应用精英反向学习策略的混合差分演化算法  

A Hybrid Differential Evolution with Elite Opposition-Based Learning

作　　者：汪慎文[1,2,3] 丁立新[1,3] 谢承旺[4] 郭肇禄[1,3] 胡玉荣[1,3] 

机构地区：[1]武汉大学软件工程国家重点实验室,湖北武汉430072 [2]石家庄经济学院信息工程学院,河北石家庄050031 [3]武汉大学计算机学院,湖北武汉430072 [4]华东交通大学软件学院,江西南昌330013

出　　处：《武汉大学学报（理学版）》2013年第2期111-116,共6页Journal of Wuhan University:Natural Science Edition

基　　金：国家自然科学基金(60975050;61165004);高等学校博士学科点专项科研基金(20070486081);中央高校基本科研业务费专项资金(6081014);江西省自然科学基金(20114BAB201025);福建省自然科学基金(2012J01248)资助项目;河北省科技支撑计划项目(11213587)

摘　　要：针对传统差分演化算法在演化后期收敛速度变慢的问题,利用精英个体的良好信息,在一般反向学习方法的基础上,提出精英反向学习策略,并融合降低参数敏感性和变异策略敏感性的机制,设计了一种基于精英反向学习策略的混合差分演化算法(EOCoDE),从理论上证明了该算法的全局收敛性.新算法使用精英反向策略初始化种群,在进化过程中,如果满足预设定的学习概率,就执行精英反向算子,否则,随机组合参数知识库和策略知识库中的知识来产生差分演化种群.对比实验结果表明,精英反向学习策略比一般反向学习策略具有更强的搜索能力,EOCoDE算法的性能具有明显优势.To solve the problem of slow convergence speed before reaching the global optimum in the conventional differential evolution （DE）, an effective approach, called elite opposition-based learning, is proposed, in which the generalized opposition-based learning strategy is improved by the elite members. A novel hybrid differential evolution algorithm （EOCoDE） is presented in this paper. The proposed algorithm incorporates the elite opposition-based learn- ing method into the mechanism of less influenced parameter and mutation strategy. It is also proven that the proposed algorithm can guarantee the convergence towards the global optimum. The novel algorithm starts with an initial popu- lation by elite opposition-based learning strategy, and then selects the knowledge from the mutation strategy base and control parameter setting base to generate the DE population. During the evolution process, the opposition population is calculated to compete with the current population according to the preset probability of learning. Experimental re- sults show that the elite opposition-based learning strategy has much better search performance than the generalized opposition-based learning strategy and the novel EOCoDE algorithms can obtain better efficiency.

关 键 词：差分演化 一般反向学习 精英反向学习 参数敏感性 变异策略 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]