计算机应用 ›› 2010, Vol. 30 ›› Issue (07): 1944-1946.

• 数据库技术 • 上一篇    下一篇

基于事件的时间序列相似性度量方法

吴学雁1,黄道平2   

  1. 1. 华南理工大学;广东工业大学
    2. 华南理工大学自动化科学与工程学院
  • 收稿日期:2009-12-14 修回日期:2010-03-05 发布日期:2010-07-01 出版日期:2010-07-01
  • 通讯作者: 吴学雁

Time series similarity matching based on event

  • Received:2009-12-14 Revised:2010-03-05 Online:2010-07-01 Published:2010-07-01

摘要: 为了在时间序列相似性度量过程中更好地体现用户的需求,提高相似性度量的准确度,提出了基于事件的时间序列相似性度量方法(SMBE)。首先将用户的需求定义为事件,将原始时间序列转化为事件序列;然后,构建了基于事件序列的相似性度量模型(SMBE),SMBE定义了不同事件序列中各元素之间的相似性,并构成相应的相似性矩阵,对相似性矩阵进行搜索得到最优路径的值作为序列之间的相似性度量;最后,提出了基于SMBE的聚类方法。实验表明,在参数设置合理的情况下,能获得接近0.90的聚类精度。

关键词: 时间序列, 相似性度量, 聚类, 数据挖掘

Abstract: In order to do the time-series similarity matching with the users’ needs and improve the accuracy, the time-series similarity matching based on event (SMBE) is proposed. First, the users’ needs are defined as the event and the original time series is translated into the event sequence. Then, SMBE is constructed. SMBE defines the similarity of the various elements in the two different sequences, constitute the corresponding similarity matrix and searches the optimal path value as the similarity measurement. Finally, the clustering method based on SMBE is proposed. Experimental results show that the clustering based on SMBE can get the accuracy of 90% with the reasonable parameters.

Key words: time series, similarity matching, clustering, data-mining