Research on PSO with Clusters and Heterogeneity

Particle Swarm Optimization(PSO) algorithm easily falls into local optimal solution when solving complex multimodal function optimization problem.Researches show that dynamic topology and variable parameters can improve the diversity of swarm to improve the situation.However,the effect of topology and parameters is rarely considered simultaneously.In this paper,a new PSO algorithm based on clustering is proposed.It takes K-means clustering method to divide the swarm into different neighborhoods dynamically.These neighborhoods have different number of particles and are heterogeneous clusters.A Ring-structure is applied to exchange information among clusters.Furthermore,a novel discriminating method is proposed to detect the exploring stage of a cluster.Each particle adjusts its parameters automatically according to the exploring stage of its cluster.The results of experiments show that the operations above can improve diversity and energetic of the particles,increase exploring ability and convergence,and reduce the dependence of initial election of parameters.