Self-Adaptive Particle Swarm Optimization

Abstract

Particle swarm optimization (PSO) has been used to solve a wide variety of optimization problems. The basic PSO algorithm contains a number of control parameters, including the inertia weight, w, and the acceleration coefficients, c1 and c2. The PSO, as an optimization algorithm, is ideally suited to optimize its own parameters. This paper proposes that the control parameters of PSO be optimized in a secondary swarm where each position vector component of each particle contains a prospective PSO control parameter (i.e. w, c1 and c2) of the main swarm. This approach relieves the user from specifying appropriate parameters when using PSO. Application of the self-adaptive particle swarm optimizer (SAPSO) to 12 well known test functions shows that SAPSO managed to reach pre-specified values quicker than an adaptive PSO using fitness rank to update the inertia weight.