A set-based particle swarm optimization method

Abstract

The representation used in Particle Swarm Optimization (PSO) is an n-dimensional vector. If you want to apply the PSO method, you have to encode your problem as fix-sized vector. But many problem domains have solutions of unknown sizes as for instance in data clustering where you often don't know the number of clusters in advance. In this paper a set-based PSO is proposed which replaces the position and velocity vectors by position and velocity sets realizing this way a PSO with variable length representation. All operations of the PSO update equations are redefined in an appropriate manner. Additionally, an operator reducing set bloating effects is introduced. The presented approach is applied to well-known data clustering problems and performs better as other algorithms on them. © 2008 Springer-Verlag Berlin Heidelberg.