Study Of Parallel Particle Swarm Optimization

Particle swarm optimization (PSO) is a swarm intelligence stochastic optimization algorithm originating from research of the behavior of birds, which accomplished search of the best results in complex space by individual cooperative and competitive. Recently, this is becoming an activity research and application field because of its simple structure, stable performance, higher efficiency and required very few algorithmic parameters to adjust.This paper presents the parallel particle swarm optimization (PPSO) by combining the nature parallel characters of standard PSO with parallel computation models. The PPSOs are designed based on three parallel computation models which include parallel computation model with central controller, ring-structure model with buffers, and BSP parallel computation model. First a PPSO based on the parallel model with controller(PPSO—Controller) is introduced. The results show that when appropriate period of communication is selected, the PPSO—Controller not only has perfect speedup ratio but also can improve the quality of results. Then a PPSO based on ring structure model with buffer storage is presented, which assures that the best particle diffuses to all sub-populations and improves diversity of sub-populations. Finally, a new PPSO is implemented based on BSP model (BSPPSO), which changes the pattern of standard PSO and improves the search efficiency. Compared BSPPSO and PSO, BSPPSO have better performance. This paper also researches the cooperative PSO that adopts different update strategy among all sub-populations to communicate periodically, which can find the best results. Analysis of the experiment data shows that the communication period is an important parameter that influences the convergence and other performance.