多源异构高维数据特征加权子空间聚类算法  

Multi⁃source heterogeneous high⁃dimensional data feature weighted subspace clustering algorithm

作　　者：郑灶贤 ZHENG Zaoxian(China Southern Power Grid Digital Enterprise Technology(Guangdong)Co.,Ltd.,Guangzhou 510000,China)

机构地区：[1]南方电网数字企业科技(广东)有限公司,广东广州510000

出　　处：《电子设计工程》2024年第20期191-195,共5页Electronic Design Engineering

基　　金：南方电网财务有限公司2022-2023年金融业务系统2.0优化建设项目(JY-KF-02-CJ-22-012)。

摘　　要：多源异构高维数据存在噪声和异常值,导致聚类中心的确定难度较大。为解决该问题,提出多源异构高维数据特征加权子空间聚类算法。计算簇的所有维信息熵,按自适应函数获取子空间最小值,确定聚类特征子空间。对两簇子空间权重分配,计算两簇相似度。通过相邻稠密单元格合并,确定质心,识别高维维度上的簇。利用数据之间互表示特性分析数据潜在低秩结构,提取数据中无杂质的低秩部分,构建多核子空间聚类目标函数。确定不同类数据之间欧氏距离,保证所有类簇融合为一类簇,实现全部子空间数据完全聚类。由实验结果可知,该算法聚类中心点与理想聚类中心点一致,且聚类结果能够达到理想效果。Because of the noise and outliers in multi-source heterogeneous high-dimensional data,it is difficult to determine the clustering center.To solve this problem,a feature weighted subspace clustering algorithm for multi source heterogeneous high dimensional data is proposed.Calculate the information entropy of all dimensions of the cluster,obtain the minimum value of the subspace according to the adaptive function,and determine the cluster feature subspace.Calculate the similarity between the two clusters by assigning subspace weights to the two clusters.By merging adjacent dense cells,the centroid is determined and clusters on high dimensional dimensions are identified.The potential low rank structure of the data is analyzed using the mutual representation characteristics between the data,and the low rank parts of the data without impurities are extracted to construct a multi kernel subspace clustering objective function.Determine the Euclidean distance between different types of data,ensure that all clusters are fused into one cluster,and achieve complete clustering of all subspace data.From the experimental results,it can be seen that the clustering center point of the algorithm is consistent with the ideal clustering center point,and the clustering results can achieve ideal results.

关 键 词：多源异构 高维数据 特征加权 子空间聚类 

分 类 号：TN101[电子电信—物理电子学]