EPSO enhanced by adaptive scaling and sub-swarms

Abstract

This paper reports the positive results derived from adopting two variants for the EPSO - Evolutionary Particle Swarm Optimization method: variable's re-scaling and sub-swarms. Sub-swarms launched from the main swarm can be applied to intensify the search in promising regions of the space. Alternatively, the information regarding the dispersion of the particles along the search space can be used to create local landscapes with a spherical/ellipsoid form in an attempt to take advantage of the excellent convergence properties of metaheuristics for spherically-shaped optimization problems. The net improvement in reducing computing effort is observed in several unconstrained optimization problems and verified with ANOVA.