Parallel Variable-Length Motif Discovery in Time Series Using Subsequences Correlation

Abstract

The repeated patterns in a long time series are called as time series motifs. As the motifs can reveal much useful information, time series motif discovery has been received extensive attentions in recent years. Time series motif discovery is an important operation for time series analysis in many fields, such as financial data analysis, medical and health monitoring. Although many algorithms have been proposed for motifs discovery, most of existing works are running on single node and focusing on finding fixed-length motifs. They cannot process very long time series efficiently. However, the length of motifs cannot be predicted previously, and the Euclidean distance has many drawbacks as the similarity measure. In this work, we propose a parallel algorithm based on subsequences correlation called as PMDSC (Parallel Motif Discovery based on Subsequences Correlation), which can be applied to find time series motifs with variable lengths. We have conducted extensive experiments on public data sets, the results demonstrate that our method can efficiently find variable-length motifs in long time series.