PSO的向量整体修订策略和局部跳出策略

Vector correction and jump out of local optimum strategy for PSO

针对传统PSO方法对CEC2005(The 2005 IEEE Congress on evolutionary computation)中的25个benchmark函数搜索效果较差的问题,提出了'向量整体修订'和'局部跳出'两种改进策略。改变PSO方法中粒子在每一维上的修订相互独立的传统机制,按某一概率将粒子作为整体进行修正,当群体最优长时间不变或变化值小于一定阈值时,为跳出局部最优,按某一概率重新定义群体最优或初始化群体。通过实验证明了改进后的PSO方法对CEC2005中的测试问题的有效性。

Using traditional Particle Swarm Optimization （PSO） the searching results for some new benchmark functions, e. g. the 25 benchmark functions in CEC2005, are not satisfactory. An improved version of PSO was designed to suit for new benchmark functions in CEC2005. Two improvement strategies, named Vector correction strategy and Jump out of local optimum strategy, were employed in this improved PSO. When the swarm optimum remains invariable for a long time, The improve PSO can revises the whole particle vector and re-initialize the swarm or generate a new swarm optimum according to certain probability. The improved PSO was tested by the 25 benchmark functions in CEC2005, and the experimental results show that the search efficiency and the ability to jump out of the local optimum of the improved PSO are significantly improved.