Time Series Similarity Search Under Piecewise Dynamic Time Warping

Nowadays we are entering a new era of data exploding.Time-series data is a collection of organized data which can be easily obtained from climate,financial and medical applications.The nature of time-series data includes:large in data size,necessary to update continuously and wide in distribution...etc.and time-series has been increasingly drawing attentions.Recent years,time-series data has initiated various of research,which can be generally categorized into query,clustering,classification,segmentation,prediction and anomaly detection.Almost every time-series mining task requires a subtle notion of similarity search,thus we can safely conclude that similarity search is of fundamental importance for varieties of time-series data mining tasks.In this article,we focused on subsequence matching method over data stream and similarity join over time-series dataset.The main work of this article are as follows:1.A comprehensive revision on existing time-series data mining research is given,the literature is divided into data representation techniques,distance measure and indexing methods.Moreover,review recent time-series data mining directions and research trends for future work.2.Propose a subsequence matching system over data stream under dynamic time warping.The idea is to adaptively segment the data stream,factorize each segment with Chebyshev polynomials and realize an efficient online matching algorithm,such that the incremental DTW calculation over data stream is implemented.Finally we confirm the effectiveness and efficiency by comparing with other classical algorithms.3.Propose the algorithm of similarity join on time-series under dynamic time warping.Concretely,time-series in the dataset are adaptively segmented,and statistical feature extraction is applied for the purpose of dimension reduction.A DTW lower bound which satisfies the triangle inequality is adopted,thus making M-tree indexing possible.Therefore,similarity join over time-series under DTW can be finally solved with index query.Experiments over different datasets confirmed that the algorithm is more stable and efficient compared to others.