面向相似性搜索的时间序列表示方法述评  

A Review on Time Series Representation for Similarity-based Pattern Search

作　　者：刘世元[1] 江浩[1] 

机构地区：[1]华中科技大学机械科学与工程学院,武汉430074

出　　处：《计算机工程与应用》2004年第27期53-59,共7页Computer Engineering and Applications

基　　金：国家自然科学基金项目(编号:50205009)资助

摘　　要：时间序列作为一种数据形式,广泛存在于各种商业、医学、工程、自然科学和社会科学等数据库中。近年来,时间序列的相似性搜索问题正得到越来越多的重视。该问题可描述为给定某个的时间序列,要求从一个大型时间序列数据库中找出与之最相似的序列。该问题的有效求解涉及到两个关键难点,即相似性度量的定义和搜索算法的时间复杂度,而这两者都依赖于时间序列的近似表示方法。因此,通过详细评述面向相似性搜索的各种时间序列近似表示方法,对这些方法进行分析和比较,总结了这些方法的优点和不足,并对进一步的研究方向作出了预测。As a data form,time series exists broadly in many applications in the databases of business,medicine,engineering,natural sciences and social sciences.Recently,the similarity-based pattern search of time series has received increasing attention.The problem can be described as:search ing the sequence most similar to a given time series from a large time series database.The efficient solution to such a problem involves two difficulties,i.e.,the definition of similarity measurement and the time complexity of searching algorithm,both of which rely on the approximate representation of time series.Thus,this paper reviews several methods of time series representation for similarity-based pattern search,compares and shows their advantages as well as limitations,and speculates the possible directions in the field.

关 键 词：时间序列 相似性数据挖掘 维规约 

分 类 号：TP182[自动化与计算机技术—控制理论与控制工程]