An Equidistant Segmentation-based Similarity Measure for Time Series

In order to make up the deficiency of high-dimension, unequal length and much noise of time-series data, a similarity measure combining equidistant segmentation and Dynamic Time Warp has been put forward. First of all, introduce l(1) trend filtration and represent time series feature as broken line of sequence trend. Segment equidistantly the segmented line so as to make the segmentation equal weight; calculate the segmentation slope to obtain a time series equal interval slope representation, which preserves the trend information as well as reduces the dimensionality. Finally, combine time series slope representation with Dynamic Time Warp to calculate unequal sequence distance. The comparison experiments show that the proposed algorithm outperforms the latest time series data similarity measure, and the clustering analysis proves the effectiveness of the algorithm.