时空事件聚类分析方法研究  

On Spatio-temporal Events Clustering Methods

作　　者：唐建波[1] 邓敏[1] 刘启亮[1,2] 

机构地区：[1]中南大学地球科学与信息物理学院,湖南长沙410083 [2]香港理工大学土地测量与地理资讯学系

出　　处：《地理信息世界》2013年第1期38-45,共8页Geomatics World

基　　金：教育部新世纪优秀人才资助计划项目(NECT-10-0831)资助;高等学校博士学科点专项科研基金项目(20110162110056)资助

摘　　要：时空事件聚类分析是当前聚类分析研究领域最前沿的一项主要内容,在疾病预警与控制、气候变化、地震预测及犯罪行为分析等领域具有重要的应用价值。本文一方面针对时空事件聚类分析的研究现状与最新进展进行归纳总结;另一方面选取四种代表性的时空事件聚类分析方法(时空重排扫描统计,ST-DBSCAN,WKN,STSNN)从聚类质量与用户操作两个方面进行试验分析,对同一方法针对不同类型数据集,以及不同方法针对同一数据集的聚类效果进行了对比分析。最后,在实验分析的基础上对现有时空事件聚类分析的优缺点及适用性进行了归纳总结,并指出了需要进一步研究的若干问题。Clustering analysis of spatio-temporal event is a forefront research in the spatio-temporal data mining domain. It has important applications in disease early warning and controlling, climate change, earthquake prediction and the analysis of crimes. Firstly, the research actuality and new progresses in spatio-temporal event clustering algorithm in recent years are investigated and summarized. Secondly, the analysis and comparison of four representative spatio-temporal event clustering methods （i.e. Space-time permutation scan statistic, ST- DBSCAN, WNN, STSNN） have been made from the views of clustering quality and user operation. Moreover, clustering conditions of the same method for different types of data sets as well as different methods for the same datasets are analyzed. Finally, the advantages, disadvantages and applicability of these four clustering methods are summarized after several experiments and comparative analyses, and a number of issues for further research are highlighted.

关 键 词：时空事件 聚类分析 对比分析 

分 类 号：O141.4[理学—数学]