Multivariate Time Series Co-evolution Shapelet Learning Method.

Shapelets are discriminative subsequences in time series that effectively capture local shape characteristics. However, the time complexity of the existing shapelet extraction process remains high. To address this issue, this paper proposes an unsupervised shapelet extraction algorithm. The algorithm utilizes the coherence of multiple wavelets and constraints on multivariate collaborative evolution intervals to constrain the generation range of shapelets, thereby improving the efficiency of shapelet learning while preserving their significance. The superiority of the proposed algorithm is validated through experiments conducted on the UEA dataset.