A new approach to finding similarities between time series using Cross Wavelet Phase Variance

摘要: This study presents a method for class identification and classification of multiple time-series data using Cross Wavelet Phase variance (Pv). The Pv method can be used to emphasize regions of frequency commonality or regions of frequency dissimilarity between datasets. By focussing on similarities in the frequency domain, the technique performs better than traditional techniques such as Euclidean distance and the euclidean distance of FFT amplitude in classifying the University of California Riverside (UCR) benchmark time series. The possibility that the Pv method will accommodate lead and lag time is hypothesized for future testing.