基于异构值差度量的SOM混合属性数据聚类算法  

Self-organizing mapping clustering algorithm based on heterogeneous value difference metric for mixed attribute data

作　　者：张宇献[1] 彭辉灯 王建辉[2] 

机构地区：[1]沈阳工业大学电气工程学院,沈阳110870 [2]东北大学信息科学与工程学院,沈阳110819

出　　处：《仪器仪表学报》2016年第11期2555-2562,共8页Chinese Journal of Scientific Instrument

基　　金：国家自然基金(61102124);辽宁省自然科学基金(2015020064)项目资助

摘　　要：针对传统聚类算法处理混合属性数据聚类质量不高且聚类结果可视化差的问题,提出了基于异构值差度量的自组织映射混合属性数据聚类算法。该算法以自组织映射神经网络为框架,采用基于样本概率的异构值差度量混合属性数据的相异性。利用分类特征项在Voronoi集合中出现频率作为分类属性数据参考向量更新规则的基础,通过混合更新规则实现数值属性和分类属性数据规则的更新。利用UCI公共数据库中的分类属性和混合属性数据集来测试所提出的聚类算法,并与SOM算法和kprototypes、SBAC、KL-FCM-GM算法进行比较。最后将所提出的聚类算法应用于轮式移动机器人的运动状态分析,获得了较好的聚类效果。Aiming at the problems of poor clustering quality and visualization of most traditional clustering algorithms in processing mixed attribute data,a self-organizing mapping clustering algorithm for mixed attribute data is proposed based on heterogeneous value difference metric. The proposed clustering algorithm takes the self-organizing mapping neural network as the framework and adopts the heterogeneous value difference based on sample probability to measure the dissimilarity of the mixed attribute data. The frequency of each category characteristic item occurring in the Voronoi set is utilized as the basis of the update rule of the category attribute data reference vectors. The mixed update rule of neurons is adopted to realize the update of the data rules of both numeric and categorical attributes simultaneously. The categorical attribute datasets and mixed attribute datasets in UCI public database were used to test the performance of the proposed clustering algorithm. The results of the proposed clustering algorithm were compared with those of the SOM algorithm,kprototypes,SBAC and KL-FCM-GM algorithms. Finally,the proposed clustering algorithm was applied to the motion status analysis of wheeled mobile robot,and good clustering results were obtained.

关 键 词：聚类 自组织映射 异构值差度量 混合属性 混合更新规则 

分 类 号：TH701[机械工程—仪器科学与技术]