基于多尺度信息熵特征的数据流快速聚类研究  

Fast clustering of data streams based on multi-scale information entropy feature

作　　者：何宇新 廖长江 何新旭 HE Yuxin;LIAO Changjiang;HE Xinxu(CSG Digital Grid Group Co.Ltd.,Digital Eenterpise Branch,Guangzhou 510000,China)

机构地区：[1]南方电网数字电网集团有限公司数字企业分公司,广东广州510000

出　　处：《电子设计工程》2024年第15期41-44,共4页Electronic Design Engineering

基　　金：南网数研院人力资源管理人才数字化建设项目(0002200000085847)。

摘　　要：由于数据流规模较大,产生和更新的速度较快,因此数据流中有价值的信息很难被充分挖掘。针对这种情况,研究一种基于多尺度信息熵特征的数据流快速聚类方法。对数据流实施降维和标准化处理,提取数据流的多尺度特征,包括压缩比数、数据流平均值、数据流峰值以及数据流峭度。计算多尺度特征的信息熵权重,选取数据流快速聚类中心,将数据流样本划分给聚类中心,完成数据流快速聚类。实验结果表明,所研究聚类方法的聚类质量系数均大于0.9,说明该方法的聚类能力更强。Due to the large scale of the data stream and the fast speed of generation and update,it is difficult to fully mine the valuable information in the data stream.In this case,a fast clustering method of data streams based on multi-scale information entropy feature is studied.The dimensionality reduction and standardization of the data stream are implemented to extract the multi-scale characteristics of the data stream,including the compression ratio,the average value of the data stream,the peak value of the data stream and the kurtosis of the data stream.Calculate the information entropy weight of multi-scale features,select the data flow fast clustering center,divide the data flow samples to the clustering center,and complete the data flow fast clustering.The experimental results show that the clustering quality coefficients of the clustering methods studied are greater than 0.9,indicating that the clustering ability of the method is stronger.

关 键 词：多尺度信息熵特征 数据流 快速聚类 信息熵权重 聚类中心 

分 类 号：TN352.22[电子电信—物理电子学]