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

李斌,谭立湘,解光军,李海鹰,庄镇泉
2002-01-01
Journal of Software
Abstract: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 a primitive and independent pattern. Then, the minimal occurrence representation of time series and the association rule discovery algorithm proposed by Agrawal is combined to extract frequent patterns of various structures from non-synchronous 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 efficiency of thealgorithm.
What problem does this paper attempt to address?