基于密度梯度的滑动窗口数据流任意形状聚类  

Arbitrary Shape Clustering of Sliding Window Data Streams Based on Density Gradient

作　　者：张华 杨磊[2] ZHANG Hua;YANG Lei(School of Artificial Intelligence and Computer Science,Jiangnan University,Wuxi Jiangsu 214000,China;University of Electronic Science and Technology of China,Chengdu Sichuan 611731,China)

机构地区：[1]江南大学人工智能与计算机学院,江苏无锡214000 [2]电子科技大学,四川成都611731

出　　处：《计算机仿真》2022年第4期316-320,共5页Computer Simulation

摘　　要：数据流为连续快速到达的数据序列,数据量大且时变性较强,对其聚类时不能准确获得聚类细节,无法高效处理演化数据流。因此,提出基于密度梯度的滑动窗口数据流任意形状聚类方法。结合数据流特征,确定数据处理时需要满足实时性、准确性等要求;根据不确定指数与不确定度分析聚类时必须处理的问题,定义核心密度单元与候选密度单元;通过点邻域、点密度、不动点以及边界点构建密度梯度模型,计算密度分布情况,分类不动点并赋予类别号,在指定聚类数与未指定聚类数两种情况下,按照合并程度函数大小排序,将拥有最小函数值的两类合并,完成数据流任意形状聚类。仿真结果表明,所提方法的聚类结果准确,对滑动窗口敏感性较低,加速比性能良好。The data stream is a data sequence that arrives quickly and continuously. The amount of data is large and time-varying, the clustering details cannot be accurately obtained when clustering, and the evolution data stream cannot be processed efficiently. Therefore, a sliding window data stream arbitrary shape clustering method based on density gradient is proposed. According to the characteristics of the data stream, it is necessary to meet the requirements of real-time and accuracy when data processing is determined;The influence of uncertainty on clustering results was analyzed according to the uncertainty index and uncertainty;The core density unit and candidate density unit were defined, and the density gradient model was constructed through point neighborhood, point density, fixed point and boundary point, the density distribution was calculated and the fixed points were classified. In the case of the number of specified clusters and the number of clusters not specified, the two classes with the minimum function value were combined according to the size of the merging degree function, so as to complete the arbitrary shape clustering of the data stream. Simulation results show that the clustering results of this method are accurate, the sensitivity to sliding window is low, and the speedup ratio is good.

关 键 词：密度梯度 滑动窗口 数据流 合并程度函数 聚类算法 

分 类 号：TP313[自动化与计算机技术—计算机软件与理论]