A Center Multi-swarm Cooperative Particle
Swarm Optimization with Ratio
and Proportion Learning
Xuemin Liu, Lili
(&)
, and Jiaoju Ge
(&)
Harbin Institute of Technology (Shenzhen), Shenzhen 518055, China
ximlli@126.com, jiaoge@hit.edu.cn
Abstract. This paper presents a center multi-swarm cooperative PSO with ratio
and proportion learning (CMCPSO-RP), employing two well-known psychol-
ogy theories. In the original MCPSO-CC, the convergence speed can be
accelerated which comes at decreasing the diversity of sub-swarms, suffering
from premature convergence. There is no mechanism to guarantee every pos-
sible region of the search space could be searched. To tackle this problem, all
best particles from each sub-swarm can be collected and sent to master swarm to
maintain a population of potential solutions. This process is less prone to
becoming trapped in local minima, but typically has lower efficiency of itera-
tions. To balance the ability of exploration and exploitation, a ratio and pro-
portion learning strategy is proposed by empowering the searching particles
with human-like thinking and cognitive process, inspired by Cognitive Load
Theory and Human Problem Solving Theory. In our approach, a reasonable ratio
design can be not only a way to exhibit a solution quality versus speed tradeoff,
but also make CMCPSO-RP more in line with the laws of regular learning in
nature. Application of the newly developed PSO algorithm on several bench-
mark optimization problems shows a marked improvement in performance over
the comparison algorithms on all test functions.
Keywords: Multi-swarm
Center communication Ratio and proportion
learning
1 Introduction
Particle swarm optimization (PSO) was first developed by Kennedy and Eberhart in
1995 [1]. For general problems, it provides efficient and satisfactory solutions like other
meta-heuristic methods and search algorithms [2], such as genetic algorithms (GA) [3]
and differential evolution (DE) [4]. Owing to its simple concept and high efficiency,
PSO has been used to solve a range of optimization problems [5], including neural
network training [6], data mining [7], feature selection [8], just to name a few. Con-
sidering the fact that the original PSO works better in simple and low-dimensional
search space, various attempts have been made to improve its performance from
aspects of topology, parameter control mechanism, learning strategy and hybridization.
For solving complex multimodal problems, numerous multi-swarm techniques, aiming
to improve the population diversity, are studied, such as dynamic multi-swarm PSO
© Springer International Publishing AG 2017
Y. Tan et al. (Eds.): ICSI 2017, Part I, LNCS 10385, pp. 189–197, 2017.
DOI: 10.1007/978-3-319-61824-1_21