A Time Series Similar Pattern Matching Algorithm Based on Singularity Event Features

Knowledge discovery from time series may help us better recognize the revolving regularity of the system. The state-of-art feature extraction methods from time series are single-scale methods that result in imprecision of the feature location and inferior quality of the discovered pattern. A novelty multiscale feature extraction method from time series is proposed based on the principle of wavelet singularity detection. It determines the number of characterized event at the analytical scale and locates the feature events precisely at finer scales. The time series are then compressed into an event sequence using singularity event feature and a dynamic time warping similarity measure of event sequence is defined. The proposed algorithm is used to match similar pattern of time series based on singularity events. The experimental result shows that it has higher matching precision and lower computing cost.