求解复杂优化问题的双层多种群粒子群优化算法  

A Bi-level multi-population particle swarm optimization algorithm for solving complicated optimization problems

作　　者：高芳[1] 崔刚[1] 吴智博[1] 刘宏伟[1] 杨孝宗[1] 

机构地区：[1]哈尔滨工业大学计算机科学与技术学院,哈尔滨150001

出　　处：《高技术通讯》2009年第5期519-524,共6页Chinese High Technology Letters

基　　金：863计划(2006AA01A103)资助项目

摘　　要：为解决粒子群优化算法存在的易早熟和精度低问题,提出了一种双层多种群粒子群优化算法。此算法采用上下两层,即下层N个基础种群和上层一个精英种群。各个基础种群相互独立进化,并从精英种群中得到优良信息指导自己的进化。上层精英种群首先通过接受各基础种群的当前最优粒子来更新自己的粒子集合,然后执行自适应变异操作,最后随机地向每一个基础种群输送出本次进化后的一个最优粒子来改进其下一轮搜索。该算法的并行双进化机制增加了群体的随机性和多样性,提高了全局搜索能力和收敛精度。实例仿真表明该算法具有较好的性能,尤其对于复杂多峰函数优化,成功率显著提高。To overcome the PSO algorithm＇s drawbacks of easily premature converging and low convergence precision, the paper proposes a new improved bi-level multi population particle swarm optimization （PSO） algorithm. This algorithm ineludes two levels： the lower level of N basic swarms and the upper level of elite swarm. These basic swarms independently evolve and obtain the advanced information to supervise their evolution. The elite swarm first accepts the current optimal particles from each basic swarm to update its particle set, then executes an adaptive mutation, and randomly outputs one of the current optimal particles to each swarm to improve its next search. The parallel dual evolving mechanism in this algorithm enhances the swarm randomicity and diversity, and improves the global search ability and converging precision. The simulations show that this algorithm has better performance, and particularly its success rate is significally increased for the multi-peak function.

关 键 词：粒子群优化(PSO) 双层多种群 精英种群 自适应变异 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]