Operating Model for Similarity Search in Streaming Time Series Using Multiresolution Index

Abstract

Similarity search in streaming time series is a challenging problem of data mining in recent years due to the complexity and difficulty in processing streaming time series and lack of an operating model to handle the problem efficiently. For this reason, the paper proposes an operating model for similarity search in streaming time series under the Euclidean distance and data normalization. The operating model exploits multithreading available in contemporary computer systems to perform similarity search for many static time-series queries in multiple streaming time series simultaneously. More especially, when there is a newly incoming subsequence of streaming time series, the operating model uses a multiresolution index to filter the candidate queries through the resolution levels to search for queries similar to the subsequence. Experiments on the range searching method implementing the operating model, and a state-of-the-art method of similarity search, show that both methods obtain the same found results; however, the execution time of the former is less than that of the latter.