一种改进的时间序列模式发现算法

如何有效地在时间序列数据库中发现时序模式是时间序列数据挖掘中一个具有重要意义的课题．本文提出一种改进的在时间序列中有效地发现时序模式的算法．在将时间序列划分为若干等长的子序列之后，根据基于关键点的线性分段算法提取每个子序列的关键点序列，该关键点序列仅保留反映数据序列的变化模式的主要关键点．接着利用每个关键点序列分隔相应的子序列，根据数据的起伏变化将相应的关键点序列分配到一系列盒子中，使得只有在同一个盒子中的序列才有可能相似，而不同盒子中的序列不可能相似．最后通过计算每个盒子中任意两个关键点序列之间的动态时间弯曲距离来发现所有的时序模式．实验结果验证了该算法的有效性．

An Improved Algorithm for Time-Series Pattern Discovery

A significant topic of time series data mining is to discover time-series patterns in time series database effectively. An improved effective algorithm for time-series pattern discovery is proposed in this paper, it divides a given sequence into several subsequences of the same length, and then a key-point series is extracted from each snbsequence by using a segmentation algorithm based on key points, only retaining the main key points which reflect its changing patterns. Separate each subsequence by its key-point series, and then distribute their key-point series into a set of boxes according to ups and downs, so that only those in the same box are possibly similar, while those in the different boxes are not. Finally all the time-series patterns will be discovered by computing Dynamic Time Warping distance between any two key-point series in each box. Experimental results show the effectiveness of the proposed algorithm.