摘要
K-Means算法对于初始中心点敏感,容易受到噪声干扰,无法处理非确定性问题等缺陷,且其改进遗传K-Means容易陷入到局部最优解中,粗糙聚类算法虽提升了算法对于不确定性问题的分析能力,但其仍有较大的提升空间。为此,提出将遗传算法与粗糙集理论结合起来,引入小生境和禁忌算法的思想,在计算数据集合的各个中心点时,采用遗传算法计算各个类别的粗糙均值点,遗传算法中的选择运算采用小生境技术,将禁忌算法作为变异算子。通过对4组UCI数据集的实验分析与比较,表明了所提算法具有更好的求解质量。
To solve defects of K-Means that it is sensitive to initial centers and noise,and that it cannot deal with uncertain problems,many improving algorithms are provided.The clustering analysis based on genetic algorithm can improve its ability but it easily plugs into a local optimum.Rough clustering algorithm can deal with uncertain problems,and it has good robustness,but it has more improving potential for its correction.The rough clustering algorithm based on niche generic and tabu search was proposed,the mentioned two algorithms were combined.Generic algorithm was used to compute the rough center of data,and niche technology was used in the selection operation,and the tabu search was used as the mutation operation.Four experiments based on UCI datasets were carried out,the results verified the efficiency of the proposed algorithm.
出处
《计算机工程与设计》
北大核心
2017年第10期2718-2722,2739,共6页
Computer Engineering and Design
基金
国家自然科学基金项目(61462008)
柳州市科学研究与技术开发计划基金项目(2016C050205)
广西科技大学创新团队基金项目(0316000209)
广西科技大学校科自基金项目(174523)
广西高校图形图像智能处理重点实验室基金项目(GIIP201508)
广西教育厅中青年教师基础能力提高基金项目(KY2016YB252)
关键词
聚类
禁忌搜索
遗传算法
粗糙集
小生境
clustering
tabu search
generic algor ithm
rough set
niche
