 6 : Pareto front of Emissions/ Active loss

Emis Generated power (MW)
Active
Figure 4 : Single line diagram of IEEE 30-bus test system sions
losses
(kg/h
(MW) Pg1 Pg2 Pg3 Pg4 Pg5 Pg6
Minimal NOx emissions: )
Table 2 gives minimal NOx emissions with corresponding 261 1.8 13.8 21.9 93.6 55.8 91.5 8.6
consumed fuel cost and active losses. 220 10.4 49.4 59. 65.8 27.7 65.8 6.9
212 40.0 49.4 59.1 56.3 23.0 59.7 9.1
Table 2 : Minimal emissions
Minimal Generated power (MW) We note that, in all three cases, the optimization of a single
emissions(kg/h) function degrades the other two functions. Hence the
211 Pg1 Pg2 Pg3 Pg4 Pg5 Pg6 usefulness of a multiobjective optimization
Corresponding C. Three-objective optimization:
Fuel cost ($/h)
645 Fig. 8 shows Pareto surface for three-objective functions:
48.7 58.7 70.4 51.3 65.6 50
Corresponding fuel cost, NOx emissions and active losses using accelerated
active losses (MW) MOPSO.
67
Table. 7 gives principal points from this Pareto surface.
Minimal active losses: Table 7 : Pareto surface of fuel cost/ NOx emissions/ Active losses
Table 3 gives minimal active losses with corresponding fuel Emi- Generated power(MW)
Losses
consumed fuel cost and NOx emissions. cost ssions
(MW) Pg1 Pg2 Pg3 Pg4 Pg5 Pg6
($/h) (kg/h)
Table3 :Minimal active losses 588 271 38 10.1 32.9 43.5 102.9 55.3 54.9
Minimal active Generated power (MW) 639 212 50 45.4 54.6 54.9 22.7 56.8 54.1
losses (MW) 626 264 1.9 5.9 27.1 92.9 57.2 94.8 6.1
1.74 Pg1 Pg2 Pg3 Pg4 Pg5 Pg6
Corresponding The values in the table 8 illustrate the extreme points of the
Emissions Pareto surface but the dispatcher can choose other compromise
(kg/h) depending on the structure of a specified electrical network.
264 8.8 24.3 99.6 57.7 86.3 23.1
Corresponding
fuel cost ($)
625
 Figure 5 : Emissions/ consumed fuel cost

Figure 8 : Pareto surface of consumed fuel cost/emissions

Then, this approach has been applied to resolve multi-
objective economic/environmental power dispatch problem.
The results obtained by simulation show that the proposed
approach is efficient for solving this problem, and generally,
for solving multi-objective problems.
In future, we will develop this approach for causing it to
resolve stochastic multi-objective problems and we will apply
it to resolve multi-objective reactive power dispatch and multi-
objective load flow problems of electric network integrating
renewable energies.
REFERENCES

[1] Dian Palupi Rini, Siti Mariyam Shamsuddin and Siti Sophiyati Yuhaniz,
Figure 6 : Pareto front of active losses/consumed fuel cost “Particle Swarm Optimization: Technique, System and Challenges”,
International Journal of Computer Applications (0975 – 8887), pp. 19-
27, 2011.
[2] Margarita Reyes-Sierra and Carlos A. Coello Coello, “Multi-Objective
Particle Swarm Optimizers: A Survey of the State-of-the-Art”,
International Journal of Computational Intelligence Research.ISSN
0973-1873 Vol.2, No.3 pp. 287–308 , 2006.
[3] Praveen Kumar Tripathi, Sanghamitra Bandyopadhyay and Sankar
Kumar Pal, “Multi-Objective Particle Swarm Optimization with time
variant inertia and acceleration coefficients”, Information Sciences 177,
pp.5033–5049, 2007.
[4] Wen-Fung Leong and Gary G. Yen, “Dynamic Population Size in PSO-
based Multiobjective Optimization”Proceedings ofIEEE Congress on
Evolutionary, ComputationSheraton Vancouver Wall Centre Hotel,
Vancouver, BC, Canada, pp. 6182-6189, 2006.
[5] Leticia Cagnina, Susana Esquivel and Carlos A. Coello Coello, “A
Particle Swarm Optimizer for Multi-Objective Optimization”, JCS&T
Vol. 5 No. 4, pp.204-210, 2005.
[6] Carlos A. Coello Coello, “Handling Multiple Objectives With Particle
Swarm Optimization”, IEEE TRANSACTIONS ON EVOLUTIONARY
COMPUTATION, VOL. 8, NO. 3, pp.256-279, 2004.
[7] Xiaodong Li, “Better Spread and Convergence: Particle Swarm
Multiobjective Optimization Using the Maximin Fitness Function “,K.
Deb et al. (Eds.): GECCO 2004, LNCS 3102, pp. 117–128, 2004.
Figure 2 : Presto front of active losses/emissions
[8] M.S. Osmana, M.A. Abo-Sinnab and A.A. Mousab,”An ε-dominance-
VI. CONCLUSIONS AND FUTURE WORK based multiobjective genetic algorithm for economic emissionload
dispatch optimization problem”, Electric Power Systems Research 79,
In this paper, we have presented a novel multi-objective pp. 1561–1567, 2009.
approach using PSO algorithm and based on matrix [9] Jiejin Cai, Xiaoqian Ma, Qiong Li, Lixiang Li, Haipeng Peng,”A multi-
computation, called AMOPSO. objective chaotic particle swarm optimization for
environmental/economic dispatch”, Energy Conversion and
Management 50, pp. 1318–132, 2009.
[10] B.K. Panigrahi, V. Ravikumar Pandi, Renu Sharma , Swagatam Das and [11] V. Vahidinasab and S. Jadid, “Joint economic and emission dispatch in
Sanjoy Das, “Multiobjective bacteria foraging algorithm for electrical energy markets: A multiobjective mathematical programming
load dispatch problem”, Energy Conversion and Management 52, pp. approach”, Energy 35, pp. 1497–1504, 2010.
1334–1342, 2011.