时间序列相似模式的有效匹配

An Effective Algorithm to Match Similar Time Series Pattern

将经验模式分解和多层前向网络的交叉覆盖算法相结合,提出一种时间序列相似模式的匹配算法.先利用经验模式分解实现时间序列趋势的提取,再把所有的趋势序列分成训练集和测试集2个部分.通过训练为每个类别做出描述,根据测试集中的每个趋势序列和覆盖中心之间的距离把它们分配到与之最匹配的类别中.实验结果表明:该算法是一种较理想的序列模式匹配方法,更擅长于维数较高的序列的匹配.

This paper proposes an effective time series matching method by combining the empirical mode decomposition （EMD） with the alternative covering algorithm. It decomposes at first a time series into a trend part and some detail parts using the EMD, and then divides all trend series into two sets： training sets and testing sets. Each pattern is learnt during the training process, and the trend series in the testing set are assigned to one of the labeled patterns based on its distance to the center of each covering. Experimental results show that the proposed method performs well and appears to be more suitable for high-dimensionality matching.