Compound Particle Optimization Using Speciation for Multimodal Function Optimization

摘要

Multimodal optimization problems pose a new challenge to evolutionary computation, since they usually not only require a search for one global optimum, but also simultaneously locating multiple optima. This paper presents a new variant of particle swarm optimization, which incorporates the notion of speciation into the compound particle optimization for solving multimodal functions. In the proposed species-based compound particle swarm optimization (SCPSO), several species containing compound particles are adaptively formed according to their similarity at each iteration step. The corresponding techniques of the compound particle, which are inspired by physics mechanisms, provides successive local improvements for each species to precisely and quickly identifying multiple global optima. Experiments on multimodal test functions suggest that SCPSO is more computationally efficient than the conventional species-based PSO.