面向多峰优化问题的双层协同差分进化算法  

Two-Layer Collaborative Differential Evolution Algorithm for Multimodal Optimization Problems

作　　者：陈宗淦[1] 詹志辉[1] CHEN Zong-Gan;ZHAN Zhi-Hui(School of Computer Science and Engineering,South China University of Technology,Guangzhou 510006)

机构地区：[1]华南理工大学计算机科学与工程学院,广州510006

出　　处：《计算机学报》2021年第9期1806-1823,共18页Chinese Journal of Computers

基　　金：国家重点研发计划(2019YFB2102102);国家自然科学基金优秀青年科学基金(61822602);面上项目(61772207)资助.

摘　　要：多峰优化问题是一类存在多个全局最优解的复杂优化问题,不仅要求算法找到尽可能多的最优解,而且要求算法尽可能提高所找到的最优解的精度.演化计算方法是求解这类问题的重要手段.但是传统演化计算方法面临多样性和收敛性两个方面的挑战.针对这两个方面的挑战,提出了一种通过探索层和精炼层协同演化的双层协同差分进化算法.在探索层中,每个个体作为一个分布式搜索单元探索并定位到一个最优解.在协同过程中,探索层引入个体寿命机制,将耗尽寿命且定位到最优解的个体存入一个外部存档,然后重新初始化这些个体以找到更多的最优解.在精炼层中,首先对探索层输送过来的外部存档中的个体进行聚类,然后对每一个类使用经典的全局优化差分进化算法进一步提升所找到的最优解的精度.因此,探索层和精炼层分别针对多样性和收敛性挑战,通过协同演化使得算法不仅能够找到尽可能多的最优解,而且使得找到的最优解的精度尽可能高.使用目前最常用的CEC’2013标准测试集中的所有20个多峰优化问题对所提出算法的性能进行测试,并与13种表现突出的和最新的多峰优化算法进行比较.实验结果显示,所提出的双层协同差分进化算法的整体性能优于所比较的13种多峰优化算法.Multimodal optimization problem is a kind of complex optimization problems with multiple global optimal solutions,which requires the algorithms not only to find as many optimal solutions as possible,but also to refine the accuracy of the found optimal solutions as much as possible.Evolutionary computation methods are promising approaches in solving this kind of problems.However,traditional evolutionary computation methods are faced with the challenges in diversity and convergence.To deal with these two challenges,a novel two-layer collaborative differential evolution algorithm is proposed,which is based on the collaboration of an exploration layer and a refinement layer.In the exploration layer,each individual acts as a distributed search unit to explore and locate an optimal solution.During the collaboration process,a lifetime mechanism is incorporated into the exploration layer,which stores the life-exhausted individuals who locate optimal solutions in an external archive and then re-initializes these individuals for finding more optimal solutions.In the refinement layer,the individuals in the external archive sent from the exploration layer are clustered.Then the classic differential evolution algorithm for global optimization is conducted in each cluster to further refine the accuracy of the found optimal solutions.Therefore,the exploration layer and refinement layer deal with the diversity and convergence challenges,respectively.Through the collaboration of the exploration layer and refinement layer,the proposed algorithm can not only find as many optimal solutions as possible,but also refine the found optimal solutions as accuracy as possible.All the 20 multimodal optimization problems in the most widely used benchmark set at present,i.e.,the CEC’2013 benchmark set,are adopted to test the performance of the proposed algorithm.The performance of the proposed algorithm is compared with 13 outstanding and latest multimodal optimization algorithms.The experimental results show that the proposed two-layer coll

关 键 词：差分进化算法 协同演化 探索层 精炼层 多峰优化问题 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]