A modified particle swarm optimization algorithm using Renyi entropy-based clustering

Abstract

An algorithm proposed using Renyi entropy clustering to improve the searching ability of traditional particle swarm optimization (PSO) is introduced in this study. Modified PSO consists of two steps. In the first step, particles in initial population are sorted according to Renyi entropy clustering method, and in the second step, some particles are removed from population and some new particles are added instead of them based on the sorted list. Thus, a reliable new initial population is created. When using sorted list from first to last with decreasing inertia weight parameter, or from last to first with increasing inertia weight parameter, a little improved search performances have been observed on three commonly used benchmark functions. However, in other two combinations of the proposed algorithm (from last to first with decreasing inertia weight and from first to last with increasing inertia weight), little worse optimization performances than traditional PSO have been noted. These four types of the proposed algorithm were run with different exchanging rate values. Thus, the representation ability of Renyi entropy clustering on initial population and the effect of organizing inertia weight parameter were evaluated together. Experimental results which were surveyed at different exchanging rate values showed the efficiency of such evaluation.