Abstract: Most of the existing similarity measurement, based on Euclidean distance, cannot be applied directly and effectively to similarity matching of the timeseries with different granularities. This paper proposed a new similarity measure based on the sample of the corresponding D-value. It firstly expounded the definition of the time-series with different granularities, and defined the sample of the corresponding D-value; secondly it put forward the similarity matching algorithm; finally, the experimental results prove that the algorithm can effectively measure the similarity of time-series with multiple granularities.

Key words: time-series, similarity matching, time granularity

摘要： 现有的时间序列的相似性度量大多基于欧氏距离，并不适用于不同粒度时间序列的相似性匹配，无法直接对其相似性进行有效的度量，为此，提出一种基于对应差值比样本的相似性度量，用于不同粒度时间序列的相似性匹配。首先对不同时间粒度的时序数据进行阐述，并定义了对应差值比样本与相似度计算方法；接着提出基于它们的相似性匹配算法；最后实验证明，该度量能够有效地度量不同粒度时间序列数据的相似性。

关键词: 时间序列, 相似性度量, 时间粒度