Abstract:

This paper presents a method based on information entropy and dynamic time warping (DTW) to measure the similarity of multivariate time series. Firstly, DTW based on the Mahalanobis Distance considers the interrelationships among the variables of the multivariate time series, through the dynamic warping to align time series of different length. Secondly, adapting the information entropy theory, the Mahalanobis distance matrix is learned by minimizing the loss function, which can obtains the global optimal Markov matrix. In order to verify the effectiveness of the proposed algorithm, the five data sets in the UCI data set were used to classify through the nearest neighbor classification algorithm. Experimental results show that this method has higher classification accuracy and less time consumption than other methods, which proves the effectiveness of the proposed method.

Key words: multivariate time series, similarity, dynamic time warping, mahalanobis distance, information entropy