A revised bare bone particle swarm optimizer and its variant

Abstract

Bare bone particle swarm optimization (BPSO), derived from particle swarm optimization, is a simple optimization technique with the advantage of without using parameters, except the number of particles and generations. Inspect the model of BPSO carefully, one can found that if a particle is restricted to move to a new position only when the new position is better than its original position, the particle then retains the best position it ever found. Based on this observation, all personal best particles are no longer required. In this paper, a revised BPSO is proposed that further eliminate personal best particle leading to more efficient utilization of memory, especially when dealing with large scale problems or in microprocessor based applications. Since this revision is comparable to BPSO, it will be referred to RBPSO in short. In addition, to enhance the performance of RBPSO, a variant, denoted as RBPSOx, is also proposed. Numerical results obtained from testing on ten benchmark functions with 30 and 50 dimensions demonstrate that the proposed modifications are feasible and outperform original BPSO especially for multimodal functions.