非同步多时间序列中频繁模式的发现算法

从多个时间序列中发现频繁模式在实际应用中具有非常重要的价值.已知文献所提供的方法均假设多时间序列是同步的,但是,在现实世界中,这一条件并不总能满足,许多情况下它们是非同步的.提出了一个从非同步多时间序列中发现频繁模式的算法.该算法首先利用线性化分段表示法和矢量形态聚类实现时间序列的特征分割与符号化转换,然后通过将Agrawal关联模式发现算法的核心思想与时间序列最短实现表示方法相结合,实现了非同步多时间序列中多种结构频繁模式的发掘.与已有算法相比,该算法更简单、更灵活,并且不要求序列严格同步.实验结果证明了该算法的有效性.

An Algorithm for Discovering Frequent Patterns in Non-Synchronous Multiple Time Series

Discovering frequent patterns in multiple time series is important in practices. Methods appeared in literatures assume that the multiple time series are synchronous, but in the real world, that is not always satisfied, in most cases they are non-synchronous. In this paper, an algorithm for discovering frequent patterns in non-synchronous multiple time series is proposed. In this algorithm, first, the time series is segmented and symbolized with the linear segment representation and the vector shape clustering method,so that each symbol can represent aprimitive and independent pattenr.Thenthe minimal occurrence representaion of time series and the association rule discovery algorithm proposed by Agrawal is combined to extrac frequent patterns of various structures from non-synchonous multiple time series.Compared with the previous methods,the algorithm is more simple and flexible,and does not require time series to be synchronous.Experimental results show the efficency of the algorithm.