局部抽象凸区域剖分差分进化算法  

Differential Evolution Algorithm with Local Abstract Convex Region Partition

作　　者：周晓根 张贵军[1] 郝小虎[1] 

机构地区：[1]浙江工业大学信息工程学院,杭州310023

出　　处：《自动化学报》2015年第7期1315-1327,共13页Acta Automatica Sinica

基　　金：国家自然科学基金(61075062);浙江省自然科学基金(LY13F030008);浙江省科技厅公益项目(2014C33088);浙江省重中之重学科开放基金(20120811);杭州市产学研合作项目(20131631E31)资助~~

摘　　要：在差分进化算法框架下,结合抽象凸理论,提出一种局部抽象凸区域剖分差分进化算法(Local partition based differential evolution,LPDE).首先,通过对新个体的邻近个体构建分段线性下界支撑面,实现搜索区域的动态剖分;然后,利用区域剖分特性逐步缩小搜索空间,同时根据下界估计信息指导种群更新,并筛选出较差个体;其次,借助下界支撑面的广义下降方向作局部增强,并根据进化信息对搜索区域进行二次剖分;最后,根据个体的局部邻域下降方向对部分较差个体作增强处理.数值实验结果表明了所提算法的有效性.Within the framework of differential evolution algorithm, a differential evolution algorithm with local abstract convex region partition is proposed in this paper incorporating the abstract convexity theory. Firstly, the partition of the search domain is performed dynamically by building piecewise linear abstract convex lower supporting hyperplanes for the neighboring individuals of new individuals. Secondly, the search domain narrows gradually by using properties of the region partition. Meanwhile, the process of population updating is guided according to the information of underestimate, and poor individuals are identified effectively. Additionally, the generalized descent directions of the lower supporting hyperpalens are used for local enhancement, and the search domain is partitioned again according to the evolutionary information. Finally, some poor individuals get enhanced according to descent directions of their local neighbourhood. Numerical experiment results have verified the effectiveness of the proposed algorithm.

关 键 词：差分进化 区域剖分 全局优化 下界估计 抽象凸 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]