A non-deterministic adaptive inertia weight in PSO

Abstract

Particle Swarm Optimization (PSO) is a relatively recent swarm intelligence algorithm inspired from social learning of animals. Successful implementation of PSO depends on many parameters. Inertia weight is one of them. The selection of an appropriate strategy for varying inertia weight w is one of the most effective ways of improving the performance of PSO. Most of the works done till date for investigating inertia weight have considered small values of w, generally in the range [0,1]. This paper presents some experiments with widely varying values of w which adapts itself according to improvement in fitness at each iteration. The same strategy has been implemented in two different ways giving rise to two inertia weight variants of PSO namely Globally Adaptive Inertia Weight (GAIW) PSO, and Locally Adaptive Inertia Weight (LAIW) PSO. The performance of the proposed variants has been compared with three existing inertia weight variants of PSO employing a test suite of 6 benchmark global optimization problems. The experiments show that the results obtained by the proposed variants are comparable with those obtained by the existing ones but with better convergence speed and less computational effort.