一种基于势能模型的数据流聚类算法  

A DATA STREAM CLUSTERING ALGORITHM BASED ON POTENTIAL FIELD MODEL

作　　者：舒越 解庆 刘永坚[1] 唐伶俐[1] Shu Yue;Xie Qing;Liu Yongjian;Tang Lingli(School of Computer Science and Technology,Wuhan University of Technology,Wuhan 430070,Hubei,China)

机构地区：[1]武汉理工大学计算机科学与技术学院,湖北武汉430070

出　　处：《计算机应用与软件》2022年第11期222-230,237,共10页Computer Applications and Software

基　　金：国家自然科学基金项目(61602353)。

摘　　要：传统的数据流聚类算法大部分将距离作为相似度度量标准,这造成对噪点敏感的问题,聚类效果不理想。针对这种情况,提出一种基于势能模型的数据流聚类算法PHAStream,该算法结合在线/离线两阶段数据流聚类框架和基于势能模型的层次聚类算法PHA,可以有效处理噪点问题。初始化阶段使用PHA聚类算法得到初始微簇;在线阶段,对每个新到达的数据点,采用融合势能和距离的相似度度量标准来更新微簇,每隔一段时间采取剪枝策略删除过期的微簇,并调整所有微簇的类型;离线阶段,对所有正常微簇使用改进的PHA聚类算法得到最终聚类结果。在两组真实数据集上的对比实验表明,PHAStream算法可以有效提高聚类质量、聚类纯度和时间效率。Most traditional data stream clustering algorithms use distance as a similarity metric,which causes the sensitivity to noise and undesirable clustering effort.In view of this,a data stream clustering algorithm based on potential field model PHAStream is proposed.It combined an online/offline two-stage data stream clustering framework and the potential-based hierarchical clustering algorithm PHA.This algorithm could effectively deal with noise.The PHA clustering algorithm was used to obtain the initial micro-clusters in the initial phase.In the online phase,for each newly arrived data point,a similarity measure based on distance and potential energy was proposed to update the micro-clusters.In the offline phase,the improved PHA algorithm for all the normal micro-clusters was used to get the final result.Comparison experiments on two real data sets show that the PHAStream algorithm can effectively improve clustering quality,clustering purity and time efficiency.

关 键 词：数据挖掘 数据流聚类 势能场模型 数据概要 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]