DMPSO: Diversity-Guided Multi-Mutation Particle Swarm Optimizer

Dongping Tian,Zhongzhi Shi,Xiaofei Zhao

doi:10.1109/access.2019.2938063

Abstract

Particle swarm optimization (PSO) is a population based meta-heuristic search technique that has been widely applied to deal with various optimization problems. However, like other stochastic methods, PSO also encounters the problems of entrapment into local optima and premature convergence in solving complex multimodal problems. To tackle these issues, a diversity-guided multi-mutation particle swarm optimizer (abbreviated as DMPSO) is presented in this paper. To start with, the chaos opposition-based learning (OBL) is employed to yield high-quality initial particles to accelerate the convergence speed of DMPSO. Followed by, the self-regulating inertia weight is leveraged to strike a balance between the exploration and exploitation in the search space. After that, three different kinds of mutation strategies (gaussian, cauchy and chaotic mutations) are used to maintain the potential diversity of the whole swarm based on an effective diversity-guided mechanism. In particular, an auxiliary velocity-position update mechanism is exclusively applied to the global best particle that can effectively guarantee the convergence of the DMPSO. Finally, extensive experiments on a set of well-known unimodal and multimodal benchmark functions demonstrate that DMPSO outperforms most of the other tested PSO variants in terms of both the solution quality and its efficiency.

Highlights

Inspired by social behavior observed in nature, such as schools of fish, flocks of birds, swarms of bees, and even human social behavior, particle swarm optimization was first introduced in the mid-1990s [25], which has the characteristics of swarm intelligence, intrinsic parallelism, simple iteration format, negligible parameter settings and inexpensive computation
EXPERIMENTS BASED ON BASIC BENCHMARK FUNCTIONS To ascertain the performance of the proposed DMPSO, a series of experiments are first conducted on a set of well-known benchmark functions consisting of six global optimization problems
To enhance its optimization performance, the chaos opposition-based learning is first applied to yield high-quality initial particles to accelerate its convergence speed, the self-regulating inertia weight is leveraged to strike a balance between the exploration and exploitation in the search space, afterwards three different mutation strategies are used to maintain the potential diversity of the whole swarm, especially an auxiliary velocity-position update mechanism is exclusively employed to the global best particle to guarantee the convergence of the proposed Particle swarm optimization (PSO)

Summary

INTRODUCTION

Inspired by social behavior observed in nature, such as schools of fish, flocks of birds, swarms of bees, and even human social behavior, particle swarm optimization was first introduced in the mid-1990s [25], which has the characteristics of swarm intelligence, intrinsic parallelism, simple iteration format, negligible parameter settings and inexpensive computation By virtue of these advantages, PSO has been extensively applied in many fields [8], [14], [15], [28], [38], [41], [49], [57], [65], [66], [71], [72] since its introduction. A large amount of research effort has been devoted to enhancing the performance of PSO From the literature, these previous works can be roughly divided into the following categories: (i) swarm initialization, (ii) parameter selection, (iii) non-parametric update, (iv) multi-swarm scheme, and (v) hybrid mechanism.

RELATED WORK

CHAOS THEORY

EXPERIMENTAL RESULTS AND DISCUSSION

CONCLUSION AND FUTURE WORK