Adaptive Discovery of Time-Warped and Length-Variable Patterns in Time Series

Time series often contains similar subsequence pattern. Extracting these frequent patterns helps us to analyze and understand the information contained in time series, such as temperature variation trends, ECG signal patterns, etc. In recent years, numerous methods have been applied to find frequent similar subsequence patterns in time series. However, most of these methods use Euclidean distance to measure the similarity between subsequences. Experimental results demonstrate that in real datasets, the Dynamic Time Warping (DTW) algorithm outperforms Euclidean distance measurement. Therefore, this paper proposes a novel model for extracting similar subsequences with varying lengths and shape warping within time series. Experiment results demonstrate the effectiveness of our model on multiple datasets.