一种改进ε支配的等度规映射方法  

Improved ε Dominance by Isomap

作　　者：杨咚咚[1] 马晶晶[1] 焦李成[1] 公茂果[1] 司晓云[1] 

机构地区：[1]西安电子科技大学智能信息处理研究所智能感知与图像理解教育部重点实验室,陕西西安710071

出　　处：《软件学报》2011年第10期2291-2304,共14页Journal of Software

基　　金：国家自然科学基金(60703107;61001202);国家高技术研究发展计划(863)(2009AA12Z210);中国博士后科学基金(20090451369);新世纪优秀人才支持计划(NCET-08-0811);中央高校基本科研业务费专项资金(JY10000902039);北京市自然科学基金(7092020);陕西省自然科学基础研究计划(2009JQ8015);陕西省"13115"科技创新工程重大科技专项(2008ZDKG-37)

摘　　要：新型支配机制的研究是进化多目标优化领域中的热点和难点之一,其中,ε支配最具代表性.但是,它的缺点在于对于不同几何形状的Pareto前沿十分敏感.提出了一种改进ε支配机制的等度规映射方法,采用等度规映射把解映射到低维流形空间,发现隐藏于非支配解的几何分布,在低维流形空间进行ε支配的剪枝操作.与传统的ε支配相比,该机制不会丢失部分有效解,能够较好地保持解分布的均匀性.另外,为了克服传统ε支配丢失部分极端解的不足,设计了极端解校验算子.与NSGAII,SPEA2,NNIA和εMOEA相比,所提出的ε支配和极端解校验算子能够较好地保持解分布的均匀性和宽广性,改进了传统的ε支配.The study of new types of dominance mechanisms is a key point in current evolutionary multi- objective optimization community and 6 dominance is a representative among them. However, their ability in diversity maintaining is sensitive to different shapes of Pareto fronts. This paper proposes an improved 6 dominance mechanism by Isomap, which employs Isomap to embed the original population to low dimensional manifold space. The intrinsic geometric structure of them is discovered and ε dominance is adopted to select data in the embedding space. Compared with traditional ε dominance, the mechanism does not lose valid solutions and can maintain a set of uniform-distributed solutions. In addition, the extreme solution-check operator is proposed to enhance the ability of holding extreme solutions of ε dominance. The detailed experimental comparison with NSGAII, SPEA2, NNIA and εMOEA shows that the two strategies in this study are beneficial to uniformity and spread maintenance, which are in the enhanced version of traditionalεdominance.

关 键 词：进化计算 多目标优化 破配 等度规映射 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]