Similarity Search for Numerous Patterns in Multiple High-Speed Time-Series Streams

Similarity search for patterns over streaming time series is a big challenge in data mining at present. In the paper, we introduce an efficient method, which performs similarity search for numerous patterns in multiple time-series streams at high-speed rates under the Euclidean distance. The search method matches approximately new-coming time-series sub sequences of time-series streams with given patterns using an overlapped segmentation of time-series data. Moreover, this method is designed for range search not only with the same search radius for all time-series patterns but also the own search radius for each time-series pattern. The experimental results reveal that the search method has the same accuracy as similarity search in static time series, and low wall-clock times.