Parallel particle swarm optimization with genetic communication strategy and its implementation on GPU

Abstract

Taking into account the advantage of high computation to communication ratio of coarse-grained parallel model, we implement coarse-grained parallel particle swarm optimization (PPSO) on Graphic Processing Unit (GPU), which is very popular for parallel computing nowadays. Meanwhile, a heuristic communication strategy called genetic migration is proposed in this paper. Numerical experimental results show that PPSO with genetic migration (PPSO_GM) can greatly improve the convergence property of particle swarm optimization (PSO), compared with PPSO with traditional unidirectional ring migration (PPSO_URM); and two orders of magnitude more speedups are achieved by PPSO_GM against serial PSO (SPSO) for all ten 100-dimensional benchmark test functions.