Multi-Sub-Swarm PSO algorithm for multimodal function optimization

Abstract

Inspired by individuals’ clustering to sub-swarm through learning between individuals and between sub-swarms, we propose a new algorithm called dynamic multi-sub-swarm particle swarm optimization (MSSPSO) algorithm for multimodal function with multiple extreme points. In the evolutionary process, the initial particles, that are separately one sub-swarm, merge into bigger sub-swarms by calculating a series of dynamic parameters, such as swarm distance, degree of approximation, distance ratio and position accuracy. Simulation results show that, in single-peak function optimization, MSSPSO algorithm is feasible but search speed is not superior to the PSO algorithm, while in a multimodal function optimization, MSSPSO algorithm is more effective than PSO algorithm, which cannot locate the required number of extreme points.