新的时间序列相似性度量方法  

New similarity measure method on time series

作　　者：张海涛[1,2] 李志华[1,2] 孙雅[1] 张华伟[2] 

机构地区：[1]江南大学物联网工程学院轻工过程先进控制教育部重点实验室,江苏无锡214122 [2]江南大学物联网应用技术教育部工程研究中心,江苏无锡214122

出　　处：《计算机工程与设计》2014年第4期1279-1284,共6页Computer Engineering and Design

基　　金：江苏省产学研前瞻基金项目(BY2013015-23);中央高校基本科研业务费专项基金项目(JUSRP211A41)

摘　　要：时间序列的相似性度量是时间序列数据挖掘的研究基础,通过对时间序列的变化趋势符号化,定以趋势距离,提出新的时间序列相似性度量方法—SMVT。该方法可以度量时间序列的变化趋势相似性,具有简洁直观,趋势描述科学,能很好地反映时间序列的整体变化趋势的特点。通过新的相似性度量方法对时间序列进行聚类计算,实验结果表明,该方法能够有效地从变化趋势的角度度量时间序列间的相似性,具有较好的实用性和良好的应用前景。Similarity measure on time series is the basis of time series data mining, a new similarity measure method on time se- ries--SMVT by describing time series with symbol based on variation trends and defining trend distance. The variation trends similarity of time series are measured with this method, which is concise and straightforward, trends description scientific, and reflects the overall variation trends of time series well. Applying the new similarity method for clustering computation of time se- ries, the experimental results show that the method can measure the similarity of time series effectively from the perspective of variation trends, which explains its well practicability and good application prospect.

关 键 词：时间序列 数据挖掘 变化趋势 相似性度量 趋势距离 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]