Journal of Engineering and Applied Sciences 9 (10-12): 405-413, 2014

ISSN: 1816-949X
© Medwell Journals, 2014

An Efficient Artificial Bee Colony Algorithm for Constrained

Optimization Problems

Soudeh Babaeizadeh and Rohanin Ahmad

Department of Mathematical Sciences, Faculty of Science, Universiti Teknologi Malaysia,
UTM 81310, Johor, Malaysia

Abstract: Artificial Bee Colony (ABC) algorithm is a relatively new swarm intelligence algorithm which has
shown a competitive performance with respect to other population-based algorithms. However, there are still
some insufficiencies in ABC algorithm such as slow convergence and easily trapped in local optima. These
drawbacks can be even more challenging when constraints are also involved. To address these issues an
Efficient Constrained Artificial Bee Colony (eABC) algorithm is proposed where two new solution search
equations are introduced respectively to employed bee and onlooker bee phases. In addition, smart flight
operator is employed to be used in scout bee phase. This algorithm is tested on several constrained benchmark
problems. The numerical results demonstrate that the eABC is competitive with other state of the art
constrained ABC algorithms under consideration.

Key words: Artificial bee colony, swarm intelligence, constrained optimization, chaotic search, smart hight

INTRODUCTION of implementation, ABC has captured much attention from

researchers and it has been applied to solve many
Optimization methods are frequently applied to numerical as well as practical optimization problems
address various real world problems. However, finding (Gao et al., 2014; Aydin et al., 2014; Xiang and An, 2013;
optimal solutions for these problems are very challenging Li et al., 2012), since its invention.
using traditional optimization methods. Because there Among optimization problems, the ones tackled in
have always been many real world problems where this paper are Constrained Optimization Problems (COPs)
derivatives are unavailable or unreliable. Swarm for Nonlinear Programming (NLP) which can be formulated
Intelligence (SI) algorithms have shown considerable as in the following problem:
success in solving nonconvex, discontinuous, non-
differentiable optimization problems and attracted more min f (x)
attention in recent years. s.t g j (x) ≤ 0, j = 1, 2,..., m (1)
The most prominent Evolutionary Algorithms (EAs) h j (x) = 0, j = m+1,...,l
have been suggested in the literatures are Genetic Where:
Algorithm (GA) (Holland, 1975), Particle Swarm x = [x1, x2,...,
Optimization (PSO) (Kennedy, 2011), Ant Colony xD]0Rn = D-dimensional decision vector and each
Optimization (ACO) (Dorigo and Blum, 2005), Rn = Bounded in the interval [xmin, xmax]
Differential Evaluation (DE) (Storn and Price, 1997) and xmin = The lower bound
Artificial Bee Colony (ABC) algorithm (Karaboga, 2005) xmax = The upper bound
and so on. f(x) = The objective function defined in D-
ABC is a recently proposed SI algorithm which dimensional search space in Rn
simulates the foraging behavior of honey bee swarms
(Karaboga, 2005). Numerical performance demonstrate In general, most of the optimization problems have
that ABC algorithm is competitive to other been primarily designed to address unconstrained
population-based algorithms such as GA, PSO, DE with optimization problems. In order to solve constrained
an advantage of employing fewer control parameters and problem, constraint handling techniques are employed to
the need for fewer function evaluations to arrive at an direct the search towards the feasible regions of the
optimal solution (Karaboga and Basturk, 2007a, b, 2008; search space. Constraint handling methods are
Karaboga and Akay, 2009). Due to its simplicity and ease categorized into four groups by Koziel and Michalewicz

Corresponding Author: Rohanin Ahmad, Department of Mathematical Sciences, Faculty of Science,

Universiti Teknologi Malaysia, UTM 81310, Johor, Malaysia
405
 J. Eng. Applied Sci., 9 (10-12): 405-413, 2014

(1999): methods based on penalty functions which where, xi = (xi1, xi2,..., xiD) and i = 1, 2,..., SN, j = 1, 2,..., D and
penalize constraints to solve a constrained problem as an D is a number of optimization parameters, xmin, j and xmax, j
unconstrained problem, methods based on reservation of are the lower and upper bounds for the dimension j,
feasible solutions by transforming infeasible solutions to respectively.
feasible solutions with some operators, methods that After initialization, the population of solution is
separate feasible and infeasible solutions, other hybrid evaluated and then is repeated in a cycle of the employed
methods. bees, onlookers and scouts. Each employed bee generates
However, ABC algorithm similar other evolutionary a new food source in their neighborhood using Eq. 3:
algorithms faces up with some challenging problems
related with the solution search equation of ABC. This
v i, j = x i, j +fi, j(x i, j -x k, j ) (3)
method has good exploration but poor exploitation
(Karaboga, 2005) which results in the poor convergence.
In this sudy, a new constrained ABC algorithm is where, k0{1, 2, ..., SN} and j0{1, 2, ...D} are randomly
proposed by employing two new solution search chosen indexes, k has to be different from i, Nij is a random
mechanisms for employed bee and onlooker bee phases number in the range [-1, 1]. Once vi = {vi1, vi2, ..., viD} is
respectively. Furthermore, smart flight operator is obtained, it will be evaluated and compared with xi using
employed into scout bee phase instead of purely random greedy selection mechanism. If the fitness of vi is better
generation of solutions in general ABC. than fitness value of xi, the vi will be replaced with x i and
xi will be removed, otherwise xi is retained in population.
MATERIALS AND METHODS After all employed bees complete their searches, they
share their information about fitness and position of
Artificial bee colony: ABC algorithm is a recently
solutions with the onlooker bees. An onlooker bee
introduced population-based method by Karaboga (2005)
chooses a solution using probability value associate with
which simulate the foraging behavior of honey bee
the solution where pi is defined as follows:
colonies. In ABC, the colony of artificial bees is classified
into three groups, employed bees, onlooker bees and
fiti
scout bees. Half of the colony includes employed bees pi = SN (4)
å fit
and the other half consists of onlooker bees. Employed
bees search around the food source and gathering i
j=1
required information. Then, they carry the information
about the position of food source back to the hive and
share this information with onlooker bees. where, fiti is the fitness value of solution i. Obviously, the
Onlooker bees choose food source with better quality higher the value of fiti has more probability that the ith
from those found by employed bees using probability solution is selected. Then as in the case of employed bee
selection mechanism as a proportional of the quality of a new solution is generated using Eq. 3. If a new solution
food source. Therefore, the food sources with good has better quality than the old solution, the old solution
quality attract more onlooker bees compared to food is replaced with new solution otherwise the old solution
source with lower quality. If the quality of the food source remained in the population.
is not improved through a predetermined number of If a solution cannot be improved further through a
iteration, the food source will be abandoned by its predetermined number of trials (limit) the solution is
employed bee and employed bee becomes a scout and assumed to be abandoned and the corresponding
starts to search for a new food source randomly in the employed bee becomes a scout. The scout produces a
neighborhood of the hive. Through, the search process, solution randomly using Eq. 2. The detailed pseudo code
scout bees are responsible for exploration while of original ABC algorithm is presented in the Algorithm 1.
exploitation is done using employed and onlooker bees.
In ABC, the position of a food source represents a Algorithm 1; Original artificial bee colony algorithm:
possible solution to the problem and the nectar amount of Initialize the population of solution
Evaluate the initial population
each food source corresponds to their fitness of the cycle = 1
related solution. The number of employed bees or the Repeat
onlooker bees is equal to the number of solutions SN in Employed bee phase
the population. At initialization step, a population of SN Apply greedy selection process
Calculate the probability values
solutions are randomly generated using the following Onlooker bee phase
equation: Scout bee phase
Memorize the best solution achieved so far
cycle = cycle+1
xi, j = x min, j +rand(0,1)(x max, j − x min, j ) (2) until cycle = maximum cycle number

406
 J. Eng. Applied Sci., 9 (10-12): 405-413, 2014

Previous work on constrained artificial bee colony: ABC probability selection mechanism is presented to enhance
algorithm has been originally introduced to address diversity by allowing infeasible solutions in the
unconstrained optimization problems (Karaboga, 2005). population where infeasible solutions are introduced
Then, this method is adapted to deal with constrained inversely proportional to their constraint violations and
optimization problems. The presence of various feasible solution defined based on their fitness values. To
constraints and interferences between them makes COPs recognize this algorithm through this paper the
more challenging than unconstrained optimization abbreviation MABC will be used.
problems. In this study, we briefly present the available Another modified constrained ABC was developed
constrained ABC algorithms in the literature. by Subotic (2011) where Multiple Onlooker Bees (MO-
The first attempt to apply ABC algorithm to solve ABC) are applied into original constrained ABC. In this
COPs is done by Karaboga and Bastruck (Karaboga and algorithm three trial solutions are applied to form a new
Basturk, 2007a, b). To cope with constraints, Deb (2000)’s solution. The numerical performance of the algorithm
mechanism is employed to be used instead of the greedy when compared with the original ABC shows comparative
selection mechanism due to its simplicity, computational results.
cost and fine tuning requirement over other constraint Mezura-Montes and Cetina-Dominguez presented a
handling methods. Because initialization with feasible Modified ABC (M-ABC). This algorithm consist of four
solutions is very time consuming and in some situation, modifications on the selection mechanism, the equality
impossible to generate a feasible solution randomly, the and boundary constraints and scout bee operators
constrained ABC algorithm does not consider the initial compared to the original constrained ABC. The
population to be feasible. As alternative Deb’s rules are mechanisms to handle equality and boundary constraints
employed to direct the solutions to feasible region of are enhanced with the aim to support a more appropriate
search space. In addition, the scout bee phase of the approach to the feasible region of the search space. A
algorithm provides a diversity mechanism that allows new binary tournament selection based on feasibility is
and probably infeasible individuals to be in the supplanted with the fitness selection of solutions applied
population. Scouts are generated at a predetermined in the original ABC. In addition, smart flight operator is
period of cycles for discovering new solution randomly. employed to be used in scout bee instead of the uniformly
This period is another control parameter called Scout random approach in constrained ABC (Karaboga and
Production Period (SPP). At each SPP cycle, it is Basturk, 2007a, b). The numerical results show that
controlled if there is an abandoned solution or not. If M-ABC provides comparable results with respect to
there is a scout production process is executed. The algorithms under comparison (Mezura-Montes and
numerical performance of the proposed ABC algorithm is Cetina-Dominguez, 2012).
evaluated and compared with constrained PSO and DE A Genetic Inspired ABC algorithm (GI-ABC) was
algorithms and results show that ABC algorithm can be introduced to adopt GA in the process of replacement of
effectively applied for solving constrained optimization exhausted solutions (Bacanin and Tuba, 2012). In this
problems. algorithm, uniform crossover and mutation operators from
Mezura-Montes et al. (2010) presented Smart Flight GAs are applied to improve the performance of ABC
ABC (SF-ABC) algorithm to improve the performance of algorithm.
constrained ABC where smart flight operator is applied in Stanarevic, Tuba and Bacanin (Stanarevic et al., 2011)
scout bee phase to direct search towards promising suggested Smart Bee ABC algorithm (SB-ABC) to solve
region of the search space. Therefore if the best solution constrained problems. In this algorithm, smart bee is used
is infeasible, the trial solution has the chance to be to memorize the solutions and their fitness. Then, the best
located near the boundaries of the feasible region of solution is replaced with a new random solution if the new
search space. However if the best solution is infeasible, solution is unfeasible or if the new solution is feasible but
the smart flight will generate a solution in promising it does not have better fitness. The numerical experiments
region of search space. In addition, the combinations of show efficiency of the method.
two dynamic tolerances are also applied into SF-ABC to ABC-BA is a hybrid algorithm presented by
transform the original COP into unconstrained Tsai (2014) that integrates ABC and Bee Algorithm (BA)
optimization. The numerical results demonstrate the to solve COPs. In this algorithm, individuals can perform
competitive performance of SF-ABC with constrained as an ABC individual in ABC sub-swarm or a BA
ABC (Karaboga and Basturk, 2007a, b). individual in the BA sub-swarm. In addition, the
A modified ABC was introduced by Karaboga and population size of the ABC and BA sub-swarms change
Akay (2011) to solve COPs. In this algorithm, a new stochastically based on current best fitness values

407
 J. Eng. Applied Sci., 9 (10-12): 405-413, 2014

achieved by the sub-swarms. Experimental results Equation 5 it is obvious that when the difference
demonstrate that ABC-BA outperforms ABC and BA between the parameters xij and xrj decreases, the
algorithms, respectively. perturbation on the position xij decreases. Therefore as
Constrained ABC algorithm was also applied to solve the search moves toward the optimum solution in the
many real-world engineering problems in recent years. search space, the step size is adaptively reduced.
Brajevic et al. (2011) proposed a Constrained Artificial After generating a new solution using Eq. 5 Deb’s
Bee Colony (SC-ABC) and applied on several standard ruels are applied in selection process to direct the
engineering benchmark problems of discrete and individuals to feasible region of search space. Using
continuous variables. The numerical results then were Deb’s mechanism, either the new solution is memorized
compared to results obtained from Simple Constrained and the current solution is removed or the current
Particle Swarm Optimization algorithm (SiC-PSO) which solution is remand.
show a very good performance. Akay and Karaboga
Algorithm 2; Employed bee phase for eABC:
(2012) used ABC to solve large scale optimization for I = 1: SN
problems as well as engineering design problems. The for j = 1: D
Produce a new solution ui using Eq. 5
numerical results show that the performance of ABC end for
algorithm is comparable to those of state of the art If no parameter is changed, choose a parameter randomly and change it form
algorithms under comparison. Upgraded Artificial Bee solution xi using Eq. 5.
Evaluate the solution vi
Colony (UABC) algorithm was also introduced for Apply the selection process between vi and xi based on Deb’s Method
constrained optimization problems by Brajevic and If solution xi does not improve triali = traili+1, otherwise traili = 0
Tuba (2013) to improve fine-tuning features of the end if

modification rate parameter and applying modified

Deb’s Method utilizes a tournament selection
scout bee phase of the ABC algorithm. This algorithm
mechanism where two solutions are compared using
was then tested on several engineering benchmark
following rules:
problems and the performance was compared with the
performance of the Akay and Karaboga (2012) algorithm.
C Feasible solution is preferred to infeasible solution
The numerical results show that the UABC produces
C Among two feasible solutions, the one having better
better results.
objective function value is preferred
C Among one feasible and one infeasible solution, the
Efficient constrained artificial bee colony: In this study, one having smaller constant violation is preferred
we proposed an efficient constrained Artificial Bee
Colony (eABC) algorithm. In the first step of this The framework of employed bee phase is given in
algorithm, initial population of SN solutions is generated Algorithm 2. After all employed bees complete their
randomly using Eq. 2. After initialization, the population searches, they share their information related to the
is evaluated and a cycle of the search procedures of the fitness values and the positions of their solutions with the
employed bees, the onlooker bees and scout bees is onlooker bees. An onlooker bee chooses a solution using
repeated. a probability mechanism as follows:
A new solution search equation is proposed for
employed bee phase using Eq. 5:  fitnessi
0.5+( SN )×0.5 if solution is feasible
 vij = x rj +γ ij (x ij -x rj ) if R j <MR


∑ fitness j
vij =  (5) pi = 
j=1
(6)
 x ij otherwise
 (1- violation i )×0.5 if solution i s infeasible
 SN
 ∑ violation j
Where:
r = Randomly chosen index has to be different  j=1
form
i and (i, j = A random number between [-1, 1] where, violationi is the constraint violation of solution xi
Rj = Uniformly distributed random number in the and finessi is the fitness value of the solution xi. Based on
range [-1, 1] this equation infeasible solutions are allowed to consider
MR = A control parameter which controls the in population as well as feasible solutions. The fitness
number of parameters to be modified value also is defined in Eq. 7:

408
 J. Eng. Applied Sci., 9 (10-12): 405-413, 2014

 1/(1+f i ) if fi ≥ 0 (7) v j = x ij + jij (x rj - x ij ) + (1 - jij )(x best, j - x ij ) (9)

fitnessi = 
1+abs(f i ) if fi < 0
where, Nij is random number in [-1,1] and r is random index
where, fi is the objective function of solution i. An that have to be different from i, xbest is the best solution
onlooker bee evaluates the information shared by found so far.
employed bees and selects a solution with a probability
Algorithm 3; Onlooker bee phase for eABC:
associated with its nectar amount. After solution t = 0, i = 1
selection, onlooker bees produce modification on the Repeat
position of the selected solution by taking advantage of if random<pi then
t = t+1
the global best and random solution to guide the for j = 1: D
candidate solution toward promising region of search Produce a new solution vi using Eq. 8
space using Eq. 8: end for
If no parameter is changed, choose a parameter randomly and change it form
solution xi using Eq. 8.
 x ij +φij (x best , j − x r1, j )+ κi, jr and(x r1, j -x r2,j ) if R j < MR Evaluate the solution vi
vj =  Apply the selection process between vi and xi based on Deb’s Method
 x ij otherwise If solution xi does not improve traili = traili+1, otherwise traili = 0
end if
(8) i = i+1
where, xr1, j and xr2, j are uniformly random solution and i = i mod (SN+1)
until t = SN
j = {1, 2, ..., D}, Ni, j and ki, j are random number in
the range [-1, 1] and [0.3, 0.6], respectively. Similar with Numerical experiments and comparisons: In order to
employed bee phase after generating new solution using evaluate the performance of ABC algorithm and
Eq. 8, the new solution is compared with current solution show the efficiency and superiority of the
using Deb’s rules. If the new solution has better equality proposed algorithm, 24 well-known benchmark
it will remained in the population and the current solution problems form CEC2006 (Liang et al., 2006) are
removed otherwise the current solution is remained. The applied.
framework of onlooker bee phase is given in The proposed algorithm is evaluated and compared
Algorithm 3. with five state of the art constrained ABC algorithms. The
After distribution of all onlooker bees if a solution eABC algorithms as well as other algorithms in
can not improve further through predetermined number of comparison are coded in MALAB environment. Each
cycles (limit) it is abandoned and replaced with a new problem runs 30 times and statistical results are provided
solution discovered by scout bees. The scout produces including the best, median, mean, worst results and
a new solution using smart flight operator defined in the standard deviation which can be seen in Table 1
Eq. 9: and 2.

Table 1: The Numerical results obtained by ABC, MABC, M-ABC, MO-ABC and eABC
Problems Parameters ABC MABC M-ABC SF-ABC MO-ABC eABC
g01 Best -1.5000000 -1.5000000 -15.000000 -15.0000000 -15.000000 -1.5.00000
Mean -1.5000000 -1.5000000 -15.000000 -14.163245 -15.000000 -1.500000
Worst -1.5000000 -1.5000000 -15.000000 -12.525128 -15.000000 -1.500000
SD 0.0000000 0.0000000 0.000000 0.92321 0.000000 0.000000
g02 Best 0.8035669 0.8035383 0.803614 -0.708944 -0.803610 -0.8036150
Mean -0.7917445 -0.792927 -0.799450 -0.471249 -0.793510 -0.8021544
Worst -0.7529237 -0.750302 -0.778176 -0.319535 -0.74458 -0.7990689
SD 0.013292 0.011051 -0.006440 0.010823 0.016310 0.001263
g03 Best -1.004657 -1.004817 -1.000 -1.000 -1.000 -1.00409
Mean -1.000096 -1.001941 -1.000 -1.000 -1.000 -1.00313
Worst -0.979659 -0.989160 -1.000 -1.000 -1.000 -1.00104
SD 0.00597911 0.0003752 0.000 0.000 0.000 0.001336
g04 Best -30665.542 -3066.542 -30665.5 -30665.539 -30665.539 -30665.54
Mean -30665.542 -3066.542 -30665.539 -30665.539 -30665.539 -30665.54
Worst -30665.542 -3066.542 -30665.539 -30665.539 -30665.539 -30665.54
SD 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
g05 Best 5126.489 5127.099 5126.734 5126.506 5126.657 5126.394
Mean 5177.239 5263.991 5178.178 5126.527 5162.506 5299.670
Worst 5307.988 5802.318 5317.183 5126.859 5229.119 5968.589
SD 57.86021 156.0343 56.0001 0.079343 47.8234 248.4226
g06 Best -6961.814 -6961.814 -6961.814 -6961.814 -6961.814 -6961.814
Mean -6961.814 -6961.814 -6961.814 -6961.813 -6961.813 -6961.814

409
 J. Eng. Applied Sci., 9 (10-12): 405-413, 2014

Table 1: Continue
Problems Parameters ABC MABC M-ABC SF-ABC MO-ABC eABC
Worst -6961.814 -6961.814 -6961.814 -6961.805 -6961.804 -6961.814
SD 0.0000000 0.0000000 0.000000 0.0002 0.0001 0.000000
g07 Best 24.46138 24.47032 24.3121 24.16452 24.32317 24.55848
Mean 24.70718 24.68698 24.41643 24.65842 24.45625 24.80918
Worst 25.16577 25.36005 24.794131 25.55104 24.92918 25.10102
SD 0.1813943 0.1786124 0.127124 0.326125 0.135021 0.1286075
g08 Best -0.095825 -0.095825 -0.095825 -0.095825 -0.095825 -0.0958250
Mean -0.095825 -0.095825 -0.095825 -0.095825 -0.095825 -0.0958250
Worst -0.095825 -0.095825 -0.095825 -0.095825 -0.095825 -0.0958250
SD 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
g09 Best 680.6381 680.6371 680.6331 680.6321 680.6312 680.6387
Mean 680.6506 680.6515 680.6472 680.6450 680.6353 680.6512
Worst 680.6757 680.6760 680.6763 680.8582 680.6362 680.6776
SD 0.008074 0.009610 0.054320 0.041213 0.004123 885.1586
g10 Best 7160.631 7220.554 7051.775 7049.517 7053.32 7304.817
Mean 7364.940 7347.843 7233.810 7116.824 7167.801 7445.860
Worst 7691.303 7924.128 7604.129 7362.741 7418.334 7647.175
SD 129.8405 134.1410 132.1284 82.12413 83.00823 87.75224
g11 Best 0.749000 0.749000 0.750000 0.750000 0.750000 0.7490001
Mean 0.749002 0.749003 0.750000 0.750000 0.750000 0.7490035
Worst 0.749010 0.749014 0.750000 0.750000 0.750000 0.7490247
SD 0.000002 0.000003 0.000000 0.000000 0.000000 0.0000005
g12 Best -1.000000 -1.000000 -1.000000 -1.000000 -1.000000 -1.000000
Mean -1.000000 -1.000000 -1.000000 -1.000000 -1.000000 -1.000000
Worst -1.000000 -1.000000 -1.000000 -1.000000 -1.000000 -1.000000
SD 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
g13 Best 0.555124 0.489597 0.05389 0.053986 0.454213 0.491074
Mean 0.949781 0.957689 0.157791 0.263854 0.456375 0.943945
Worst 1.492954 1.437534 0.441978 1.000000 0.489036 1.325281
SD 0.146915 0.161358 0.017232 0.216234 0.021103 0.171501

Table 2: The Numerical results obtained by ABC, MABC, M-ABC, MO-ABC and eABC
Problems Parameters ABC MABC M-ABC SF-ABC MO-ABC eABC
g14 Best -45.11878 -45.32082 -47.64541 -46.6651370 -46.450835 -45.97169
Mean -42.68215 -42.65421 -47.27156 -46.468243 -45.998013 -42.20681
Worst -40.60165 -40.05962 -46.53698 -43.87123 -45.316798 -39.11480
SD 1.171236 1.195831 0.245761 0.520124 0.257124 1.461280
g15 Best 941.2191 951.4375 961.7152 961.7151 961.7151 940.1215
Mean 958.8476 960.8922 961.7188 961.7155 961.8831 957.7468
Worst 972.9578 970.6846 961.7912 961.7201 964.3398 970.6761
SD 7.512742 4.87894 0.014319 0.1592 0.542672 7.754218
g16 Best -1.905155 -1.905155 -1.905155 -1.905155 -1.905155 -1.905155
Mean -1.905155 -1.905155 -1.905155 -1.905155 -1.905155 -1.905155
Worst -1.905155 -1.905155 -1.905155 -1.905155 -1.905155 -1.905155
SD 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000
g17 Best 8886.685 8879.576 8866.599 8927.598 8939.125 8871.616
Mean 9053.597 9053.567 8987.459 8928.865 8946.134 9052.462
Worst 9249.174 9215.365 9165.2543 8938.617 8956.127 9.259310
SD 123.0898 122.6397 95.6532 3.1213 9.52825 1.320255
g18 Best -0.840568 -0.859365 -0.866023 -0.866025 -0.865976 -0.842418
Mean -0.840573 -0.710702 -0.795019 -0.740748 -0.767198 -0.741887
Worst -0.0508290 0.0677663 0.093789 0.145231 0.096120 0.061626
SD 0.0508290 0.0677663 0.093789 0.145231 0.096120 0.061626
g19 Best 36.77401 37.58086 33.2547 32.66271 33.76983 37.08395
Mean 39.29784 39.83492 34.2656 33.10714 35.31478 39.70235
Worst 42.70161 42.42735 35.7368 34.91401 37.36458 42.47243
SD 1.457124 1.174349 0.63124 0.51321 0.687514 1.363563
g23 Best - - -159.754 -350.126 - -704.385
Mean - - -35.2847 -121.375 - -221.191
Worst - - 109.1275 276.0038 - 57.88116
SD - - 82.76981 157.8952 - 196.7631
g24 Best -5.508013 -5.508013 -5.508013 -5.508013 -5.508013 -5.508013
Mean -5.508013 -5.508013 -5.508013 -5.508013 -5.508013 -5.508013
Worst -5.508013 -5.508013 -5.508013 -5.508013 -5.508013 -5.508013
SD 0.000000 0.000000 0.000000 0.000000 0.000000 0.000000

Benchmark test problems and parameter settings: The functions (linear, nonlinear, polynomial, quadratic and
main characteristics of 24 benchmark functions are shown cubic) with different numbers of decision variables,
in Table 3. Table 3 describes various kinds of these test different types (linear inequalities, linear equalities,

410
 J. Eng. Applied Sci., 9 (10-12): 405-413, 2014

nonlinear inequalities and nonlinear equalities) and is the number of linear inequality constraints, NI is the
numbers of constraints. Table 1 D is the estimated ratio number of nonlinear inequality constraints, LE is the
between the feasible region and the search space, LI number of linear equality constraints, NE is the number of
nonlinear equality constraints, a is the number of
Table 3: The main characteristics of the test problems constraints active at the optimal solution and n is the
Function Type n D LI NI LE NE a number of variables of the problem. However as all the
g01 Quadratic 13 0.0111 9 0 0 0 6 algorithms considered in comparison were not able to
g02 Nonlinear 20 99.9971 1 1 0 0 1 obtain feasible solutions for g20-22 we exclude these
g03 polynomial 10 0.0000 0 0 0 1 1 problems from our experiments. In addition, the value of
g04 Quadratic 5 52.1230 0 6 0 0 2
g05 Cubic 4 0.0000 2 2 0 3 3
each parameters used are given in Table 4.
g06 Cubic 2 0.0066 0 5 0 0 2
g07 Quadratic 10 0.0003 3 2 0 0 6 RESULTS AND DISCUSSION
g08 Nonlinear 2 0.8560 0 4 0 0 0
g09 polynomial 7 0.5121 0 3 0 0 2
The numerical performance of eABC algorithm was
g10 linear 8 0.0010 3 0 0 0 3
g11 polynomial 2 0.0000 0 1 0 1 1 compared with original ABC (Karaboga and Basturk,
g12 Quadratic 3 4.7713 0 1 0 0 0 2007a, b), MABC (Karaboga and Akay, 2011), M-ABC
g13 Quadratic 5 0.0000 0 0 0 3 3 (Mezura-Montes and Cetina-Dominguez, 2012), SF-ABC
g14 Nonlinear 10 0.0000 0 0 3 0 3
(Mezura-Montes et al., 2010) and MO-ABC (Subotic,
g15 Quadratic 3 0.0000 0 0 1 1 2
g16 Nonlinear 5 0.0204 4 34 0 0 4 2011) algorithms. From Table 1 and 2 it is obvious that the
g17 Nonlinear 6 0.0000 0 0 0 4 4 eABC algorithm in problems g02, g03, g04, g08, g11, g23
g18 Quadratic 9 0.0000 0 13 0 0 6 outperforms compare with other algorithms. For g05, g10,
g19 Nonlinear 15 33.4761 0 5 0 0 0
g20 linear 24 0.0000 0 6 2 12 16
g15, g17, g19 the performance of SF-ABC was superior to
g21 linear 7 0.0000 0 1 0 5 6 all other algorithms. However, MO-ABC is superior in
g22 linear 22 0.0000 0 1 8 11 19 problems g09 and g14. In problem g01, g06, g12, g16 and
g23 linear 9 0.0000 0 2 3 1 6 g24 all the algorithm can find the optimal results. The
g24 linear 2 79.6556 0 2 0 0 2
numerical performance showed that eABC provided
comparable result with respect to other state of the art
Table 4: Parameters setting
Parameters Symbols Values
algorithms in solving COPs. In order to compare the
Solutions number SN 20 convergence ability of eABC with the other state of the
Maximum cycle number MCN 6000 art algorithms, three sample plots are presented in Fig. 1-3
Modification rate MR 0.8
which clearly show that eABC was able to converge faster
Population size PS 40
Limit Limit 150 than other algorithms which confirms that the new search
Scout production period SPP 150 equations can accelerate the constrained ABC
Epsilon g 0.001 convergence.

-0.1

-0.2

-0.3

-0.4
The value of functions

-0.5

-0.6
ABC
MABC
0.7 M-ABC
SF-ABC
-0.8 MO-ABC
eABC

-0.9
100 101 102 103 104
No. of function evaluations

Fig. 1: Iterations to convergence for problem g02

411
 J. Eng. Applied Sci., 9 (10-12): 405-413, 2014

0.92

0.90
0.88

The value of function

0.86

0.84
0.82
ABC
0.80
MABC
0.78 M-ABC
SF-ABC
0.76 MO-ABC
eABC
0.74
100 101 102 103 104
No. of function evaluations

Fig. 2: Iterations to convergence for problem g11

600
ABC
400 MABC
M-ABC
200 SF-ABC
The value of funcion

MO-ABC
0 eABC

-200

-400

-600

-800

-1000
100 101 102 103 104
No. of function evaluations

Fig. 3: Iterations to convergence for problem g23

CONCLUSION REFERENCES

In this study, we have introduced a modified Akay, B. and D. Karaboga, 2012. Artificial bee colony
constrained ABC called eABC algorithm to solve algorithm for large-scale problems and engineering
constrained optimization problems in which two new design optimization. J. Intell. Manuf., 23: 1001-1014.
search equations proposed for employed bee and Aydin, D., S. Ozyon, C. Yasar and T. Liao, 2014. Artificial
onlooker bee phases to enhance the global convergence bee colony algorithm with dynamic population size to
of ABC algorithm to solve COPs. In addition, smart flight combined economic and emission dispatch problem.
mechanism was applied to generate new solution in scout Int. J. Electr. Power Energy Syst., 54: 144-153.
bee phase. The experimental results were tested Bacanin, N. and M. Tuba, 2012. Artificial Bee Colony
on 24 benchmark functions and show that eABC is
(ABC) algorithm for constrained optimization
competitive with state of the art constrained ABC
improved with genetic operators. Stud. Inform.
algorithms. The Future researches include testing other
Control, 21: 137-146.
constraint handling mechanisms and using local search
operators to improve ABC algorithm further. Brajevic, I. and M. Tuba, 2013. An upgraded Artificial Bee
Colony (ABC) algorithm for constrained optimization
ACKNOWLEDGEMENT problems. J. Intell. Manuf., 24: 729-740.
Brajevic, I., M. Tuba and M. Subotic, 2011. Performance of
The researchers would like to thank Universiti the improved artificial bee colony algorithm on
Teknologi Malaysia for the financial funding through the standard engineering constrained problems. Int. J.
grant RUG 08H47. Math. Comput. Simul., 5: 135-143.

412
 J. Eng. Applied Sci., 9 (10-12): 405-413, 2014

Deb, K., 2000. An efficient constraint handling method for Koziel, S. and Z. Michalewicz, 1999. Evolutionary
genetic algorithms. Comput. Methods Applied Mech. algorithms, homomorphous mappings and
Eng., 186: 311-338. constrained parameter optimization. Evol. Comput.,
Dorigo, M. and C. Blum, 2005. Ant colony optimization 7: 19-44.
theory: A survey. Theor. Comput. Sci., 344: 243-278. Li, G., P. Niu and X. Xiao, 2012. Development and
Gao, W.F., S.Y. Liu and L.L. Huang, 2014. Enhancing investigation of efficient artificial bee colony
artificial bee colony algorithm using more algorithm for numerical function optimization.
information-based search equations. Inform. Sci., Applied Soft Comput., 12: 320-332.
270: 112-133. Liang, J.J., T.P. Runarsson, E. Mezura-Montes, M. Clerc,
Holland, J.H., 1975. Adaptation in Natural and Artificial
N. Suganthan, C.A.C. Coello and K. Deb, 2006.
Systems: Introductory Analysis with Applications to
Problem definitions and evaluation criteria for the
Biology, Control and Artificial Intelligence.
CEC 2006 special session on constrained real-
University of Michigan Press, Ann Arbor, MI., USA.,
parameter optimization. J. Applied Mech., 41: 2-24.
ISBN: 9780472084609, Pages: 183.
Karaboga, D. and B. Akay, 2009. A comparative study of Mezura-Montes, E. and O. Cetina-Dominguez, 2012.
artificial bee colony algorithm. Applied Math. Empirical analysis of a modified artificial bee colony
Comput., 214: 108-132. for constrained numerical optimization. Applied
Karaboga, D. and B. Akay, 2011. A modified Artificial Bee Math. Comput., 218: 10943-10973.
Colony (ABC) algorithm for constrained optimization Mezura-Montes, E., M. Damian-Araoz and O. Cetina-
problems. Applied Soft Comput., 11: 3021-3031. Domingez, 2010. Smart flight and dynamic tolerances
Karaboga, D. and B. Basturk, 2007a. A powerful and in the artificial bee colony for constrained
efficient algorithm for numerical function optimization. Proceedings of the IEEE Congress on
optimization: Artificial Bee Colony (ABC) algorithm. Evolutionary Computation, July 18-23, 2010,
J. Global Optim., 39: 459-471. Barcelona, pp: 1-8.
Karaboga, D. and B. Basturk, 2007b. Artificial Bee Colony Stanarevic, N., M. Tuba and N. Bacanin, 2011. Modified
(ABC) optimization algorithm for solving constrained artificial bee colony algorithm for constrained
optimization problem. Proceedings of the 12th problems optimization. Int. J. Math. Models Methods
International Fuzzy Systems Association World Applied Sci., 5: 644-651.
Congress, June 18-21, 2007, Cancun, Mexico, Storn, R. and K. Price, 1997. Differential evolution: A
pp: 789-798. simple and efficient heuristic for global optimization
Karaboga, D. and B. Basturk, 2008. On the performance of
over continuous spaces. J. Global Optim., 11: 341-359.
Artificial Bee Colony (ABC) algorithm. Applied Soft
Subotic, M., 2011. Artificial bee colony algorithm with
Comput., 8: 687-697.
multiple onlookers for constrained optimization
Karaboga, D., 2005. An idea based on honey bee swarm
problems. Proceeding of the European Computing
for numerical optimization. Technical Report-TR06,
Conference, April 28-30, 2011, Paris, France,
Erciyes University, Engineering Faculty, Computer
Engineering Department, Kayseri, Turkey, October pp: 251-256.
2005. http://mf.erciyes.edu.tr/abc/pub/tr06_2005.pdf. Tsai, H.C., 2014. Integrating the artificial bee colony and
Kennedy, J., 2011. Particle Swarm Optimization. In: bees algorithm to face constrained optimization
Encyclopedia of Machine Learning, Sammut, C. and problems. Info. Sci., 258: 80-93.
G.I. Webb (Eds.). Springer Science and Business Xiang, W.L. and M.Q. An, 2013. An efficient and robust
Media, New York, USA., ISBN: 9780387307688, artificial bee colony algorithm for numerical
pp: 760-766. optimization. Comput. Operat. Res., 40: 1256-1265.

413