Time series clustering based on shape dynamic time warping using cloud models

A growing attention has been paid to mining time series knowledge recently. Euclidean distance measure is used for comparing time series. However, it may be a brittle distance measure as less robustness. In this paper, a modification algorithm, which applied linguistic variable concept tree to describe the slope feather of time series, was presented named dynamic time warping. For reducing the computational time and local shape variance disturbing, the piecewise linear representation was used to process warping path. Moreover, the linguist concept tree was developed based on cloud models theory which integrities randomness and probability of uncertainty, so that made conversion between qualitative and quantitative knowledge. Experiments about cluster analysis on the basis of this algorithm, compared with Euclidean measure, were implemented on synthetic control chart time series. The results showed that this method, presented in this paper, have strong robustness to loss of feather data due to piecewise segment preprocessing. Moreover, after the construction of shape concept tree, we can discover knowledge of time series on different time granularity.