基于时态逻辑的多时间序列挖掘模型

A new model for mining multiple time series based on temporal logic

为了从多时间序列之间发现的定性的时态相关模式可而更全面的理解和把握系统的演化特性,提出了一种基于时态逻辑的多时间序列挖掘模型。它首先将多时间序列转化为多事件序列,然后将预处理后的多事件序列利用区间时态逻辑(ITL)关系子集来定义多事件序列中事件间的时态相关模式。其次进行多状态序列融合和局部时态观测序列的生成,之后采用频繁模式挖掘算法发现多时间序列的频繁时序模式。该模型有助于解决时间序列挖掘所面临的若干挑战和难题,有助于扩展现有时间序列挖掘系统的功能,从而指导时间序列等复杂类型数据的知识发现过程。实验结果表明了该模型及算法的有效性和优越性。

In order to extract connotative pattern for scientific decision-making, identify the qualitative temporal pattems from the multiple time series and have a comprehensive understanding on the evolution of the system, a new mining model for multiple time series is proposed based on temporal logic. Firstly, the multiple time series are transformed into multiple event sequences, and then temporal relative pattern in multiple event sequences is defined using a subset of ITL relation. Secondly, the multiple state sequences are synthesized into one state sequence, and then the frequency pattern mining algorithm is applied to discover the frequency time series in multiple time series. The proposed model is useful in time series mining, and expends the capacity of existing time series mining system. The case study shows that the proposed model is valid and has advantages.