An Improved Clonal Algorithm in Multiobjective
Optimization
Jianyong Chen, Qiuzhen Lin and Qingbin Hu
Department of Computer Science and Technology, Shenzhen University, Shenzhen, P.R. China
Email: jychen@szu.edu.cn, qiuzheng658@163.com, huqb@szu.edu.cn
Abstract—In this paper, we develop a novel clonal algorithm
for multiobjective optimization (NCMO) which is improved
from three approaches, i.e., dynamic mutation probability,
dynamic simulated binary crossover (D-SBX) operator and
hybrid mutation operator combining with Gaussian and
polynomial mutations (GP-HM operator). Among them, the
GP-HM operator is controlled by the dynamic mutation
probability. These approaches adopt a cooling schedule,
reducing the parameters gradually to a minimal threshold.
By this means, they can enhance exploratory capabilities,
and keep a desirable balance between global search and
local search, so as to accelerate the convergence speed to the
true Pareto-optimal front. When comparing NCMO with
various state-of-the-art multiobjective optimization
algorithms developed recently, simulation results show that
NCMO evidently has better performance.
Index Terms—multiobjective optimization; immune
algorithm; clonal selection; hybrid mutation
I. INTRODUCTION
In some real world problems, there exist many
multiobjective optimization problems (MOPs) in practical
engineering and scientific applications. Without any
further information about the relationship among the
objectives, it is possible to obtain a set of optimal
solutions in which each solution is equally preferable
when regarding all criteria considered. As a result, in
order to provide multiple candidate selections for
different applications, we aim at as many representative
optimal solutions as possible.
Evolutionary algorithms have the ability to process sets
of solutions in parallel and explore big search spaces in
reasonable time. Therefore, Evolutionary algorithms have
been recognized to be well suited to MOPs. The ability to
handle complex problems, involving features such as
discontinuities, multimodality, disjoint feasible spaces
and noisy function evaluations, reinforces the potential
effectiveness of evolutionary algorithms in MOPs [1]. In
the last few years, numerous competent Evolutionary
algorithms have been proposed as the state-of-the-art
algorithms for MOPs. For example, NSGA-II [2] was
proposed with a fast nondominated sorting technology,
elitism and crowding-distance assignment. SPEA-II [3]
was proposed with a fine-grained fitness assignment
strategy, an enhanced archive truncation method and a
new density estimation technique. The pareto archived
evolution strategy (PAES) [4]
was proposed with simple
(1+1) evolution strategy. All of them tried to design
effective and efficient technologies to improve the
abilities of the convergence and the diversity.
On the other hand, Artificial Immune Systems (AIS)
have been developed since 1990s as a new branch in
computational intelligence, which simulate the defense of
the human immune system against bacteria, viruses and
other invaders [5]. A number of AIS models have been
found applications in various fields such as machine-
learning and pattern-recognition tasks [6], network
security [7], scheduling [8], data mining [9] and others
[10, 11]. In recent years, AIS are also applied in MOPs
and studies show their remarkable performances. For
example, Coello Coello and Cortes [12] presented a
multiobjective immune system algorithm (MISA) based
on the clonal selection principle. Freschi and Repetto [13]
proposed a vector artificial immune system (VAIS) based
on the artificial immune network. Gong and Jiao et al. [14]
proposed a nondominated neighbor-based immune
algorithm (NNIA), by using a novel nondominated
neighbor-based selection technique, proportional cloning,
heuristic search operators and elitism.
For MOPs, one of the key points is to find a uniformly
distributed approximation set that is as close as possible to
the Pareto-optimal front. Based on the notion that the
preservation of representative solutions can effectively
drive the algorithm toward the Pareto-optimal front [15],
most of current studies pay much attention to fitness
assignment and population maintenance such as
crowding-distance [2], grid mechanism [16] and clustering
technology [17]. However, rare studies pay attention to
search operators, which can obviously accelerate the
convergence speed.
In this paper, we propose a novel clonal algorithm for
MOPs (NCMO) based on the improvement of search
operators, aiming to speed up the convergence. Three
novel approaches, i.e., dynamic mutation probability,
dynamic simulated binary crossover (D-SBX) operator
and hybrid mutation operator combining with Gaussian
and polynomial mutations (GP-HM operator), are
proposed in this paper. Similar to the temperature
parameter in simulated annealing, these approaches also
Manuscript received January 1, 2009. The work was supported b
y
the National Natural Science Foundation of China (Grant No:
60703112).
Jianyong Chen, Qiuzhen Lin and Qingbin Hu are with the
Department of Computer Science and Technology, Shenzhen
University, Shenzhen, P.R. China (e-mail: jychen@szu.edu.cn;
qiuzheng658@163.com; huqb@sz u.edu.cn)
976 JOURNAL OF SOFTWARE, VOL. 4, NO. 9, NOVEMBER 2009
doi:10.4304/jsw.4.9.976-983