Multiobjective sorting-based learning particle swarm optimization for continuous optimization

Abstract

Canonical particle swarm optimization (PSO) utilizes the historical best experience and neighborhood’s best experience of particle through linear summation to guide its search direction. Such a learning strategy is easy to use, but is inefficient when searching in the complex problem space since the global best individual only considers the fitness value but always ignores the diversity information. Hence, designing learning strategies where the guidance exemplar simultaneously considers the fitness value and diversity have become one of the most salient and active PSO research topics. In this paper, a multiobjective sorting-based learning (MSL) strategy for PSO is proposed and this modified PSO is named as multiobjective sorting-based learning particle swarm optimization. The MSL strategy can guide particles to fly in better direction by constructing a guidance exemplar with better fitness value and diversity. Since the fitness value and diversity are simultaneously considered to construct the guidance exemplar instead of the global best individual in canonical PSO, a better balance between exploration and exploitation can be achieved. The proposed strategy is applied to the original PSO algorithm, as well as several advanced PSO variants. Experimental results on sixteen benchmark problems show that the proposed strategy is an effective approach to enhance the performance of most PSO algorithms studied in terms of solution quality, convergence speed and algorithm reliability.