Statistics Based Model Evaluation and Parameter Selection for Particle Swarm Optimization Algorithm

It is proper that to evaluate algorithm's performance using statistics for stochastic search optimization such as PSO. In this paper, we do performance statistics and analysis for some different situation taking Rosen Brock function as example, from statistics we think that local model is better than global model in avoiding premature, and neighborhood size is not important, number of particles should be enough large to distributed as uniformly as possible in search space, and the same times it should smaller than iteration times to complete information flow among particles. Based on this statistics we present a weighted PSO model, test result shows that our model's performance is better than basic PSO model.