基于正交设计的反向学习差分进化算法  

Differential evolution algorithm using orthogonal design opposition-based learning

作　　者：閤大海[1,2] 李元香[1,2] 祝婕[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]武汉大学计算机学院,湖北武汉430072 [3]长江工程职业技术学院测绘工程系,湖北武汉430212

出　　处：《华中科技大学学报（自然科学版）》2017年第5期23-27,44,共6页Journal of Huazhong University of Science and Technology(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61672391);国家重大仪器专项资助项目(2011YQ170065.4)

摘　　要：为了克服反向学习带来的维度退化现象,生成部分反向解增强对反向解空间的勘测能力,提出了一种基于正交设计的反向学习差分进化算法,利用正交设计仅生成若干具有代表性的部分反向解,在增强算法勘测能力与减少函数评价次数上达到了一个良好的平衡.实验结果表明:该算法相对于其他几种反向学习差分进化算法有更好的收敛精度及速度,同时对函数维度变化不敏感,鲁棒性较强.When opposition-based learning technique is used to generate an opposite individual, it is usually taking the inverse value of all dimensions and makes some dimensions have the potential to degenerate, however, all the partial opposite solutions will greatly increase the function evaluation times, thus a differential evolution algorithm was proposed based on orthogonal design. The orthogo- nal design was used to generate a number of representative partial opposite solutions, which can achieve a good balance between the ability to enhance the algorithm and decrease the number of func- tion evaluation. Experimental results show that the proposed algorithm has better convergence accura- cy and speed than other opposition-based learning differential evolution algorithms, and it is not sensi- tive to the change of function dimension with a good robustness.

关 键 词：差分进化算法 反向学习 正交设计 部分反向解 维度退化 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]