Abstract: Time series have the characteristics of large amount of data, high dimensionality and fast update speed, which makes it difficult for general piecewise linear algorithms to characterize the global trend of the original time series. Aiming at the characteristics of time series, this paper proposes an autonomous piecewise representation method based on temporal edge operators (APLR_TEO), which can describe the shape features of time series effectively. Firstly, the temporal edge operator is convolved with the original time series and the edge extremum is obtained according to association rules. Then, according to the changing characteristics of the time series, the association rules of trend turning distance are adopted to carry out autonomous linear segmentation and get the key points. Finally, the sequence of key points can represent approximately time series. The experiments show that APLR_TEO can effectively depict shape features of the sequence. This algorithm has good adaptability and stability among data sets of different sizes and effectively reduces the fitting error.

Key words: time series, temporal edge operator, autonomous piecewise linear, edge extreme point, fitting error