摘要
针对使用多目标聚类集成算法得到的聚类解集中包含大量质量较差解，影响后续集成操作的问题，提出一种基于热点解搜索和差分进化的多目标聚类集成算法。根据热点解的概念找出聚类解集中质量较好的解，以这些解引导种群的搜索方向，加强潜在最优区域的搜索；在后续集成操作中只采用热点解及其邻域个体，去除较差解对最终结果的影响。在优化过程中采用改进的差分进化算子提高全局寻优的能力，去除编码长度不一对算子使用的影响。对3组UCI数据的测试结果表明，该算法优于2种对比算法，其RI取值提高了0．0021～0．0524，FM取值提高了0．0134～0．0591。
The effect of ensemble operator in the multiobjective clustering ensemble algorithm is weakened by the bad solutions of the obtained solution set. A novel multi-objective clustering ensemble algorithm based on searching knees solutions and differential evolution was proposed to solve the problem. The promising solutions based on the notion of “knees” were found out and used to guide the search direction for enhancing the promising regions. During the ensemble operation process, only knees solutions and their neighbors were used so that the bad solutions did not influence the final results. An advanced differential evolution operator （DEO） was designed to improve the global searching ability and to solve the problem caused when using the DEO for varying codes. The test results show that the proposed algorithm works better than the other two algorithms, the RI values are improved 0. 0021 to 0. 0524, and the FM values are improved 0. 0134 to 0. 0591.
出处
《计算机工程与设计》
CSCD
北大核心
2014年第8期2912-2916,共5页
Computer Engineering and Design
基金
国家青年基金项目(61301232)
河南省教育厅科学技术研究重点基金项目(13A520148)
关键词
多目标聚类
聚类集成
热点解
差分进化
全局寻优
multi-objective clustering
clustering ensemble
knees solution
differential evolution
global optimization
