International Journal of Recent Research and Review, Vol.

VII, Issue 3, September 2014

ISSN 2277 8322

Fast-Convergent Artificial Bee Colony with an Adaptive Local

Search
Aishwarya Varshney1, R. L. Yadav2, Vijay Kumar3
1

Research Scholar, Kautilya Institute of Technology, Jaipur, Rajasthan, India; Email: [email protected]
Assistant Professor, Kautilya Institute of Technology Jaipur, Rajasthan, India; Email: [email protected]
3
Assistant Professor, Kautilya Institute of Technology Jaipur, Rajasthan, India; Email: [email protected]

optimization (ACO) [1], particle swarm optimization

(PSO) [2] & bacterial foraging optimization (BFO) [3] are
some examples of swarm intelligence based techniques.
The work presented in the articles [4-5] proved its
efficiency and potential to deal with non linear, non
convex and discrete optimization problems. D.Karaboga
[6] contributed the recent addition to this category known
as artificial bee colony (ABC) optimization algorithm.
The ABC algorithm mimics the foraging behavior of
honey bees while searching food for them. ABC is a
simple and population based optimization algorithm. Here
the population consists of possible solutions in terms of
food sources for honey bees whose fitness is regulated in
terms of nectar amount which the food source contains.
The swarm updating in ABC is due to two processes
namely, the variation process and the selection process
which are responsible for explo-ration and exploitation,
respectively. However the ABC achieves a good solution
at a significantly faster rate but, like the other
optimization algorithms, it is also weak in refining the
already explored search space. On the other part, it is also
required to tune the ABC control parameters based on
problem. Also literature says that basic ABC itself has
some drawbacks like stop proceeding toward the global
optimum even though the population has not converged to
a local optimum [7] and it is observed that the position
update equation of ABC algorithm is good at exploration
but poor at exploitation [8] i.e, has not a proper balance
between exploration and exploitation. Therefore these
drawbacks require a modification in position update

Abstract - Fast Convergent Artificial Bee Colony (ABC)

algorithm has been emerged as one of the latest swarm
intelligence based algorithm. Though, there are some areas
where ABC outperforms the other optimization techniques
but, the drawbacks like preference on exploration at the
cost of exploitation and skipping the true solution due to
large step sizes, are also associated with it. In this paper, to
improve the exploitation capability of the ABC algorithm, a
new local search strategy is developed and incorporated
with ABC. The proposed local search strategy is based on
linearly decreasing inertia weight which has been applied in
particle swarm optimization (PSO) algorithm. The
proposed local search strategy greedily reduces the interval
of step size for the best solution in the colony. The modified
ABC is named as fast convergent ABC with an adaptive
local search (FABCLS). To validate the performance of the
proposed algorithm FABCLS, 15 benchmark optimization
functions of different complexities are considered and
results comparison section shows the clear superiority of the
proposed modification over the basic ABC.
Keywords - Artificial Bee Colony, Local Search Algorithm,
Exploration-Exploitation, Swarm Intelligence.
I.

INTRODUCTION

Swarm Intelligence is one of the recent outcomes of the

research in the field of Nature inspired algorithms.
Collaborative trial and error method is the main concept
behind the Swarm Intelligence which enables the
algorithmic procedure to find the solution. Researchers
are analyzing such collaboration among the social insects
while searching food for them and creating the intelligent
structures known as Swarm Intelligence. Ant colony


As mentioned earlier, if ij and difference between
randomly selected solution and current solution is high in
position update equation of ABC then there will be
sufficient chance to skip the global optima. In this
situation, some local search strategy can help the search
procedure. During the iterations, local search algorithm
illustrates very strong exploitation capability [14].
Therefore, the exploitation capability of ABC algorithm
may be enhanced by incorporating a local search strategy
with ABC algorithm.
In this way, the exploration and exploitation capability of ABC algorithm could be balanced as the global
search capability of the ABC algorithm explores the
search space or tries to identify the most promising search
space regions, while the local search strategy will exploit
the identified search space.
Therefore, in this paper a self adaptive local search
strategy is proposed and incorporated with the ABC. The
proposed local search strategy is inspired from linearly
decreasing inertia weight (LDIW) [13].
In the proposed local search strategy, the required
step size (i.e.,(x xrandom)) to update an individual is
reduced self adaptively to exploit the search area in the
proximity of the best candidate solution. Thus, the
proposed strategy is named as self adaptive local search
(SALS). In SALS, the step size is reduced as a
logarithmic function of iteration counter as shown in
equations (2) and (3). In SALS, the random component
(ij) of basic ABC algorithm is optimized to direct the
best solution to update its position. This process can be
seen as an optimization problem solver which minimizes
the unimodal continuous objective function f(ij) in the
direction provided by the best solution xbest over the
variables w1,w2 in the interval [1, 1] or simply it
optimizes the following mathematical optimization
problem:
min f() in [1, 1];
(1)

equation and/or a local search approach to be

implemented in ABC. These drawbacks have also
addressed in earlier research. To enhance the exploitation,
Wei-feng Gao et.al [9] improved position update equation
of ABC such that the bee searches only in neighborhood
of the previous iterations best solution. Anan Banharnsakun et al. [10] proposed the best-so-far selection in
ABC algorithm and incorporated three major changes:
The best-so-far method, an adjustable search radius, and
an objective-value-based comparison in ABC. To solve
constrained optimization problems, D. Karaboga and B.
Akay [11] used Debs rules consisting of three simple
heuristic rules and a probabilistic selection scheme in
ABC algorithm. Dervis Karaboga [9] examined and
suggested that the limit should be taken as SN D, where,
SN is the population size and D is the dimension of the
problem and coefficient ij in position update equation
should be adopted in the range of [-1, 1]. Further, Kang et
al. [12] introduced exploitation phase in ABC using
Rosenbrocks rotational direction method and named
modified ABC as Rosenbrock ABC (RABC). In this
paper, the effective modification in ABC that we made is
the incorporation of a local search strategy in ABC after
all its three (emplyoed, onlooker and scout bee) phases,
based on linearly decreasing inertia weight (LDIW) [13].
This local strategy exploits only the global best position
of the current swarm after each iteration.
Rest of the paper is organized as follows: In section 2,
proposed modified ABC (FABCLS) is explained. In
Section 3, performance of the proposed strategy is
analyzed via numerical experiment. Finally, in section 4,
paper is concluded.
II . FAST-CONVERGENT ARTIFICIAL BEE COLONY WITH
AN ADAPTIVE LOCAL SEARCH
This section explains the proposed modified ABC
algorithm. In this paper a new self adaptive local search
strategy is incorporated with ABC. The motivation for the
proposed modifications is described as follows:

SALS process starts with initial range (w1 = 1, w2 = 1)

and generates two points in this interval by diminishing
the edges of the range through a greedy way using the
equations (2) and (3). At a time, either of these equations


is executed depends which f(w1) or f(w2) has better
fitness. If w1 provides better fitness then edge w2 of the
range shrinks towards w1 otherwise w1 shifts itself near
to w2. The detailed implementation of SALS can be seen
in Algorithm 1.
w1 = w1 + (w2 w1) t / maxiter
w2 = w2 (w2 w1) t / maxiter

Algorithm 2;
Calculate f(xnew1) and f( xnew2);
if f(xnew1) < f(xnew2) then
w2 = w2 (w2 w1) itercoun / maxiter ;
if f(xnew1) < f(xbest) then
xbest = xnew1;
end if
else
w1 = w1 + (w2 w1) itercount / maxiter ;
if f(xnew2) < f(xbest) then
xbest = xnew2;
end if
end if
Set itercount = itercount + 1;
end while

(2)
(3)

where, t and maxiter are the current iteration counter and

maximum allowable iterations in the local search
algorithm respectively.
Algorithm 1 terminates when either iteration counter
exceeds the maximum iteration allowable in local search
or absolute difference between w1 and w2 falls below a
user defined parameter .
In algorithm 2, D is the dimension of the problem,
U(0, 1) is a uniformly distributed random number in the
range (0, 1), pr is a perturbation rate and is used to
control the amount of disturbance in the best solution xbest
and xk is a randomly selected solution in the population.
As the modifications are proposed to improve the
convergence speed of ABC, while maintaining the
diversity in the swarm. Therefore, the proposed strategy is
named as Fast-convergent Artificial Bee Colony with an
adaptive Local Search (FABCLS).
The FABCLS is composed of four phases: employed bee
phase, onlooker bee phase, scout bee phase. The last
phase, namely self adaptive local search phase is executed
after the completion of scout bee phase. The pseudo-code
of the proposed FABCLS algorithm is shown in Algo 3.

Algorithm 2: New solution generation:

Input: w and best solution xbest ;
for j = 1 to D do
if U(0, 1) < pr then
xnewj = xbestj + w(xbestj xkj );
else
xnewj = xbestj ;
end if
end for
Return xnew
Algorithm 3: Accelerating ABC using adaptive local
search (FABCLS):
Initialize the population and control parameters;
while Termination criteria is not satisfied do
Step 1: Employed bee phase.
Step 2: Onlooker bees phase to update the food
sources based on their profitability.
Step 3: Scout bee phase to determine the new
food sources for exhausted food sources.
Step 4: Apply self adaptive local search
strategy phase to exploit the best solution found
so far using Algorithm 1
end while

Algorithm 1: Self Adaptive Local Search (SALS)

Strategy:
Input: Optimization function M inf(x), the best solution
xbest;
Initialize termination parameters , maximum number of
iteration counter maxiter and variables w1 = 1, w2 =1
and itercount = 1;while (|w1 w2| > and itercount
maxiter ) do
Generate two new solutions xnew1 and xnew2 from xbest by
using w1 and w2, respectively using

Return the best solution.


III . EXPERIMENTAL RESULTS AND DISCUSSION

and other considered algorithms in terms of average

number of function evaluations (AFE),
standard
deviation (SD), mean error (ME) and success rate (SR). It
can be observed from Table II that FABCLS outperforms
the considered algorithms most of the time in terms of
accuracy, reliability and efficiency. Some other statistical
tests like acceleration rate (AR) [19], and boxplots have
also been done in order to analyze the algorithms output
more intensively.

A. Test problems under consideration

To validate the effectiveness of the proposed algorithm

FABCLS, 15 mathematical optimization problems ( f1 to
f15 ) of different characteristics and complexities are taken
into consideration (listed in Table 1).These all problems
are continuous in nature. Test problems f1 f15 are taken
from [15].
B. Experimental setting

D. Statistical Analysis

The results obtained from the proposed FABCLS are

stored in the form of success rate, average number of
function evaluations, standard deviation of the fitness and
mean error. Results for these test problems (Table I) are
also obtained from the basic ABC and recent variant of
ABC named Modified AB (MABC) [16] for the
comparison purpose. The following parameter setting is
adopted while implementing our proposed and other
considered algorithms to solve the problems:

The number of simulations/run =100,

Colony size NP = 50 [17-18] and Number of food

sources SN = NP/2, ij = rand[1, 1] and limit =

DimensionNumber of food sources=D SN[11,
16],

C = 1.5 [8],

The terminating criteria: Either acceptable error

(Table I) meets or maximum number of function
evaluations (which is set to be 200000)is reached,

The proposed local search in ABC runs either 10

times (based on empirical experiment) for each
iteration or = 0.001 whichever comes earlier in
algorithm.

Parameter settings for the other considered

algorithms ABC and MABC are adopted from
their original articles.
C. Results Analysis of Experiments Results Analysis

Algorithms ABC, MABC and FABCLS are compared

based on SR, AFE, and ME. First SR of all these
algorithms is compared and if it is not possible to
distinguish the performance of algorithms based on
SR then comparison is made on the basis of AFE. ME is
used for comparison if the comparison is not possible on
the basis of SR and AFE both. It is clear from Table II,
when the results of all functions are evaluated together,
the FABCLS algorithm is the cost effective algorithm for
most of the functions.
Since boxplot [20] can efficiently represent the
empirical distribution of results, the boxplots for average
number of function evaluations for all algorithms
FABCLS, ABC, and MABC have been represented in
Figure 1.Figure 1 shows that FABCLS is cost effective in
terms of function evaluations as interquartile range and
median of average number of function evaluations are
very low for FABCLS.
Further, we compare the convergence speed of
the considered algorithms by measuring the AFEs. A
smaller AFEs means higher convergence speed. In order
to minimize the effect of the stochastic nature of the
algorithms, the reported function evaluations for each test
problem is averaged over 100 runs.
In order to compare convergence speeds, we use
the acceleration rate (AR) which is defined as follows,
based on the AFEs for the two algorithms

of

Experiments
Table II presents the numerical results for benchmark
problems of Table 1 with the experimental settings shown
in section B. Table II shows the results of the proposed


V. REFERENCES
[1]

[2]

[3]

Fig 1: Boxplots graph for Average number of function

evaluation

[4]

ALGO and FABCLS: the AFEs for the two algorithms

ALGO and FABCLS:

[5]

AR AFEALGO/AFEFABCLS

(4)

where, ALGO2 {ABC, and MABC} and AR > 1 means

FABCLS is faster. In order to investigate the AR of the
proposed algorithm as compare to the considered
algorithms, results of Table II are analyzed and the value
of AR is calculated using equation (4).
Table III shows a comparison between FABCLS ABC, and FABCLS - MABC in terms of AR. It is clear
from the Table III that convergence speed of FABCLS is
better than considered algorithms for most of the
functions.
IV . CONCLUSION

[6]

[7]

[8]

[9]

ABC is a simple algorithm with very less parameters with

drawbacks like premature convergence and poor in
exploitation. This article proposed a self adaptive local
search strategy and incorporated it with ABC to enhance
the exploitation capability. The proposed algorithm has
been extensively compared with other recent variants of
ABC namely, MABC. Through the extensive
experiments, it can be stated that the proposed algorithm
is a competitive algorithm to solve the continuous
optimization problems.

[10]

[11]

[12]

B. Akay and D. Karaboga. A modified artificial bee

colony algorithm for real-parameter optimization.
Information Sciences, doi:10.1016/j.ins.2010.07.015,
2010.
J. Kennedy and R. Eberhart. Particle swarm
optimization.
In
Neural
Networks,
1995.
Proceedings.,IEEE International Conference on, volume
4, pages 19421948. IEEE, 1995.
K.M. Passino. Biomimicry of bacterial foraging for
distributed optimization and control. Control Systems
Magazine, IEEE, 22(3):5267, 2002.
K.V. Price, R.M. Storn, and J.A. Lampinen. Differential
evolution: a practical approach to global optimization.
Springer Verlag, 2005.
J. Vesterstrom and R. Thomsen. A comparative study of
differential evolution, particle swarm optimization, and
evolutionary algorithms on numerical benchmark
problems. In Evolutionary Computation, 2004.
CEC2004. Congress on, volume 2, pages 19801987.
IEEE, 2004.
D. Karaboga. An idea based on honey bee swarm for
numerical optimization. Techn. Rep. TR06, Erciyes
Univ. Press, Erciyes, 2005.
D. Karaboga and B. Akay. A comparative study of
artificial bee colony algorithm. Applied Mathematics and
Computation, 214(1):108132, 2009.
G. Zhu and S. Kwong. Gbest-guided artificial bee colony
algorithm for numerical function optimization. Applied
Mathematics and Computation, 217(7):31663173,
2010.
W. Gao and S. Liu. A modified artificial bee colony
algorithm. Computers & Operations Research, 2011.
A. Banharnsakun, T. Achalakul, and B. Sirinaovakul.
The best-so-far selection in artificial bee colony
algorithm. Applied Soft Computing, 11(2):28882901,
2011.
Dervis Karaboga and Bahriye Akay. A modified
artificial bee colony (abc) algorithm for constrained
optimization problems. Applied Soft Computing,
11(3):30213031, 2011.
F. Kang, J. Li, and Z. Ma. Rosenbrock artificial bee
colony algorithm for accurate global optimization of
numerical
functions.
Information
Sciences,
181(16):35083531, 2011.


[13]

Yuhui Shi and Russell C Eberhart. Empirical study of

particle swarm optimization. In Evolutionary
Computation, 1999. CEC 99. Proceedings of the 1999
Congress on, volume 3. IEEE, 1999.
[14] H.Wang, D.Wang, and S. Yang. A memetic algorithmwith
adaptive hill climbing strategy for dynamic optimization
problems. Soft Computing-A Fusion of Foundations,
Methodologies and Applications,13(8):763780, 2009.
[15] M.M. Ali, C. Khompatraporn, and Z.B. Zabinsky. A
numerical evaluation of several stochastic algorithms on
selected continuous global optimization test problems.
Journal of Global Optimization,31(4):635672, 2005.
[16] M. Dorigo and G. Di Caro. Ant colony optimization: a
new meta-heuristic. In Evolutionary Computation, 1999.
CEC 99. Proceedings of the 1999 Congress on, volume
2. IEEE, 1999.
[17] K. Diwold, A. Aderhold, A. Scheidler, and M.
Middendorf. Performance evaluation of artificial bee
colony optimization and new selection schemes.
Memetic Computing, pages 114, 2011.
[18] M. El-Abd. Performance assessment of foraging
algorithms vs. evolutionary algorithms. Information
Sciences, 182(1):243263, 2011.
[19] S. Rahnamayan, H.R. Tizhoosh, and M.M.A. Salama.
Opposition-based differential evolution. Evolutionary
Computation, IEEE Transactions on, 12(1):6479, 2008.
[20] D.F. Williamson, R.A. Parker, and J.S. Kendrick. The
box plot: a simple visual method to interpret data.
Annals of internal medicine, 110(11):916, 1989.


Table I
Test Problems

TestProblem

ObjectiveFunction
f1( ) =

Parabola Sphere

f2

DeJong'sf4
f3

Griewank
Rosenbrock**

Ackley

f5

f7

SalomonProblem SAL
f 0,0,0,0,0,0 0

2 2

) +

) )

f8

1cos 2

Axisparallel
hyperellipsoid

20

0.1

f9( )=
f10

Sumofdifferentpowers
Stepfunction 100,100 f
0.5 x
0.5 0

f11
f12

Acceptable
Error

5.12,5.12

f 0 0

30

1.0E 05

5.12,5.12

f 0 0

30

1.0E 05

600,600

f 0 0

30

1.0E 05

30,30

f 0 0

30

1.0E 02

1,1

f 0 0

30

1.0E 05

10,10

f 0 0

30

1.0E 05

0,

fmin 9.66015

10

1.0E 05

100,100

f 0 0

30

1.00E01

5.12,5.12

f 0 0

30

1.0E 05

f 0 0

30

1.0E 05

30

1.0E 05

f 0
10
D 1

1.0E 05

1,1

100,100

xI

5,5

whereI cos 4

f 0.5
0.5

x
0

fmin

Neumaier3Problem
NF3 Neumaier,2003b

f13

Rotatedhyperellipsoid
function

D2,D2
f14

f15

10sin2

10sin2

2x 1
2 where

OptimumValue

f6

Michalewiczfunction

Levimontalvo1

f4( )=

Alpine

Invertedcosinewave
function Masters 5,5
f 000..0 D 1

SearchRange

10

1.00E01

65.536,
65.536

f 0 0

30

1.0E 05

10,10

f 0 0

30

1.0E 05


Table II
Comparison of the results of test problems


Table III
Acceleration Rate (AR) of FABCLS as compared to the ABC and MABC, TP: Test Problems

TP
f1
f2
f3
f4
f5
f6
f7
f8
f9
f10
f11
f12
f13
f14
f15

ABC
1.789789021
1.630366328
1.427561747
0.893905381
1.538661587
1.732237123
1.692745642
6.837846389
1.715578036
4.593910368
1.404979294
2.563734002
15.03419457
1.645067827
1.831326494

MABC
1.96079635
3.844045848
1.36838962
1.034286084
1.35774579
4.462132855
2.262139944
1.190251731
1.942186565
2.708310738
1.935523278
1.952887338
10.30569318
1.945601992
2.123712958