Similarity Search in Multiple High Speed Time Series Streams under Dynamic Time Warping

Finding all similar time-series patterns in real time under Dynamic Time Warping (DTW) is a huge challenge in nowadays data mining. A vital requirement of the critical task is data normalization so that the search results are accurate. However, DTW and data normalization, particularly in the streaming context, cost great deals of computation time and memory space; so many techniques are required to reduce the time and space complexity. In the paper, we introduce an efficient method, which similarly searches numerous time-series queries over multiple streaming time-series under DTW. The search method utilizes many advanced techniques as cascading lower bounding functions, incrementally updating the envelopes of time-series subsequences, and incrementally normalizing timeseries data so that the computation time is minimized. Furthermore, we exploit multi-threading technique so that the method has a fast response to time-series streams at high-speed rates. The experimental results reveal that the method obtains the same accuracy as similarity search in static time series.