针对簇类中心自适应的密度峰值聚类算法  

Adaptive density peak clustering algorithm for cluster centers

作　　者：王慧玲 宋威[2] 谢国伟 WANG Huiling;SONG Wei;XIE Guowei(Laboratory of IOT Application Technology,Wuxi Taihu University,Wuxi 214064,China;School of Internet of Things Engineering,Jiangnan University,Wuxi 214122,China;Wuxi Aerospace Jiangnan Data System Technology Co Ltd,Wuxi 214125,China)

机构地区：[1]江苏省物联网应用技术重点建设实验室无锡太湖学院,江苏无锡214064 [2]江南大学物联网工程学院,江苏无锡214122 [3]无锡航天江南数据系统科技有限公司,江苏无锡214125

出　　处：《传感器与微系统》2020年第12期119-122,126,共5页Transducer and Microsystem Technologies

基　　金：国家自然科学基金资助项目(61673193);中央高校基本科研业务费专项资金资助项目(JUSRP51635B);中国博士后科学基金资助项目(2017M621625);江苏省自然科学基金资助项目(BK20181341)。

摘　　要：针对密度峰值聚类(CFSFDP)算法无法自动确定簇类中心点和需要人为主观通过决策图截取簇类中心的局限性的问题,提出了一种针对簇类中心自适应的密度峰值聚类算法。首先应用马氏距离计算方式计算数据的距离矩阵,其次根据簇类中心的分布规律提出自适应化的簇类中心提取机制,最后基于簇类中心的刻画自适应化离群点的分辨,从而得到最终的聚类效果。为验证所提方法的有效性,实验中对多个数据集进行广泛测试,结果表明:所提算法在人工测试数据集和UCI数据集上都比原CFSFDP算法具有更好的聚类结果。In order to overcome the limitation of clustering by fast search and find of density peaks(CFSFDP)which cannot automatically determine cluster centers and need human-centered interception of cluster centers through decision maps,an adaptive density peak clustering algorithm for cluster centers is proposed.Firstly,the distance matrix of data is calculated by using the method of Mahalanobis distance calculation.Secondly,according to the distribution rule of cluster centers,an adaptive cluster center extraction mechanism is proposed.Finally,the outlier resolution is adapted based on the description of cluster centers,so as to obtain the final clustering effect.In order to verify the effectiveness of the proposed method,experiments are carried out on a wide range of datasets.The results show that the proposed algorithm has better clustering results than the original CFSFDP algorithm on both manual test datasets and UCI datasets.

关 键 词：聚类 密度聚类 密度峰值 簇类中心自适应 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]