Accelerating Artificial Bee Colony Algorithm with
New Multi-Dimensional Selection Strategies
1
st
Wenqi Xiao, 2
nd
Haolun Li, and 4
th
Hao Gao
The Institute of Advanced Technology. Nanjing
University of Posts and Telecommunications,
Nanjing, China
wuhanxwq@gmail.com, lhl219319@gmail.com,
tsgaohao@gmail.com
3
rd
Jiajun Yan
The college of electronic and optical engineering
&colleg
e of microelectronics. Nanjing
University
of Posts
and
Telecommunications,
Nanjing,
Ch
ina
Abstract—As a new intelligent swarm optimization algorithm,
artificial bee colony (ABC) algorithm has been used to solve a lot
of function optimization problems and successfully applied in
many engineering fields. However, the single-dimensional search
feature of the ABC algorithm results in a slower convergence rate.
In this paper, we develop an improved ABC algorithm with new
multi-dimension selection strategies (MDSABC) to enhance the
search efficiency and improve the accuracy of the solution by
selecting how many dimensions and which dimensions are
updated. It specifically includes a multi-dimensional update
strategy, neighbor and dimension selection strategies. The
property of the MDSABC algorithm is tested on variety of
benchmark functions with the original ABC algorithm and some
classic improved ABC algorithms published in recent years. The
experimental results show that the MDSABC algorithm can
obviously improve the search efficiency and better than other
algorithms.
Keywords—artificial bee colony; multi-dimensional; search
strategy; convergence rate
I. I
NTRODUCTION
Optimization is a crucial problem in many fields of science
and engineering. Many optimization problems in practical
engineering need to find an optimal solution in complex and
large search space, including linear, non-linear; continuous,
discrete; single peak, multi-peak and other types [1,2]. It is
difficult to get the desired results with traditional optimization
methods.
The rapid development of science and technology make
people pay more and more attention to efficiency, people get
inspiration by summarizing the laws of evolution in nature to
deal with intricate optimization problems. Through the study of
the intelligent behavior of the biological groups such as ants,
bees, birds and anything else, researchers have proposed a
number of intelligent optimization algorithms. For example, ant
colony optimization (ACO) imitates ants selecting the shortest
path of food source to solve discrete problems, this model was
presented by the Italian scientist Dorigo in 1992 [3]. Learning
the experiences of birds looking for habitat laws, Dr. Kennedy
and Eberhart proposed a particle swarm algorithm (PSO) in
1995 [4]. Through the different roles of the bee division of labor,
the colony showed complex intellectual behavior during the
feeding process [5], Teodorovic proposed a colony optimization
algorithm based on the propagation mechanism of bees and
Karaboga proposed artificial bee colony algorithm (ABC)
according to the bee feeding principle in 2005 [6].
ABC algorithm has the advantages of simple principle, less
control parameters, strong global search capability and great
robustness compared to other algorithms. However, since ABC
uses a single-dimensional search mechanism, it has certain
limitations and impacts on performance, such as the slow
convergence speed and the poor local search ability [7]. In recent
years, many researchers have improved the original ABC.
GABC improve the accuracy of standard ABC by learning from
the best location [8]. SABC adopts three iterative formulas to
adaptive balance the exploration and the exploitation [9]. The
updating mechanism enables bees always to track better
individuals in STABC to improve search efficiency [10]. SESA
uses the successful search directions, the neighbor selection and
the failure reuse three methods to accelerate update effectiveness
together [11].
After ample theoretical evidence and experimental results,
we present a MDSABC to improve the precision of the original
ABC based on the idea of a multi-dimensional change and an
appropriate dimension updates strategies. At the stage of main
exploration, the dimension of each update is incremented, which
gradually improves the search efficiency and conquers
premature convergence. At the stage of main exploitation,
MDSABC chooses another one dimension which is most likely
to improve the result by comparing to the best position, which
improves the search precision with a high probability.
The paper is organized as follows: the original ABC
algorithm is reviewed in Section II, Section III describes
MDSABC in details, compared on variety of benchmark
functions with some classic improved ABC algorithms in recent
years and lists the results in Section IV. Finally, writes the
conclusion of the article in Section V.
2018 Tenth International Conference on Advanced Computational Intelligence (ICACI)
March 29–31, 2018, Xiamen, China
978-1-5386-4362-4/18/$31.00 © 2018 IEEE