Online classification of abnormal patterns in time series during continuous casting process based on optimal shapelet combination

Abstract

Aiming at the different abnormal patterns of high resolution time series data in industrial control process, an optimal shapelet combination algorithm based on maximum information gain is proposed to complete the classification of time series. First, genetic algorithm is used to extract the candidate shapelet set with a certain shape from the training set of time series data, and then the optimal shapelet combination is extracted from the training set via the maximum information gain, and the time series data set is transformed into a data matrix. Second, the time series classification is completed via combining the data matrix with the conventional classification methods. The experimental results show that the computational efficiency of the proposed method is improved by more than 7 times based on the standard time series data set in UEA & UCR data warehouse. Finally, it is applied to the online classification of abnormal pattern in time series data of casting speed during continuous casting process. The results show that the classification time of the proposed method is reduced by more than 10 times, reaching within 0.5 seconds, which can meet the real-time requirements of industrial anomaly detection.