基于两阶段变异交叉策略的差分进化算法  

Differential Evolution Algorithm Based on Two-stage Mutation and Crossing Strategy

作　　者：张大斌[1,2] 江华[1] 徐柳怡 张文生[2] 

机构地区：[1]华中师范大学信息管理学院,武汉430079 [2]中国科学院自动化研究所,北京100190

出　　处：《计算机工程》2014年第8期183-189,共7页Computer Engineering

基　　金：国家自然科学基金资助项目(70971052);中国博士后基金资助项目(2012M510607);湖北省自然科学基金创新群体项目(2011CDA116)

摘　　要：针对差分进化算法存在的收敛速度慢、稳健性差等问题,借鉴多种变异优化策略,提出一种基于两阶段不同变异交叉策略的差分进化算法。引入反向混沌搜索的初始化方法,将初始种群分为较好和较差2个子种群,两阶段依次对上一阶段改进的较好和较差2个子种群采用不同的差分进化策略,并定期将较好和较差2个子种群重新按适应值排列组合进入下一阶段,以提高种群的质量,同时克服单一差分策略的缺陷。函数仿真结果表明,与其他差分进化算法相比,该算法的收敛速度和寻优精度均得到明显改善。Differential Evolution Based on Two-stage Mutation and Crossing Strategy （TMCDE） aims to accelerate convergence and improve accuracy of Differential Evolution （DE）.The TMCDE algorithm introduces the initialization method of opposition-based chaos and stochastic diffusion search strategy,and the initial group is divided into sub-groups of both better and worse.The two stages successively improve the better and worse of two sub-groups with different DE strategies.At certain time,the two sub-groups combine to one group,and which enters to the next stage according to their fitness values.It is conducive to improve the group＇ s quality,and overcomes the shortcomings of a single differential strategy.By the benchmark function experiments,the TMCDE algorithm performs better convergence speed and optimization capability by the comparison with other DE algorithms,and the results prove the effectiveness of the TMCDE algorithm.

关 键 词：差分进化 差分策略 反向学习 混沌搜索 两阶段变异交叉 函数优化问题 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]