Applied Soft Computing 33 (2015) 15–23

Contents lists available at ScienceDirect

Applied Soft Computing

journal homepage: www.elsevier.com/locate/asoc

The continuous artificial bee colony algorithm for binary optimization

Mustafa Servet Kiran ∗
Department of Computer Engineering, Faculty of Engineering, Selçuk University, 42075 Konya, Turkey

a r t i c l e i n f o a b s t r a c t

Article history: Artificial bee colony (ABC) algorithm, one of the swarm intelligence algorithms, has been proposed for
Received 2 January 2015 continuous optimization, inspired intelligent behaviors of real honey bee colony. For the optimization
Received in revised form 19 March 2015 problems having binary structured solution space, the basic ABC algorithm should be modified because its
Accepted 2 April 2015
basic version is proposed for solving continuous optimization problems. In this study, an adapted version
Available online 22 April 2015
of ABC, ABCbin for short, is proposed for binary optimization. In the proposed model for solving binary
optimization problems, despite the fact that artificial agents in the algorithm works on the continuous
Keywords:
solution space, the food source position obtained by the artificial agents is converted to binary values,
Artificial bee colony
Binary optimization
before the objective function specific for the problem is evaluated. The accuracy and performance of the
Conversion of continuous values proposed approach have been examined on well-known 15 benchmark instances of uncapacitated facility
Uncapacitated facility location problem location problem, and the results obtained by ABCbin are compared with the results of continuous particle
swarm optimization (CPSO), binary particle swarm optimization (BPSO), improved binary particle swarm
optimization (IBPSO), binary artificial bee colony algorithm (binABC) and discrete artificial bee colony
algorithm (DisABC). The performance of ABCbin is also analyzed under the change of control parameter
values. The experimental results and comparisons show that proposed ABCbin is an alternative and simple
binary optimization tool in terms of solution quality and robustness.
© 2015 Elsevier B.V. All rights reserved.

1. Introduction leaf-constrained minimum spanning tree problem, Rao et al. [5]

proposed ABC for optimization of distribution network configura-
In recent years, many swarm intelligence-based meta-heuristic tion for loss reduction. The ABC was implemented to solve quadratic
approaches have been proposed in order to solve NP-hard opti- minimum spanning tree problem by Sundar and Singh [6]. While
mization problems. Artificial bee colony (ABC) algorithm, which is a modified ABC for real parameter optimization was proposed by
one of them, has been proposed for solving continuous optimiza- Akay and Karaboğa [7]; Karaboğa and Akay [8] also modified ABC
tion problems, inspired waggle dance and foraging behavior of real for solving constrained optimization problems by using Deb’s rules
honey bee colonies [1]. If the solution space of the problem is binary [9] and Pan et al. [10] developed a discrete model of ABC for lot-
structured, either working of the method should be adapted to run streaming flow shop scheduling problem. The ABC was also used
on the binary solution space by using binary vectors or continuous for neural networks training [11], solving reliability redundancy
values in the solution vectors in the method should be transformed allocation problems [12], software test suite optimization [13]. In
to binary values. In this study, by inspiring an approach proposed addition, Alatas [14] proposed a chaotic ABC algorithm for avoiding
for particle swarm optimization (PSO) by Güner and Şevkli [2], the to get stuck on local solutions and a Gbest -guided ABC for numer-
continuous values in the solutions produced by ABC have been ical function optimization was proposed by Zhu and Kwong [15].
moved to binary space and the results obtained by the proposed DisABC which means discrete ABC algorithm for solving uncapaci-
approach named as ABCbin have been compared with continuous tated facility location problems (UFLPs) has been also proposed by
PSO. Kashan et al. [16]. Artificial agents of ABC algorithm is modified to
Since invention of ABC, studies on ABC in the literature have work on binary structured solution space to solve binary optimiza-
increased significantly. The ABC algorithm was used for designing tion problems by using logic operators in [17], and an ABC variant
of digital IRR filters by Karaboga [3], Singh [4] used it for solving is proposed for binary optimization in [18].
According to above applications and modifications of ABC, the
algorithm is highly successful for solving optimization problems.
∗ Tel.: +90 332 223 1992; fax: +90 332 241 06 35. But according to best knowledge of the authors, the usage of ABC
E-mail address: [email protected] algorithm that its artificial agents work on continuous solution in

http://dx.doi.org/10.1016/j.asoc.2015.04.007
1568-4946/© 2015 Elsevier B.V. All rights reserved.
16 M.S. Kiran / Applied Soft Computing 33 (2015) 15–23

this study is first effort for solving a binary optimization problem. population and between dimensionality of the problem, respec-
The paper is organized as follows; Section 1 presents introduction tively.
and a brief literature review about the method, the ABC algorithm The fitness value of the new food source is calculated by using
is given in Section 2. Conversion of continuous values to binary val- Eq. (2) and if new one is better than old one, the new food source
ues is presented in Section 3 and the general model of the UFLP position is memorized by the employed bee and the AC of the food
is presented in Section 4. The computational experiments are con- source is reset. Otherwise, AC is increased by 1.
ducted in Section 5. Finally, conclusion and future works are given Onlooker bee phase: The employed bee share the position infor-
in Section 6. mation about the self-food source by dancing in the dance area of
the hive. The onlooker bees watch the dance and select an employed
2. Artificial bee colony algorithm bee by using the fitness values of the food sources of the employed
bees and roulette wheel. The selection probability is calculated as
In the artificial hive of ABC, there are two types of forager follows:
called as employed and unemployed foragers. A new food source
is produced for each employed forager in the initialization of the fiti
pi = N (4)
algorithm. Employed foragers move not only nectar foraged from j=1
fitj
food source but also the position information about food sources
to the hive. Onlooker bees which are one of the unemployed for-
where pi is probability of being selected ith employed bee. After
agers try to improve food source positions of employed foragers by
the selection, the onlooker bees try to improve the solutions of
considering information moved by employed foragers to the hive.
employed bees by using Eq. (3). If new solution obtained by the
In the algorithm, food sources represent feasible solutions for the
onlooker bee is better than the solution of employed bee, the
optimization problem. The other type of unemployed foragers is
employed bee memorizes the solution of onlooker bee and the AC
scout bee. If a food source could not be improved by the employed
is reset. Otherwise, AC is increased by 1.
or onlooker bees in a certain time (limit), the employed bee assigned
Scout bee phase: In this phase, the AC with highest content is
to this food source becomes a scout bee. ABC algorithm consists of
fixed and compared with predetermined limit value. If value of the
four phases named as initialization, employed bee, onlooker bee
AC with maximum content is higher than the limit, the employed
and scout bee, which are sequentially realized. It should be also
bee of food source belongs to this AC with maximum content
mentioned that ABC algorithm is an iterative algorithm and half
becomes a scout bee. A new solution is generated for this scout
of the population in the hive is employed bee and the other half is
bee by using Eq. (1) and the AC of the new food source is reset. The
onlooker bee and only one scout bee can occur at the each iteration.
scout bee generated new solution for itself returns to instatement
Phases of the algorithm are expressed as follows [8]:
(employed bee).
Initialization phase: This phase is realized only one time. The
population size is first determined and half of the population con-
sists of employed bee and the other half is onlooker bees. A new
food source is produced for each employed bee by using Eq. (1). 3. Solution transformation and proposed algorithm

xi,j = xjmin + (xjmax − xjmin ) i = 1, . . ., N, j = 1, . . ., D (1) The continuous values of the food source position in the ABC
algorithm should be converted to the binary values in order to solve
where xi,j is jth dimension of ith employed bee, xjmin and xjmax is the binary optimization problem. Despite the fact that Eq. (5) has
lower and upper bounds of jth parameters, respectively,  is a ran- been proposed for conversion of continuous values to binary solu-
dom number in range of [0,1], N is the number of employed bees tion in Şevkli and Güner [19], we used a bit modified version (Eq.
and D is the dimensionality of the optimization problem. In addi- (6)) of Eq. (5) in order to increase search ability of the ABC algorithm
tion, the abandonment counter (AC) of each employed bee is reset and provide quick change for the binary values in this study.
in this phase. And then, the fitness values of the food sources of the
employed bees are calculated as follows: zi = floor(|yi mod 2|) (5)
⎧
⎨ 1
if (fi ≥ 0)
fiti = 1 + fi (2) where yi is the solution vector included continuous values and zi
⎩ 1 + abs(f ) otherwise is the binary vector. The continuous value is first divided by 2 and
i
absolute value of the remainder is floored. Therefore, the binary
where fiti is the fitness value of food source of ith employed bee, fi is number is obtained. An example is given as follows:
the objective function value specific for the optimization problem
of food source of ith the employed bee. In addition, limit value is zi = floor(| − 3.724 mod 2|) = floor(| − 1.724|) = floor(|1.724|) = 1
determined and a counter (abandonment counter-AC) is created
and reset for each food source in this phase.
According to this conversion, if the absolute value of the remain-
Employed bee phase: Each employed bee try to find a new food
der is between 0 and 0.9999, the binary number will be 0 and if the
source in order to improve self-solution by using Eq. (3).
absolute value of the remainder is between 1 and 1.9999, the binary
vi,j = xi,j + (xi,j − xk,j ) i, k ∈ {1, 2, . . ., N}, number will be 1. When absolute value of the remainder is higher
than 0.5, despite the fact that absolute value of the remainder is
j ∈ {1, 2, . . ., D} and i =
/ k (3) closer to 1 than 0, it is obtained 0 by Eq. (5). The same manner,
when absolute value of the remainder is higher than 1.5, despite
where vi,j jth dimension of ith candidate solution, xi,j is jth dimen-
the fact that absolute value of the remainder is closer to 2 than 1,
sion of ith employed bee, xk,j is jth dimension of kth employed
it is obtained 1 by Eq. (5). Therefore, we propose round operation
bee,  is a random number in range of [−1,+1], N is the number of
instead of floor and double mod process given by Eq. (6).
employed bee and D is the dimensionality of the optimization prob-
lem. Also, the neighbor of candidate solution (k) and dimension of
the problem (j) are randomly selected among the employed bee zi = round(|yi mod 2|)mod 2 (6)
 M.S. Kiran / Applied Soft Computing 33 (2015) 15–23 17

Table 1
Conversion of continuous values to binary values.

Food Source Dimensions

1 2 3 4 5 6

Food source position on the continuous solution space 5.624 −4.645 1.892 3.328 −1.342 8.456
Food source position on the binary solution space 0 1 0 1 1 0

An example of the solving Eq. (6) is presented as follows: binary variable: yi = 1 if a facility is located in i ∈ I, yi = 0 otherwise.
While Eq. (8) provides to satisfy all demand of customers, Eq. (9)
zi = round(| − 3.724 mod 2|)mod 2
ensures that a customer can be served from a facility only if a facility
= round(| − 1.724|) mod 2 is opened. Eqs. (10) and (11) define the decision variables in binary
= round(|1.724|) mod 2 structure. Due to single assignment property of UFLP, demand of
a customer is always entirely fulfilled by its nearest facility [21].
= 2 mod 2
When the locations of facilities which will be opened are deter-
=0 mined, which customer will be served by a facility can be obtained
By using Eq. (6), if the absolute value of remainder is between 0 easily. Therefore, we address a vector A of n variables where each
and 0.4999 or 1.5 and 1.9999, the binary number is obtained as 0. variable in A is bit. Each bit (Ai , i = 1,2,. . .,n) indicates that whether
If the absolute value of remainder is between 0.5 and 1.4999, the a facility is opened at location i or not. This vector plays a critical
binary number is obtained as 1. role in ABCbin because each employed bee represents this vector.
In the ABC algorithm, before the continuous solution vector UFLP is one of the most important NP-hard problems in loca-
is evaluated by objective function specific for the problem, it is tion theory [16,22] and many studies are presented on it in the
converted to binary solution vector by using Eq. (6). This is also literature. In order to solve UFLPs, many exact methods such as
illustrated in Table 1. the branch-and-bound [23], linear programming and Lagrangian
After transformation mechanism and ABC algorithm are relaxation [24], dual-based procedure [25] have been proposed.
explained, the adaptation of the mechanism with ABC algorithm Despite the fact that these methods ensure the optimality, com-
is shown in Fig. 1. putation time of these methods may be too much. For this reason,
According to Fig. 1, it is shown that before the objective func- some approximate methods have been proposed for solving UFLPs.
tion specific for the problem, which is objective of UFLP in this These methods cannot guarantee to find the optimal solution but
study, is evaluated, the continuous solutions should be converted can obtain optimum or near optimum solution in a reasonable
to binary solution. Like the basic ABC algorithm, the proposed time. For instance, genetic algorithm [26], tabu search [27,28], con-
approach works on continuous solution space but the binary solu- tinuous and discrete PSO [2,19] have been proposed in order to
tions are used before evaluating the objective function by nature of solve UFLPs. In addition, a comparative study for binABC [17], Dis-
the problem. ABC [16], discrete binary particle swarm optimization (BPSO) [29],
improved BPSO (IBPSO) [30] on solving UFLPs is presented in [17].
4. Uncapacitated facility location problem (UFLP)
5. Computational experiment
A brief description about UFLP is given in this section. The per-
formance and accuracy of the proposed ABCbin algorithm are tested In order to test performance and accuracy of proposed approach
on UFLP. In the basic formulation, UFLP consists of a set of potential with PSO, 15 UFLPs taken from OR-Lib [31] were used and the prob-
facility sites I where a facility can be opened and has no capacity lem size and optimum cost for the problems were given in Table 2.
constraint, and a set of customer location J that must be serviced. The results obtained by ABCbin were compared with the results of
The goal (Eq. (7)) is to find a subset F of I facilities that is cor- CPSO (continuous PSO) algorithm, taken from Şevkli and Güner
responded demand of customers J. The objective function of this [19]. In comparison tables, the better results are written in bold
problem is to minimize sum of the shipment costs between F and face font type. According to Şevkli and Güner [19], the population
J and the opening costs of the facilities. The general model of the size of the method is selected as number of facilities and the upper
UFLP can be mathematically expressed as following [20]: and lower bound of the continuous solution space are taken as −10
⎧ ⎫ and +10, respectively. c1 cognitive and c2 social components of the
⎨  ⎬
f (UFLP) = min cij xij + fi yi (7)
⎩ ⎭ Table 2
i∈I j∈J i∈I The test suite used for comparison of the methods.

Subject to: Problem name Problem size Cost of optimal solution

 Cap71 16 × 50 932,615.75
xij = 1, j∈J (8) Cap72 16 × 50 977,799.40
i∈I Cap73 16 × 50 1,010,641.45
Cap74 16 × 50 1,034,976.98
xij ≤ yi , i ∈ I and j ∈ J (9) Cap101 25 × 50 796,648.44
Cap102 25 × 50 854,704.20
xij ∈ {0, 1}, i ∈ I and j ∈ J (10) Cap103 25 × 50 893,782.11
Cap104 25 × 50 928,941.75
yi ∈ {0, 1}, i∈I (11) Cap131 50 × 50 793,439.56
Cap132 50 × 50 851,495.33
where I = {1,2,. . .,n} is the set of possible facility locations, Cap133 50 × 50 893,076.71
J = {1,2,. . .,m} is the set of customer demand points, fi is the fixed Cap134 50 × 50 928,941.75
cost of opening a facility in i ∈ I, cij is the shipment cost between CapA 100 × 1000 17,156,454.48
ith facility location and jth customer point. The decision variable xij CapB 100 × 1000 12,979,071.58
CapC 100 × 1000 11,505,594.33
is the demand of customer j corresponded ith facility and yi is the
18 M.S. Kiran / Applied Soft Computing 33 (2015) 15–23

1. Initialization
a. Set N as the number of food sources
b. Set AC for each food source.
c. Set the limit control parameter value.
d. Set the termination condition.
e. Set D as dimensionality of the problem.
f. Create food sources (NxD) using Eq. 1
g. Convert continuous solutions to binary solutions using Eq.6.
h. Evaluate the fitness of the solutions by using objective function specific for the problem and
Eq. 2
i. Memorize the best solution
2. Employed Bee Phase
a. For each employed bee
i. Select a neighbor solution randomly.
ii. Produce a candidate food source by using Eq. 3
iii. Convert to continuous solution to binary solution using Eq.6
iv. Evaluate the fitness of the solution by using objective function specific for the
problem and Eq. 2.
v. If fitness of the new solution is better than old one, memorize the new solution and
reset AC of this food source; otherwise increase AC by 1.
3. Onlooker Bee Phase
a. Calculate designation probability of each food source by using Eq. 4
b. For each onlooker bee
i. Select a food source by using designation probability.
ii. Select a food source randomly.
iii. Produce a candidate food source by using Eq. 3
iv. Convert to continuous solution to binary solution using Eq.6
v. Evaluate the fitness of the solution by using objective function specific for the
problem and Eq. 2.
vi. If fitness of the new solution is better than old one, memorize the new solution and
reset AC of this food source; otherwise increase AC by 1.
4. Scout Bee Phase
a. Fix the AC with maximum content
b. If this content is higher than limit,
i. Remove this solution from the population
ii. Create a new solution by using Eq. 1
iii. Convert continuous solution to binary solution using Eq.6.
iv. Evaluate the fitness of the solution by using objective function specific for the
problem and Eq. 2
v. Reset AC of new solution.
5. Termination
a. If the best solution of the population is better previous best solution, memorize the new best
solution.
b. If the termination condition is met, report the best solution; otherwise go to Employed Bee
Phase

Fig. 1. The detailed pseudo-code of proposed approach.

PSO are set to 2, the upper and lower bound of velocities for the PSO The gap in the graph is used in order to show the accuracy of the
are taken as −4 and +4, respectively. Inertia weight for the PSO is a method and it is calculated by using Eq. (13). As it is seen from the
random number in the range [0.5,1]. The termination condition of Fig. 2, the ABCbin has converged to the optimum or near optimum
the algorithms is maximum iteration number and it is set as 1000 because global search and exploitation ability of the algorithm is
iterations. Also, the limit value for ABCbin is calculated as follows: quite good.

limit = N × D (12) f (sol) − f (opt)

Gap = × 100 (13)
f (opt)
where N is the population size and D is the number of potential
facility locations. The limit is N square due to equality of N and D. It where f(opt) is the cost of the optimum solution, f(sol) is the cost of
must be mentioned that if the number of the potential facility loca- the best solution obtained by ABCbin at each iteration.
tion (D) is odd number, the population size of the ABCbin is selected As seen from Table 3, ABCbin is superior to CPSO in terms of solu-
as D − 1 in order to avoid odd number for population size of ABCbin , tion quality and also ABCbin algorithm is more robust than CPSO
because half of the population of ABCbin consists of the employed according to standard deviations. Especially, in respect of CPU time,
bees and the other half consists of onlooker bees. CPSO consumes more CPU time than ABCbin algorithm. Finally, it
The methods were repeated 30 times with random seeds and should be mentioned that ABCbin algorithm is faster than CPSO
the best, the worst, standard deviations, number of runs yielded to algorithm, with respect to running time of the algorithms.
optimum (Hit) and CPU times of the algorithms were reported in the
Table 3. The results of CPSO are taken from Şevkli and Güner [19]. 5.1. Effect of the population size value on the performance of the
In order to make a fair comparison between the algorithms, CPSO ABCbin algorithm
algorithm is re-coded on MATLAB with same PC and CPU times
are obtained by tic and toc functions of MATLAB. The convergence In order to analyze the behavior of the proposed ABCbin , it has
graph of the ABCbin algorithm drawn for each problem is presented been run with different population sizes. The effect of the popula-
in Fig. 2. tion size is tested on four problems that are Cap73, Cap103, Cap133
 M.S. Kiran / Applied Soft Computing 33 (2015) 15–23 19

Fig. 2. Convergence graph of the ABCbin algorithm.

Cap73 Cap103 Cap133

1,0015
1,00135
1,0012
1,00105
1,0009
1,00075
%

1,0006
1,00045
1,0003
1,00015
1
10 20 30 40 50 60 70 80 90 100
Populaon size

Fig. 3. Evolution of mean values for Cap73, 103 and 133 based on different population size.

CapB

1,045
1,04
1,035
1,03
1,025
%

1,02
1,015
1,01
1,005
1
10 20 30 40 50 60 70 80 90 100
Populaon size

Fig. 4. Evolution of mean values for CapB based on different population size.
20 M.S. Kiran / Applied Soft Computing 33 (2015) 15–23

Table 3
The results obtained by CPSO and ABCbin algorithms.

Problem CPSO [19] ABCbin

Best Worst Std. Dev. Hit CPU Best Worst Std. Dev. Hit CPU

Cap71 932,615.75 934,199.14 562.23 25 12.32 932,615.75 932,615.75 0.00 30 9.09

Cap72 97,799.40 983,713.81 1324.30 25 11.96 977,799.40 977,799.40 0.00 30 8.98
Cap73 1,010,641.45 1,012,643.69 702.13 22 11.78 1,010,641.45 1,010,641.45 0.00 30 8.99
Cap74 1,034,976.98 1,045,342.23 2124.54 0 11.99 1,034,976.98 1,034,976.98 0.00 30 8.95
Cap101 796,648.44 802,457.23 1480.72 0 18.34 796,648.44 796,648.44 0.00 30 14.45
Cap102 854,704.20 857,380.85 1015.64 10 18.28 854,704.20 854,704.20 0.00 30 14.48
Cap103 893,782.11 899,424.91 1695.79 0 18.41 893,782.11 894,008.14 85.67 25 14.38
Cap104 928,941.75 944,394.83 3842.64 18 18.26 928,941.75 928,941.75 0.00 30 14.33
Cap131 795,291.86 804,549.64 2429.54 0 38.62 793,439.56 794,910.64 1065.73 6 33.85
Cap132 851,495.33 865,667.16 4297.07 0 38.95 851,495.33 851,636.70 213.28 14 33.83
Cap133 893,076.71 909,908.70 4210.93 0 38.62 893,076.71 895,407.93 561.34 5 33.62
Cap134 928,941.75 951,803.25 6619.05 7 38.35 928,941.75 928,941.75 0.00 30 33.51
CapA 18,351,465.4 24,638,983.7 1,608,650.3 0 1106.34 17,180,540.6 18,144,628.9 268,685.2 0 1008.8
CapB 13,390,061.3 15,356,618.4 532,161.6 0 1065.44 13,203,262.1 13,344,479.4 88,452.8 0 1011.6
CapC 11,970,113.1 13,572,876.1 350,412.5 0 1063.63 11,609,406.5 11,739,020.4 78,162.2 0 1004.2

Table 4
Mean values of Cap73, 103, 133 and B under different population sizes.

Population size Problem

Cap73 Cap103 Cap133 CapB

10 1,010,641.45 893,997.03 894,356.68 13,451,664.52

20 1,010,641.45 893,895.13 893,655.89 13,498,396.50
30 1,010,641.45 893,849.92 893,511.90 13,435,810.82
40 1,010,641.45 893,872.52 893,675.84 13,332,636.28
50 1,010,641.45 893,782.11 893,604.60 13,309,742.24
60 1,010,641.45 893,804.72 893,313.47 13,238,283.50
70 1,010,641.45 893,782.11 893,344.83 13,302,798.87
80 1,010,641.45 893,804.72 893,252.13 13,326,588.35
90 1,010,641.45 893,782.11 893,343.51 13,228,199.82
100 1,010,641.45 893,782.11 893,414.05 13,259,758.37
Optimal 1,010,641.45 893,782.11 893,076.71 12,979,071.58

and CapB to represent of each test problem group. Each of the exper- 5.2. Effect of the limit value on the performance of the ABCbin
iments was repeated 10 times with the limit value = 1000. The mean algorithm
values with different population sizes varying as 10–100 by step 10
have been presented in Table 4 and shown in Figs. 3 and 4. As mentioned in Section 2, the scout bee production is controlled
From Table 4 and Figs. 3 and 4, it can be concluded that as the by the control parameter limit in the ABC algorithm. According
population size increases, as expected the algorithm produces bet- to Karaboğa and Baştürk [28], there is an inverse proportionality
ter results. For the Cap73, optimal value is obtained in all population between the value of limit and the scout bee production frequency.
sizes. For the Cap103, after the population size is 50, algorithm can As the value of limit approaches to infinity, the total number of
find the optimal value. However, optimal values are not achieved the scout bees produced goes to zero. The limit value used in this
in each different population size for the test problems Cap133 and paper is equaled to N × D, where, N is the population size and D is
B. the number of potential facility locations. The limit is N square due
Finally, increasing the population size from 10 to 100 does not to equality of N and D. In order to show the effects of changing the
make any sense on optimal convergence of Cap73 and increases limit value, in other words, to show the effect of the scout bee pro-
the convergence to optimal of Cap103, 133 and B 0.024%, 0.11% duction on the performance of the ABCbin , the mean values for the
and 1.43%, respectively. In the frame of effect of population size on test problems Cap73, 103, 133 and B are obtained for the different
the test problems, the population size that is greater the 50 can limit values (N × D/4; N × D/2; N × D; N × D × 2; N × D × 4) and the
provide an acceptable and effective convergence speed for search. results are given in Table 5 and Figs. 5 and 6.

Table 5
Mean values of Cap73, 103, 133 and B under different limit values.

Limit value Problem

Cap73 Cap103 Cap133 CapB

N × D/4 1,010,641.45 893,804.72 893,415.37 13,288,275.16

N × D/2 1,010,641.45 893,804.72 893,229.74 13,228,042.66
N×D 1,010,641.45 893,906.62 893,581.58 13,279,562.22
N×D×2 1,010,641.45 893,804.72 893,498.48 13,262,862.84
N×D×4 1,010,641.45 894,110.43 893,504.71 13,246,322.41
Optimal 1,010,641.45 893,782.11 893,076.71 12,979,071.58
 M.S. Kiran / Applied Soft Computing 33 (2015) 15–23 21

Cap73 Cap103 Cap133

1,0006
1,0005
1,0004
1,0003
%

1,0002
1,0001
1
N*D/4 N*D/2 N*D N*D*2 N*D*4
Limit value

Fig. 5. Evolution of mean values for Cap73, 103 and 133 based on different limit values.

CapB

1,025
1,024
1,023
1,022
%

1,021
1,02
1,019
1,018
N*D/4 N*D/2 N*D N*D*2 N*D*4
Limit value

Fig. 6. Evolution of mean values for CapB based on different limit values.

As seen from Table 5 and Figs. 5 and 6, for all test problems, value will be updated dynamically. With mentioned parameter
when the scout bee production frequency is very high (N × D/4) or application, the mean values for the test problems Cap73, 103,
close to zero (N × D/4), the results obtained by the proposed ABCbin 133 and B are obtained and the results are given in Table 6 and
are get worse that the moderate value for limit value is N × D/2. Figs. 7 and 8.
The results that are nearest to optimal values are obtained when As seen from Table 6 and Figs. 7 and 8, optimal values are
the limit value is equaled to N × D/2. obtained for the test problem Cap73 in each population size and
This results show that the global search ability of the proposed limit value. Increasing the population size and accordingly the limit
method is as important as local search ability of the method. And value (N × D) also increases the optimal convergence of test prob-
also it shows that scout bee production should not be neglected for lems Cap103, 133 and B. As expected, increasing the number of
this kind of problems. population size and limit value at the same time results in increas-
ing the chance of getting the optimal solution by the proposed
algorithm. Almost for all problems, choosing a greater number
5.3. Effect of the population size and limit to the performance of of population size, results in improving the quality of solutions
ABCbin simultaneously obtained by ABCbin .

In addition to the analyses of the effects of changing popula-

tion sizes and limit values on the performance of the ABCbin , in this 5.4. The comparison of ABCbin with swarm intelligence-based
section, simultaneous analyses of these two decisive parameters binary optimizers
is investigated. For that reason, the population size is increased
from 10 to 100 step by 10 while the limit value is equaled to In order to validate performance of the proposed algorithm, the
N × D. Therefore, while the population size is increased, the limit performance of the proposed approach is compared with the some

Table 6
Mean values of Cap73, 103, 133 and B under different population sizes and limit values.

Population Size Problem

Cap73 Cap103 Cap133 CapB

10 1,010,641.45 893,804.72 894,469.92 13,512,861.39

20 1,010,641.45 893,804.72 894,012.87 13,393,571.81
30 1,010,641.45 893,804.72 893,926.47 13,394,279.57
40 1,010,641.45 893,827.32 893,809.11 13,324,492.44
50 1,010,641.45 893,804.72 893,572.61 13,364,560.11
60 1,010,641.45 893,782.11 893,573.93 13,338,066.64
70 1,010,641.45 893,849.92 893,749.11 13,284,705.69
80 1,010,641.45 893,804.72 893,565.42 13,316,676.15
90 1,010,641.45 893,827.32 893,569.72 13,283,700.50
100 1,010,641.45 893,804.72 893,397.89 13,267,806.92
Optimal 1,010,641.45 893,782.11 893,076.71 12,979,071.58
22 M.S. Kiran / Applied Soft Computing 33 (2015) 15–23

Cap73 Cap103 Cap133

1,00165
1,0015
1,00135
1,0012
1,00105
1,0009
%

1,00075
1,0006
1,00045
1,0003
1,00015
1
10 20 30 40 50 60 70 80 90 100
Population size

Fig. 7. Evolution of mean values for Cap73, 103 and 133 based on different population size and limit value.

CapB

1,045
1,04
1,035
1,03
1,025
%

1,02
1,015
1,01
1,005
1
10 20 30 40 50 60 70 80 90 100
Population size

Fig. 8. Evolution of mean values for CapB based on different population size and limit value.

binary versions of PSO and ABC algorithms. The DisABC [16] and For comparison of ABCbin and BPSO, IBPSO, binABC and
binABC [17] algorithms are binary structured methods and they DisABC, the results of these methods on the test problems
work on the binary solution space. BPSO [29] is the discrete binary are taken directly from [17]. The control parameter setting
version of classic PSO algorithm. IBPSO [30] is a variant of BPSO but of the methods are summarized in Table 7, and the compar-
the particles of IBPSO works on binary structured solution space. ative results are given in Table 8. Based on Table 8, it is
While DisABC uses similarity-based search on the solution space, shown that the proposed method produces competitive results
the binABC and IBPSO uses logic operator for producing candidate for the binary test problems like binABC and DisABC algo-
solutions. rithms.

Table 7
Control parameters of compared algorithms.

BPSO IBPSO binABC DisABC ABCbin

Population size 40 40 40 40 D
Maximum number of iterations 2000 2000 2000 2000 1000
c1 2 2 N/A N/A N/A
c2 2 2 N/A N/A N/A
Limit N/A N/A N × D/4 2.5 × Nb × D N × D/2

Table 8
Comparison of ABCbin with the binary structured optimization algorithms.

Problem BPSO IBPSO binABC DisABC ABCbin

Gap Std.dev. Gap Std.dev. Gap Std.dev. Gap Std.dev. Gap Std.dev.

Cap71 0.0000 0.00 0.0374 587.49 0.0000 0.00 0.0000 0.00 0.00 0.00
Cap72 0.0000 0.00 0.2749 1844.64 0.0000 0.00 0.0000 0.00 0.00 0.00
Cap73 0.0242 634.62 0.1980 1513.78 0.0000 0.00 0.0000 0.00 0.00 0.00
Cap74 0.0088 500.27 0.4031 4426.67 0.0000 0.00 0.0000 0.00 0.00 0.00
Cap101 0.0462 566.44 0.5968 3799.52 0.0000 0.00 0.0000 0.00 0.00 0.00
Cap102 0.0148 386.76 0.7317 3249.38 0.0000 0.00 0.0000 0.00 0.00 0.00
Cap103 0.0422 485.26 0.6410 4978.98 0.0000 0.00 0.0000 0.00 0.0051 85.67
Cap104 0.0810 1951.81 0.9964 10,845.26 0.0000 0.00 0.0000 0.00 0.00 0.00
Cap131 0.1317 1207.63 2.4236 4244.29 0.0000 0.00 0.6196 2337.64 0.1967 1065.73
Cap132 0.0914 1196.19 3.6014 11,569.02 0.0000 0.00 0.0945 813.37 0.0199 213.28
Cap133 0.1115 821.28 5.2626 14,905.27 0.1215 200.24 0.0309 359.03 0.0747 561.34
Cap134 0.1346 2285.42 7.6338 15,788.86 0.0000 0.00 0.0000 0.00 0.00 0.00
CapA 2.1785 374,302.81 37.8862 3,357,138.19 2.9622 236,833.5 0.1522 74,782.61 3.1723 268,685.2
CapB 1.9490 176,206.07 55.2701 1,406,575.7 2.5081 91,430.13 3.3027 109,738.5 2.8154 88,452.8
CapC 1.4870 92,977.85 45.5561 1,245,252.2 2.5800 82,312.70 4.6968 95,778.78 2.0374 78,162.2
 M.S. Kiran / Applied Soft Computing 33 (2015) 15–23 23

6. Conclusion [6] S. Sundar, A. Singh, A swarm intelligence approach to the quadratic minimum
spanning tree problem, Inf. Sci. 180 (17) (2010) 3182–3191.
[7] B. Akay, D. Karaboga, A modified Artificial Bee Colony algorithm for real-
In this paper, an adapted version of ABC algorithm which is parameter optimization, Inf. Sci. 192 (2012) 120–142.
called ABCbin is introduced to solve binary problems. In the pro- [8] D. Karaboga, B. Akay, A modified Artificial Bee Colony (ABC) algorithm for con-
posed algorithm for solving binary optimization problems, despite strained optimization problems, Appl. Soft Comput. 11 (3) (2011) 3021–3031.
[9] K. Deb, An efficient constraint handling method for genetic algorithms, Comput.
the fact that artificial agents in the algorithm works on the con- Methods Appl. Mech. Eng. 186 (2–4) (2000) 311–338.
tinuous solution space, the food source position obtained by the [10] Q.K. Pan, et al., A discrete artificial bee colony algorithm for the lot-streaming
artificial agents is converted to binary values, before the objective flow shop scheduling problem, Inf. Sci. 181 (12) (2011) 2455–2468.
[11] D. Karaboga, C. Ozturk, Neural networks training by Artificial Bee Colony algo-
function specific for the problem is evaluated. The effectiveness
rithm on pattern classification, Neural Netw. World 19 (3) (2009) 279–292.
of the proposed approach is tested on benchmark test problem [12] W.C. Yeh, T.J. Hsieh, Solving reliability redundancy allocation problems using an
instances of the UFLP, and compared with continuous particle artificial bee colony algorithm, Comput. Oper. Res. 38 (11) (2011) 1465–1473.
[13] D.J. Mala, V. Mohan, ABC tester-artificial bee colony based software test suite
swarm optimization technique that is applied for UFLP in the
optimization approach, Int. J. Softw. Eng. 2 (2009) 15–43.
literature, where the results demonstrate that our approach is [14] B. Alatas, Chaotic bee colony algorithms for global numerical optimization,
competitive. In addition to the comparison, the behavior of ABCbin Expert Syst. Appl. 37 (8) (2010) 5682–5687.
algorithm under different control parameter values has also been [15] G.P. Zhu, S. Kwong, Gbest-guided artificial bee colony algorithm for numerical
function optimization, Appl. Math. Comput. 217 (7) (2010) 3166–3173.
analyzed. When compared ABCbin algorithm with PSO variants, [16] M.H. Kashan, N. Nahavandi, A.H. Kashan, DisABC: a new artificial bee colony
proposed method is better than the other methods in terms of solu- algorithm for binary optimization, Appl. Soft Comput. 12 (1) (2012) 342–352.
tion quality. When we compare ABCbin with the ABC variants, the [17] M.S. Kiran, M. Gunduz, XOR-based artificial bee colony algorithm for binary
optimization, Turk. J. Electr. Eng. Comput. Sci. 21 (2013) 2307–2328.
results are competitive and implementation and adaptation of the [18] M.S. Kıran, E. Özceylan, T. Paksoy, Artificial bee colony algorithm for solv-
proposed method is easier than the ABC variants. Generally, the ing uncapacitated facility location problems, in: 25th European Conference on
results obtained by the proposed approach are very promising, and Operational Research, Vilnius, Lithuania, 2012, p. 165.
[19] M. Şevkli, A.R. Güner, A continuous particle swarm optimization algorithm
the future works include hybridization of proposed approach with for uncapacitated facility location problems, Lect. Notes Comput. Sci. 4150
local search modules to improve intensification capability of the (316–323) (2006) 316.
proposed method. [20] R.D. Galvâo, L.A. Raggi, A method for solving to optimality uncapacitated loca-
tion problems, Ann. Oper. Res. 18 (1) (1989) 225–244.
[21] J. Krarup, P.M. Pruzan, The simple plant location problem – survey and synthe-
Acknowledgements sis, Eur. J. Oper. Res. 12 (1) (1983) 36–81.
[22] M.L. Cornuejols, G.L. Nemhauser, L.A. Wolsey, The uncapacitated facility loca-
tion problem, in: R.L. Francis, P. Mirchandani (Eds.), Discrete Location Theory,
The author thanks Dr. Eren Özceylan and Dr. Turan Paksoy for
Wiley Interscience, New York, 1990.
their helps at the beginning of the study. The author also wishes [23] K. Holmberg, Exact solution methods for uncapacitated location problems with
to thank Selcuk University Scientific Research Project Office and convex transportation costs, Eur. J. Oper. Res. 114 (1) (1999) 127–140.
[24] J. Barcelo, et al., Lagrangian relaxation and constraint generation procedures
Scientific and Technological Research Council of Turkey for their
for capacitated plant location problems with single sourcing, OR Spectr. 12 (2)
institutional support. (1990) 78–79.
[25] D. Erlenkotter, A dual-based procedure for uncapacitated facility location, Oper.
Res. 26 (6) (1978) 992–1009.
References
[26] J.H. Jaramillo, J. Bhadury, R. Batta, On the use of genetic algorithms to solve
location problems, Comput. Oper. Res. 29 (6) (2002) 761–779.
[1] D. Karaboga, An idea based on honey bee swarm for numerical optimization, in: [27] K.S. Al-Sultan, M.A. Al-Fawzan, A tabu search approach to the uncapacitated
Technical Report-TR06, Erciyes University, Engineering Faculty, Department of facility location problem, Ann. Oper. Res. 86 (1999) 91–103.
Computer Engineering, 2005. [28] M.H. Sun, Solving the uncapacitated facility location problem using tabu search,
[2] A.R. Güner, M. Şevkli, A discrete particle swarm optimization algorithm for Comput. Oper. Res. 33 (9) (2006) 2563–2589.
uncapacitated facility location problem, J. Artif. Evol. Appl. 2008 (2008) 1–9. [29] J. Kennedy, R.C. Eberhart, A discrete binary version of the particle swarm algo-
[3] N. Karaboga, A new design method based on artificial bee colony algorithm for rithm, in: Smc’97 Conference Proceedings – 1997 IEEE International Conference
digital IIR filters, J. Franklin Inst. Eng. Appl. Math. 346 (4) (2009) 328–348. on Systems, Man, and Cybernetics, vols. 1–5, 1997, pp. 4104–4108.
[4] A. Singh, An artificial bee colony algorithm for the leaf-constrained minimum [30] X.H. Yuan, et al., An improved binary particle swarm optimization for unit
spanning tree problem, Appl. Soft Comput. 9 (2) (2009) 625–631. commitment problem, Expert Syst. Appl. 36 (4) (2009) 8049–8055.
[5] R.S. Rao, S. Narasimham, M. Ramalingaraju, Optimization of distribution net- [31] J.E. Beasley, OR-library – distributing test problems by electronic mail, J. Oper.
work configuration for loss reduction using artificial bee colony algorithm, Int. Res. Soc. 41 (11) (1990) 1069–1072.
J. Electr. Power Energy Syst. Eng. 1 (2) (2008) 116–122.