Differential mutation and novel social learning particle swarm optimization algorithm

Abstract

Social Learning Particle Swarm Optimization (SLPSO) is an improved Particle Swarm Optimization (PSO) algorithm, which greatly improves the optimization performance of PSO. However, SLPSO still has some deficiency, such as poor balance between exploration and exploitation and low search efficiency, so that it cannot yet do well in solving many complex optimization problems. Thus, this paper proposes an improved SLPSO algorithm, that is, Differential mutation and novel Social learning PSO (DSPSO). Firstly, in order to balance exploration and exploitation better, a dynamic inertia weight is introduced to replace the random inertia weight of SLPSO, and a single-example learning approach and an example-mean learning one are proposed to replace the imitation component and the social influence component of SLPSO respectively. Secondly, the dimension-based velocity updating equation of SLPSO is divided into two particle-based updating equations with the two approaches, and the two are executed alternately to form a novel social learning PSO (NSLPSO), which enhance the exploitation of SLPSO. Finally, a dynamic differential mutation strategy is used in NSLPSO to update the three best particles to enhance the exploration to obtain DSPSO. Experimental results on the complex functions from CEC2013 reveal that DSPSO outperforms SLPSO and quite a few state-of-the-art and classic PSO variants.