Study of particle swarm optimization algorithms using message passing interface and graphical processing units employing a high performance computing cluster

Abstract

Particle Swarm Optimization (PSO) is a heuristic technique that have been used to solve problems where many events occur simultaneously and small pieces of the problem can collaborate to reach a solution. Among its advantages are fast convergence, large exploration coverage, and adequate global optimization; however to address the premature convergence problem, modifications to the basic model have been developed such as Aging Leader and Challengers (ALC) PSO and Bioinspired Aging (BAM) PSO. Being these algorithms parallel in nature, some authors have attempted different approaches to apply PSO using MPI and GPU. Nevertheless ALC-PSO and BAM-PSO have not been implemented in parallel. For this study, we develop PSO, ALC-PSO and BAM-PSO, through MPI and GPU using the High Performance Computing Cluster (HPCC) Agave. The results suggest that ALC-PSO and BAM-PSO reduce the premature convergence, improving global precision, whilst BAM-PSO achieves better optimal at the expense of significantly increasing the algorithm computational complexity.