All particles driving particle swarm optimization: Superior particles pulling plus inferior particles pushing

Abstract

In particle swarm optimization (PSO), the velocity vector is a conjecture to the descending direction of the objective function. The traditional PSO obtains such a direction using only two attractors (i.e., pb and pg). In fact, all particles may carry useful information. The particles with good fitness are capable of guiding other particles to explore promising regions; meanwhile, the particles with poor fitness values are capable of indicating the possible hopeless regions. To make use of the information carried by the whole population, an all particles driving PSO (APD-PSO) is proposed in this paper. APD-PSO utilizes superior particles as attractors and inferior particles as repellers when updating the velocity. An information interaction operator is also developed in this paper for better modeling the fitness landscape and bringing helpful noise. Comprehensive simulation experiments with statistical analysis on the results validate the excellent performance of our proposed APD-PSO. The experimental comparison to several PSO competitors shows that the proposed APD-PSO achieves very competitive optimization performance.