反距离加权和微簇合并的密度峰值聚类算法  

Density peaks clustering with inverse distance weighting and micro cluster merging

作　　者：范强 吕莉[1,2] 邱日轩 崔希 张宸源 樊棠怀[1,2] FAN Qiang;LYU Li;QIU Rixuan;CUI Xi;ZHANG Chenyuan;FAN Tanghuai(School of Information Engineering Nanchang Institute of Technology,Nanchang 330099,China;Nanchang Key Laboratory of IoT Perception and Collaborative Computing for Smart City,Nanchang Institute of Technology,Nanchang 330099,China;Information and Telecommunication Branch,State Grid Jiangxi Electric Power Company,Nanchang 330012,China;JXIC Energy Tech.Research Institute Co.,Ltd.,Nanchang 330096,China;Jiangxi Shenzhou Information Security Assessment Center Co.,Ltd.,Nanchang 33002,China)

机构地区：[1]南昌工程学院信息工程学院 [2]南昌工程学院南昌市智慧城市物联感知与协同计算重点实验室,江西南昌330099 [3]国网江西省电力有限公司信息通信分公司 [4]江西江投能源技术研究有限公司,江西南昌330096 [5]江西神舟信息安全评估中心有限公司,江西南昌330002

出　　处：《南昌工程学院学报》2024年第6期43-50,90,共9页Journal of Nanchang Institute of Technology

基　　金：江西省重点研发计划一般项目(20203BBGL73225)。

摘　　要：密度峰值聚类算法不需要进行复杂的迭代计算,具有高的计算效率。但在聚类流形数据时,不能准确找到聚类中心且分配剩余样本时易引发样本的连续误分配。针对上述缺陷,提出了一种基于反距离加权和微簇合并的密度峰值聚类(DPC-IDW-MCM)算法。该算法结合样本的K近邻信息计算近邻密度,再引入反距离加权系数重新定义样本的局部密度,能更好地适应流形数据的结构特征,使算法更准确地找到类簇中心;同时定义微簇间相似性度量准则,将相似性高的微簇合并为类簇,有效避免了样本的误分配。将DPC-IDW-MCM算法与5种聚类算法在流形数据集和UCI数据集上进行了对比实验。实验结果表明,DPC-IDW-MCM算法在综合性能方面优于上述5种对比算法。Density peaks clustering does not need to carry out complex iterative calculation and it has high computational efficiency.However,when clustering the manifold data,it is easy to cause continuous misallocation of samples when the clustering center cannot be found accurately,and the remaining samples are allocated.In this paper,an inverse distance weighting and microcluster merging density peak clustering(DPC-IDW-MCM)algorithm is proposed.The algorithm combines the K-nearest neighbor information of the sample to calculate the nearest neighbor density,and then introduces the inverse distance weighting coefficient to redefine the local density of the sample,which can better adapt to the structural characteristics of the manifold class cluster,so that the algorithm can find the center of the class cluster more accurately.At the same time,we define the similarity measurement criteria among microclusters,and combine the microclusters with high similarity into class clusters,which effectively avoids the misallocation of samples.In this paper,DPC-IDW-MCM algorithm is compared with five clustering algorithms on manifold data sets and UCI data sets.The experimental results show that DPC-IDW-MCM algorithm has better comprehensive performance than the comparison algorithm.

关 键 词：聚类 流形数据 反距离加权 局部密度 微簇合并 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]