Computational intelligence PC tools
Computational intelligence PC tools
Hi-index | 0.00 |
Population members of the classical PSO quickly converge onto a smaller region of the objective function landscape, which helps to refine the discovered optimum, but the searching ability of the algorithm collapses. This paper proposes modification to the way information flows between the global best particle and the rest of the particles to resist particles clustering. Particles are ranked according to their fitness value such that each rank is single particle and a particle learns only from the particle one rank above. Global best particle learns only from its own experience. The proposed version of PSO is named as RPSO and experiments on test bed functions show not only RPSO particles resisted clustering but stability has also been observed in RPSO results. The downside of RPSO was its slow rate of convergence, which was improved by nominating certain particles from the whole population as diggers with learning topology of the classical PSO. This version was named RPSO-D and experiments were conducted to show its superiority over the classical PSO, both in terms of stability and rate of convergence.