An Effective Constraint-Based Anomaly Detection Approach on Multivariate Time Series

Zijue Li, Xiaoou Ding, Hongzhi Wang

Abstract: With the development of IoT, various sensors are deployed in industry applications. Sensors produce multivariate time series, while error data and abnormal values often exist in the data. Correlation in multivariate time series can be used to identify such anomaly. In this paper, we propose an efficient method to utilize the correlation between multivariate time series with constraint-based anomaly detection. We develop a DP algorithm to execute the detection process, and optimize the algorithm efficiency with 2D range tree. Experiments on real IIoT dataset demonstrate the superiority of our proposed method compared to the prediction based models.