引用本文:杨紫晴,姚加林,伍国华,陈学伟,毛成辉.集成协方差矩阵自适应进化策略与差分进化的优化算法[J].控制理论与应用,2021,38(10):1493~1502.[点击复制]
YANG Zi-qing,YAO Jia-lin,WU Guo-hua,CHEN Xue-wei,MAO Cheng-hui.Ensemble optimization algorithm from covariance matrix adaptive evolution strategy and differential evolution[J].Control Theory and Technology,2021,38(10):1493~1502.[点击复制]
集成协方差矩阵自适应进化策略与差分进化的优化算法
Ensemble optimization algorithm from covariance matrix adaptive evolution strategy and differential evolution
摘要点击 3550  全文点击 984  投稿时间：2021-01-02  修订日期：2021-08-03
查看全文  查看/发表评论  下载PDF阅读器
DOI编号  10.7641/CTA.2021.10002
  2021,38(10):1493-1502
中文关键词  智能优化算法  差分进化  协方差矩阵自适应进化策略  算法集成  连续优化
英文关键词  intelligent optimization algorithm  differential evolution  covariance matrix adaptation evolution strategy  ensemble algorithm  continuous optimization
基金项目  国家自然科学基金项目(62073341), 中南大学研究生自主探索创新项目(1053320190633)资助.
作者单位E-mail
杨紫晴 中南大学 maimai@csu.edu.cn 
姚加林 中南大学  
伍国华 中南大学  
陈学伟 中南大学  
毛成辉* 中南大学 mchh99@csu.edu.cn 
中文摘要
      不同智能优化算法在求解优化问题时通常表现出显著的性能差异. 差分进化(DE)算法具备较好的全局搜 索能力, 但存在收敛慢、效率低的不足, 协方差矩阵自适应进化策略(CMA–ES)局部搜索能力强, 具备旋转不变性, 但容易陷入局部最优, 因此, DE和CMA–ES之间具有潜在的协同互补能力. 针对上述问题, 提出了一种集成协方差 矩阵自适应进化策略与差分进化的优化算法(CMADE). 在CMADE框架中, DE算法负责全局搜索, CMA–ES算法进 行局部搜索. 通过周期性解交换机制实现CMA–ES和DE两个算法间协同交互和反馈控制. 在解交换时, 从DE种群 中选择优秀个体, 利用CMA–ES算法在优秀个体周围进行局部搜索. 同时在DE和CMA–ES的混合种群中, 综合考虑 解的多样性和最优性, 选取一定比例的解作为DE算法的新种群进行全局搜索, 实现全局搜索与局部搜索的动态平 衡. 将CMADE算法与CMA–ES, DE, SaDE, jDE, EPSDE, ACODE和SHADE算法在CEC2014标准测试集上进行比较 实验. 结果表明, CMADE整体性能显著优于其它比较算法.
英文摘要
      Different metaheuristic algorithms often show significant performance differences in solving optimization problems. Differential evolution (DE) algorithm has a good global search ability, while it has disadvantages of slow convergence speed and low efficiency. Covariance matrix adaptive evolution strategy (CMA–ES) has strong local search capabilities and rotation invariance, but it is easy to fall into local optimum. Therefore, there is a potential synergy between DE and CMA–ES. To address these problems, this paper proposes an integrated covariance matrix adaptive evolution strategy and differential evolution optimization algorithm (CMADE). Under the framework of the CMADE, DE algorithm is designed for global search, and CMA–ES algorithm is designed for local search. The two algorithms are integrated by a periodic information exchange strategy to realize collaborative interaction and feedback control. During the information exchange, the best individual is selected from the population of DE algorithm, then CMA–ES algorithm is used for local search around the best individual. Considering the diversity and optimality of solutions, we choose a new population from the mixed populations of DE and CMA–ES for global search , to realize the dynamic balance between global search and local search. We conduct extensive experiments on the suit of CEC 2014 benchmark functions and comprehensive comparisons with the seven algorithms including CMA–ES, DE, SaDE, jDE, EPSDE, ACODE and SHADE, which demostrates the superiority of the proposed CMADE.