constrained optimization using gravitational search algorithm
TRANSCRIPT
RESEARCH ARTICLE
Constrained Optimization Using Gravitational Search Algorithm
Anupam Yadav • Kusum Deep
Received: 12 January 2013 / Revised: 29 March 2013 / Accepted: 1 May 2013 / Published online: 8 October 2013
� The National Academy of Sciences, India 2013
Abstract Gravitational search algorithm is a newly born
metaheuristic which is inspired by the working force of
attraction between two masses. In this article gravitational
search algorithm is employed to solve the constrained
optimization problems. A variety of state-of-the-art
benchmark problems are taken into account to justify the
performance of the gravitational search algorithm. The
results of gravitational search algorithms are compared
with two state-of-the-art Particle Swarm Optimization
algorithms in various aspects. Pairwise one tailed t test is
applied to justify the statistical significance of the results
and time complexity analysis is also performed. Finally the
conclusions are drawn based on the experimental results.
keywords Gravitational search algorithm �Particle swarm optimization � Constrained optimization
Introduction
Constrained optimization problems are very significant to
the science and engineering community because many real
life and engineering design problems can be modeled as
constrained optimization problems. Mathematically a
constrained optimization problem can be modeled as
Maximize or Minimize f ðx1; x2; x3; . . .; xnÞ; ð1Þ
subject to inequality constraints
gjðx1; x2; x3; . . .; xnÞ 6 0; j ¼ 1; 2; 3; . . .q ð2Þ
and equality constraints
hjðx1; x2; x3; . . .; xnÞ ¼ 0; j ¼ qþ 1; qþ 2; . . .m ð3Þ
where f : Rn ! R is a real valued function where g and
h are constraints. These problems have much importance in
engineering and design for obtaining the optimal
parameters involve in the design and engineering. There
are two class of methods available for solving these
problems, the first one is deterministic techniques and the
other one is non deterministic techniques. The role of non
deterministic techniques gets much more importance due to
the limitations of deterministic techniques. The differential
free nature of non deterministic techniques stimulates its
importance in comparison to the deterministic techniques.
Nature inspired optimization techniques are recent
paradigm to the non deterministic optimization methods.
Genetic Algorithm, Differential Evolution, Particle Swarm
Optimization etc. are some examples of nature inspired
algorithms. In recent years Deep and Bansal [1] designed a
PSO for solving economic dispatch problem as a
constrained optimization problem. Deep et al. [2] applied
PSO for evaluating the hypocentral parameters of an
earthquake by utilizing some real life data. The promising
solution of these problems and the well applicability of
nature inspired algorithms on constrained optimization
problems directs to figure out the applicability of some
more nature inspired optimization techniques to the
constrained optimization problems. In the present article
a newly developed gravitational search algorithm (GSA) is
utilized to solve the constrained optimization problems.
GSA is proposed by Rashedi [3] and Rashedi et al. [4].
It is inspired by the Newton’s theory on force of attraction
A. Yadav (&) � K. Deep
Department of Mathemtics, Indian Institute of Technology
Roorkee, Roorkee, India
e-mail: [email protected]
K. Deep
e-mail: [email protected]
123
Natl. Acad. Sci. Lett. (September–October 2013) 36(5):527–534
DOI 10.1007/s40009-013-0165-8
between two masses. The proposed technique is success-
fully applied on unconstrained optimization problems [4]
as well as many real life applications [5, 6]. In this article
constrained optimization problems are tried to solve with
the help of GSA. A set of 24 state-of-the-art problems
presented in IEEE CEC 2006 [7] are solved with the help
of GSA. A parameter free constraint handling method is
employed to handle the constraints. The results of the GSA
is compared with two state-of-the-art variants of Particle
Swarm Optimization.
The organization of the paper is done in the following
way, In the Sect. 2 a brief idea of GSA is presented, Sect. 3
a brief idea of Particle Swarm Optimization is discussed. In
Sect. 4 the performance and evaluation of GSA is pre-
sented based on experiments over the benchmark functions,
finally the conclusions are drawn with future scope.
Gravitational Search Algorithm
GSA [3, 4] is a recent heuristic optimization algorithm. It is
inspired by Newton’s basic physical theory that a force of
attraction works between every particle in the universe and
this force is directly proportional to the product of their
masses and inversely proportional to the square of distance
between their positions. All these particles are named as
agents (particles). In GSA each particle is equipped with
four kind of properties: position, mass, active gravitational
mass and passive gravitational mass. The position of the
mass provides the solution of the problem. Gravitational
masses and inertial masses can be evaluated using fitness
function. Each kind of masses follows the following two
basic laws of physics:
1. Law of gravity: Each particle attracts every other
particle and the gravitational force between two
particles is directly proportional to the product of their
masses and inversely proportional to the square of the
distance between them.
2. Law of motion: The current velocity of any mass is
equal to the sum of the fractions of its previous
velocities and the variations of the velocity. Variation
in the velocity of acceleration of any mass is equal to
the force acted on the system divided by the mass of
inertia.
Inspired by the definitions above we are able to define
physics of GSA. Let the position of the ith particle at any
instant t in a D-dimensional search space be
Xtiðxt
i1; xti2; . . .; xt
iDÞ for i ¼ 1. . .ps. The force of attraction
on the ith particle to jth particle is defined as in the fol-
lowing equation
FtijD ¼ Gt �
Mtpi �Mt
aj
Rtij
� xtid � xt
jd
� �ð4Þ
where d ¼ 1; 2; . . .D;Mtpi is the passive gravitational mass
related to ith particle at time t, Mait is the active
gravitational mass related to jth particle at time t, Gt is
the gravitational constant at time t, � is a small constant and
Rijt is the Euclidian distance between the two particles i and
j given by the following equation:
Rtij ¼ kXt
i ;Xtjk2 ð5Þ
The value of gravitational constant Gt can be calculated as
Eq. 6
Gt ¼ Gt0 � exp �aiter
itermax
� �� �ð6Þ
where a and Gt0 are descending coefficient and initial value
respectively, iter is the current iteration and itermax is the
maximum number of iterations.
The total force of attraction exerted by the ith particle at
time t in a D-dimensional space is given by Eq. 7
Ftid ¼
Xps
i¼1;i 6¼j
randðÞFtijd ð7Þ
where d ¼ 1; 2; . . .D and rand() is a random number in the
interval [0,1], which is added to provide the stochastic
nature to the algorithm. By using the law of motion the
acceleration of ith particle is given by the following
equation:
actid ¼
Ftid
Mtii
ð8Þ
where Mtii is the inertial mass of the ith particle. The
velocity and position of particles are calculated as follow:
Vtþ1id ¼ randðÞ � Vt
id þ actid ð9Þ
xtþ1id ¼ xt
id þ Vtþ1id ð10Þ
where rand() is uniform random variable in [0,1]. The
gravitational and inertial masses are simply calculated by
the fitness evaluations. A greater mass can be treated as
better particle and having higher force of attraction so that
they can influence other particles with high level of
attraction. The gravitational and inertial mass will be
updated with the help of following equations:
Mai ¼ Mpi ¼ Mii for i ¼ 1; 2. . .ps ð11Þ
mti ¼
fitti � worstt
bestt � worsttð12Þ
Mti ¼
mtiPps
i¼1 mti
ð13Þ
528 A. Yadav, K. Deep
123
where fitit represents the fitness value of the ith particle at time t.
and bestt & worstt may be defined as in the following equations:
bestt ¼ minðfittjÞ; j 2 f1; . . .psg ð14Þ
worstt ¼ maxðfittjÞ; j 2 f1; . . .psg ð15Þ
The exhaustive procedure of GSA is explained in Table 1
Particle Swarm Optimization
Particle swarm optimization is a nature inspired stochastic
population based optimization search technique, inspired by
the social behavior of fish and birds. PSO was first introduced
by Kennedy and Eberhart. It uses the learning, information
sharing and position updating strategy of each solution.
Mathematically, PSO can be formulated as: Let the position of
the ith particle in a D-dimensional search space be
Xtiðxt
i1; xti2; . . .; xt
iDÞ with a flag of velocity Vti ðvt
i1; vti2; . . .; vt
idÞat any moment t, where i = 1 to ps, where ps is the swarm size.
Let Pbestit and Gbesti
t denote the latest best position of the
particle (personal best) and global best at the moment t. Initially
Pbestit and Xi
t are same. From the theory of the PSO the velocity
update equation and position update equation can be written as:
Vtþ1i ¼ c1Vt
i þ c2ðXti � Gbestt
iÞ þ c3ðXti � Pbestt
iÞ ð16Þ
Xtþ1i ¼ Xt
i þ Vtþ1i ð17Þ
The updated position of particles may be obtained by
applying Vit?1 on the position Xi
t (Eq. 17). The exhaustive
procedure of PSO is explained in Table 2
Performance and Evaluation of GSA
Test Functions
To judge the optimization ability of the GSA over con-
strained optimization problems, a set of state-of-the-art
problems from CEC 2006 [7] has been taken for the jus-
tification. A brief detail of the functions are listed in
Table 3. Few good constraint handling methods [8–11] are
available which may also utilize with the GSA. The fol-
lowing constraint handling method is applied to deal with
the constraint violations. The parameter-exempt constraint
dealing approach by computing constraint violation [7] is
used to deal with the constraints. The grade of constrained
violation [7] of single x for the jth constraint may be
computed as follows:
GjðxÞ ¼ maxfgjðxÞ; 0g1� j� q ð18Þ
GjðxÞ ¼ maxfjhjðxÞ � ej; 0gqþ 1� j�m; ð19Þwhere e is a positive tolerance esteem for equality con-
straints. The grade of constraints violation for an single
root x is G(x) =P
j=1m Gjx. To compare the results of the
GSA two popular variants of PSO is taken from the liter-
ature. There algorithms are
1. Basic PSO
2. SPSO 2011
The parameter setting of all the algorithms are assumed
same as cited in the concerned articles.
Table 1 Pseudo code of
gravitational search algorithm
Gravitational Search Algorithm 529
123
Results and Analysis
All the algorithms are coded on Matlab 2011b platform.
100 runs experiment are performed for each algorithm, a
run consists of maximum 4,000 iterations, the population
size for each algorithm is uniformly taken as 60. The
results of the performed experiments are presented in the
form of best, mean, worse and standard deviation of fitness
value along with the mean infeasibility of the particles in
the final population of a median run. It is recorded in the
results that the out of 24 problems GSA is able to solve 17
problems. GSA solved six problems for which the other
two algorithms are not able to solve. For the problems
g01, g02, g04, g06, g08, g11, g12, g13, g14, g15, g16, g18,
g19 and g24 the best recorded value is either better or
comparable to other two algorithms. The mean infeasibility
of almost all the problems by GSA is near to 0. Even for
few problems the worse recorded value is better than the
best recorded value of rest two algorithms. The results for
which fully infeasible solution is obtained is represented as
’NaN’. The results of the problems g20-g23 are not listed
in the Table 4, since none of the algorithms are able to
provide the feasible solution. Due to the very small ratio of
the feasible region and search region as discussed in the
Table 3, GSA is not able to reach towards the optimal
solution for few problems, since this very small ratio
increases the complexity of the problems and that is why
the working force of attraction among the particles become
less effective to move them towards global optima.
t Test
Pair wise one tailed t test is applied with 98� of freedom at
0.05 level of significance, over the fitness value of the
problems. This test involves the problems for which all the
algorithms are successful to give feasible result. Table 5
gives the results of the t test. The pairwise mean, standard
deviation(STD), Standard error mean, p value along with
conclusion of the test is listed. ’a-’, ’a’ and ’a?’ shows,
the Algo1 is significantly worse, alike and significantly
better to Algo2 respectively. It is observed that GSA per-
forms significantly better than other two algorithms for
more than seven problems.
Time Complexity
To compare the time complexity of the GSA in a median
run, its time consumption at key points are measured by in
built ’Matlab Profiler’, that determines the proportionality
of time consumption at key points. Table 6 shows the
comparative results of CPU time at key points of the
algorithm, for GSA these key process points are
Table 2 Pseudo code of
particle swarm optimization
algorithm
530 A. Yadav, K. Deep
123
Table 3 Characteristics of test problems
Prob. D Type q (%) LI NI LE NE a
g01 13 Quadratic 0.0111 9 0 0 0 6
g02 20 Nonlinear 99.9971 0 2 0 0 1
g03 10 Polynomial 0.0000 0 0 0 1 1
g04 5 Quadratic 52.1230 0 6 0 0 2
g05 4 Cubic 0.0000 2 0 0 3 3
g06 2 Cubic 0.0066 0 2 0 0 2
g07 10 Quadratic 0.0003 3 5 0 0 6
g08 2 Nonlinear 0.8560 0 2 0 0 0
g09 7 Polynomial 0.5121 0 4 0 0 2
g10 8 Linear 0.0010 3 3 0 0 6
g11 2 Quadratic 0.0000 0 0 0 1 1
g12 3 Quadratic 4.7713 0 1 0 0 0
g13 5 Nonlinear 0.0000 0 0 0 3 3
g14 10 Nonlinear 0.0000 0 0 3 0 3
g15 3 Quadratic 0.0000 0 0 1 1 2
g16 5 Nonlinear 0.0204 4 34 0 0 4
g17 6 Nonlinear 0.0000 0 0 0 4 4
g18 9 Quadratic 0.0000 0 13 0 0 6
g19 15 Nonlinear 33.4761 0 5 0 0 0
g20 24 Linear 0.0000 0 6 2 12 16
g21 7 Linear 0.0000 0 1 0 5 6
g22 22 Linear 0.0000 0 1 8 11 19
g23 9 Linear 0.0000 0 2 3 1 6
g24 2 Linear 79.6556 0 2 0 0 2
D is the number of decision variables, q = |F|/|S| is the estimated ratio between the feasible region and the search space, LI is the number of
linear inequality constraints, NI the number of nonlinear inequality constraints, LE is the number of linear equality constraints and NE is the
number of nonlinear equality constraints. a is the number of active constraints at x
Table 4 Comparative results of objective function values for 4,000 Iterations
Problem Algorithm Best Mean Worse STDEV Mean infeas.
g01 Basic PSO -1.213E?01 -1.173E?01 -1.127E?01 2.947E-01 1.00E-02
SPSO 2011 -1.294E?01 -1.131E?01 -9.157E?00 1.115E?00 1.00E-02
GSA -1.348E?01 -1.450E?02 -1.348E?01 5.038E?01 1.67E-04
g02 Basic PSO -3.896E-01 -3.475E-01 -3.088E-01 2.347E-02 1.00E-02
SPSO 2011 -4.715E-01 -4.263E-01 -3.756E-01 2.881E-02 1.00E-02
GSA -4.768E-01 -1.063E-01 -4.711E-02 4.247E-02 1.00E-02
g03 Basic PSO -5.393E-01 -2.411E-01 -6.085E-02 1.443E-01 7.73E-04
SPSO 2011 -7.998E-01 -5.990E-01 -4.398E-01 1.091E-01 2.87E-03
GSA -2.356E-02 -1.816E-02 -1.244E-03 4.085E?02 0.00E?00
g04 Basic PSO -3.065E?04 -3.063E?04 -3.062E?04 8.971E?00 1.00E-02
SPSO 2011 -3.057E?04 -3.042E?04 -3.031E?04 8.335E?01 1.00E-02
GSA -3.065E?04 -2.782E?04 -2.332E?04 2.146E?03 2.83E-03
g05 Basic PSO NaN NaN NaN NaN 0.00E?00
SPSO 2011 NaN NaN NaN NaN 7.60E-03
GSA NaN NaN NaN NaN 0.00E?00
g06 Basic PSO -6.933E?03 -6.897E?03 -6.855E?03 2.983E?01 1.00E-02
SPSO 2011 -6.892E?03 -5.936E?03 -4.667E?03 7.435E?02 1.00E-02
Gravitational Search Algorithm 531
123
initialization of particles, calculation of mass, calculation
of acceleration position update and fitness calculation. The
key process points of both the variants of PSO are initial-
ization of particles, initialization of velocity, velocity
update and fitness calculation. Two problems g01 and g02
are considered to compare the consuming time. It can be
observed from Table 6 that the time consumption in cal-
culation of acceleration and mass is less in comparison to
Table 4 continued
Problem Algorithm Best Mean Worse STDEV Mean infeas.
GSA -6.934E?03 2.264E?05 7.859E?05 2.082E?05 1.67E-04
g07 Basic PSO 8.204E?01 1.215E?02 1.732E?02 2.928E?01 9.79E-03
SPSO 2011 4.101E?01 1.080E?02 3.207E?02 9.673E?01 1.00E-02
GSA 4.286E?01 2.151E?02 4.652E?02 9.207E?01 1.67E-04
g08 Basic PSO -9.583E-02 -9.583E-02 -9.583E-02 3.440E-14 1.00E-02
SPSO 2011 -9.583E-02 -9.583E-02 -9.583E-02 1.600E-17 1.00E-02
GSA -9.583E-02 -9.440E?00 1.401E-01 1.704E?01 0.00E?00
g09 Basic PSO 7.112E?02 7.309E?02 7.470E?02 1.159E?01 1.00E-02
SPSO 2011 6.809E?02 6.892E?02 7.324E?02 1.515E?01 1.00E-02
GSA 7.043E?02 1.355E?06 9.020E?06 2.170E?06 1.67E-04
g10 Basic PSO 1.079E?04 1.190E?04 1.332E?04 8.307E?02 8.75E-03
SPSO 2011 7.702E?03 1.156E?04 1.606E?04 2.948E?03 1.00E-02
GSA NaN NaN NaN NaN 0.00E?00
g11 Basic PSO 7.509E-01 7.692E-01 8.265E-01 2.183E-02 1.70E-03
SPSO 2011 7.503E-01 7.747E-01 8.805E-01 4.423E-02 6.81E-03
GSA 7.452E-01 8.570E-01 1.253E?00 1.022E?00 1.67E-04
g12 Basic PSO -1.000E?00 -1.000E?00 -1.000E?00 1.150E-05 1.00E-02
SPSO 2011 -1.000E?00 -1.000E?00 -1.000E?00 0.000E?00 1.00E-02
GSA -1.000E?00 -1.000E?00 -1.000E?00 0.000E?00 1.00E-02
g13 Basic PSO NaN NaN NaN NaN 0.00E?00
SPSO 2011 NaN NaN NaN NaN 9.60E-03
GSA 9.978E-01 1.394E?06 7.659E?07 9.903E?06 1.67E-04
g14 Basic PSO NaN NaN NaN NaN 0.00E?00
SPSO 2011 NaN NaN NaN NaN 2.61E-03
GSA -4.330E?01 -1.039E?03 -4.330E?01 2.620E?02 1.67E-04
g15 Basic PSO NaN NaN NaN NaN 0.00E?00
SPSO 2011 9.618E?02 9.641E?02 9.699E?02 2.950E?00 1.00E-02
GSA 9.610E?02 8.124E?02 9.881E?02 1.039E?02 1.67E-04
g16 Basic PSO -1.894E?00 -1.886E?00 -1.880E?00 3.794E-03 1.00E-02
SPSO 2011 -1.800E?00 -1.633E?00 -1.457E?00 1.115E-01 1.00E-02
GSA -1.821E?00 3.379E-01 3.012E?00 1.096E?00 1.67E-04
g17 Basic PSO NaN NaN NaN NaN 0.00E?00
SPSO 2011 NaN NaN NaN NaN 2.80E-03
GSA NaN NaN NaN NaN 0.00E?00
g18 Basic PSO NaN NaN NaN NaN 0.00E?00
SPSO 2011 NaN NaN NaN NaN 5.60E-03
GSA -5.621E-01 1.452E?00 1.451E?02 5.861E?01 1.67E-04
g19 Basic PSO 1.163E?02 1.497E?02 1.785E?02 1.600E?01 1.00E-02
SPSO 2011 9.809E?01 1.319E?02 2.210E?02 3.238E?01 1.00E-02
GSA 5.170E?01 1.793E?03 3.433E?03 9.090E?02 2.83E-03
g24 Basic PSO -5.508E?00 -5.508E?00 -5.508E?00 5.270E-06 1.00E-02
SPSO 2011 -5.508E?00 -5.504E?00 -5.489E?00 6.173E-03 1.00E-02
GSA -5.508E?00 -3.757E?00 -3.628E-01 1.484E?00 5.00E-03
532 A. Yadav, K. Deep
123
the velocity initialization and velocity updation, so the total
consuming time for GSA to solve the problem g01 and g02
is less while others take little more time. This is a great
advantage of GSA, because while dealing with large scale
optimization problems the CPU time consumed by GSA
will be considerably less.
Conclusion
In the current article a new GSA is employed to solve the
constraint optimization problems. A set of stat-of-the-art
problems are solved using GSA which are taken from CEC
2006. 100 runs experiment is performed for maximum 4000
iterations. The results of GSA are compared with Basic PSO
and SPSO 2011 in the form of best, mean, worse and
standard deviation of the objective function values. The
statistical significance of the results are justified with the
one tailed t test. It is concluded that the performance of the
GSA is good for most of the problems and for few problems
its results are comparable with Basic PSO as well as SPSO
2011. While doing experimental analysis it is observed that
a more compatible constraint handling technique to GSA
may provide some better results to the existing.
Table 5 Pairwise t test results over objective function values with 95 % confidence interval at 0.05 level of significance
Problem Algo1 vs Algo2 Mean STD Std. error mean p value Conclusion
g01 GSA-Basic PSO 0.44 0.68 0.14 0.004 a
GSA-SPSO 2011 2.44 0.68 0.14 0.000 a?
g02 GSA-SPSO 2011 0.09 0.03 0.01 0.000 a?
GSA-SPSO 2011 0.42 0.03 0.01 0.000 a?
g04 GSA-Basic PSO 694.99 364.18 72.84 0.000 a?
GSA-SPSO 2011 781.66 157.88 31.58 0.000 a?
g07 GSA-Basic PSO 320.95 283.78 56.76 0.000 a?
GSA-SPSO 2011 353.07 283.49 56.70 0.000 a?
g09 GSA-Basic PSO 10.65 6.85 1.37 0.000 a?
GSA-SPSO 2011 24.92 6.30 1.26 0.000 a?
g12 GSA-Basic PSO 0.00 0.00 0.00 0.020 a-
GSA-SPSO 2011 0.00 0.00 0.00 0.021 a-
g18 GSA-Basic PSO -0.20 0.06 0.01 0.000 a?
GSA-SPSO 2011 0.24 0.07 0.01 0.000 a?
g19 GSA-Basic PSO 481.71 111.77 22.35 0.000 a?
GSA-SPSO 2011 526.02 110.26 22.05 0.000 a?
g24 GSA-SPSO 2011 0.00 0.00 0.00 0.000 a?
GSA-SPSO 2011 0.00 0.00 0.00 0.000 a?
Table 6 Comparision of CPU time at some key processing points of the algorithms
g01 Initializate X Initialization V V calculation Fitness Total time (s)
Basic PSO 0.01 0.01 0.2 3 3.22
SPSO 2011 0.1 0.02 0.02 3.1 3.24
Initializate X Mass Acceleration Fitness
GSA 0.01 0.001 0.003 2.9 2.914
g02 Initializate X Initialization V V calculation Fitness
Basic PSO 0.02 0.02 0.35 5 5.39
SPSO 2011 0.02 0.03 0.034 6 6.084
Initializate X Mass Acceleration Fitness
GSA 0.02 0.002 0.023 4 4.045
Gravitational Search Algorithm 533
123
Acknowledgments The first author is thankful to Council of Sci-
entific and Industrial Research, HRDG Group for the financial support
with grant No. 9834-11-44. I would also like to thanks to the anon-
ymous reviewers for their valuable suggestion and comments for
revising this article.
References
1. Deep K, Bansal JC (2012) Solving economic dispatch problems
with valve-point effects using particle swarm optimization.
J Univers Comput Sci 18(13):1842–1852
2. Deep K, Yadav A, Kumar S (2012) Improving local and
regional earthquake locations using an advance inversion
technique: Particle swarm optimization. World J Model Simul
8(2):135–141
3. Rashedi E (2007) Gravitational search algorithm. MSc Thesis,
Shahid Bahonar University of Kerman, Kerman (in Farsi)
4. Rashedi E, Nezamabadi-pour H, Saryazdi S (2009) Gsa: a grav-
itational search algorithm. Inf Sci 179(13):2232–2248
5. Li C, Zhou J (2011) Parameters identification of hydraulic turbine
governing system using improved gravitational search algorithm.
Energy Convers Manag 52(1):374–381
6. Rashedi E, Nezamabadi-Pour H, Saryazdi S (2011) Filter mod-
eling using gravitational search algorithm. Eng Appl Artif Intell
24(1):117–122
7. Liang J, Runarsson T, Mezura-Montes E, Clerc M, Suganthan P,
Coello C, Deb K (2006) Problem definitions and evaluation cri-
teria for the cec 2006 special session on constrained real-
parameter optimization. J Appl Mech 41:1–24
8. Coath G, Halgamuge S (2003) A comparison of constraint-han-
dling methods for the application of particle swarm optimization
to constrained nonlinear optimization problems. In: Evolutionary
Computation, 2003. CEC’03. The 2003 Congress on, IEEE, vol 4,
pp 2419–2425
9. Coello Coello C (2002) Theoretical and numerical constraint-han-
dling techniques used with evolutionary algorithms: a survey of the
state of the art. Comput Method Appl Mech Eng 191(11):1245–1287
10. Fuentes Cabrera J, Coello Coello C (2007) Handling constraints
in particle swarm optimization using a small population size. In:
Gelbukh A, Kuri Morales AF (eds) Advances in Artificial Intel-
ligence. Springer, Berlin, Heidleberg, pp 41–51
11. Hu X, Eberhart R (2002) Solving constrained nonlinear optimi-
zation problems with particle swarm optimization. In: Proceed-
ings of the sixth world multiconference on systemics, cybernetics
and informatics, Citeseer, vol 5, Springer, Berlin, Heidelberg,
pp 203–206
534 A. Yadav, K. Deep
123