一种新型的多种群微粒群算法

Novel multi-population particle swarm optimizer

针对微粒群算法容易出现早熟问题,提出一种动态种群与子群混合的微粒群算法(SPSDPSO)。该算法在微粒群搜索停滞时对微粒进行分群,在子群内部通过微粒随机初始化以及个体替代策略提高优化性能,在子群进化一定代数后重新混合为一个种群继续优化,种群进化与子群进化交替进行直至满足算法终止条件。SPSDPSO的种群与子群混合进化策略增强了群体多样性,并且使得子群体之间能够进行充分的信息交流。收敛性分析表明,SPSDPSO以概率1收敛到全局最优解。函数测试结果表明,新算法的全局收敛性能有了显著提高。

To prevent the problem of premature convergence frequently appeared in particle swarm optimizer（PSO）,a shuf-fled population and subpopulations dynamic of PSO（SPSDPSO） is proposed.In the approacht,he whole population is divided into different subpopulations when particles search stagnated for certain iterations.Moreover,some individuals of subpopula-tions are re-initiated randomly and some individuals are substituted to improve the search ability further.Particles of different subpopulations are shuffled together to search for the destination after certain iterations.The processes of population and sub-populations optimization alternate are repeated until the terminal conditions satisfied.The strategy of shuffled population and subpopulations dynamic enhances the diversity of the swarm and subpopulations can exchange useful optimization informa-tion among themselves.The SPSDPSO is guaranteed to converge to the global solution with probability one.The functional test shows that SPSDPSO algorithm has advantages of convergence property.