Time Series Anomaly Detection via Rectangular Information Granulation for Sintering Process
Sheng Du, Xian Ma, Min Wu, Weihua Cao, Witold Pedrycz
Abstract
Time series anomaly in the sintering process is a direct manifestation of equipment failure and abnormal operating mode, and effective detection of time series anomaly is important to improve the stability of the sintering process. This paper presents a time series anomaly detection via rectangular information granulation, whose originality is to apply the similarity of information granules as a reference for anomaly detection. It converts time series into rectangular granules, and the similarity of time series is measured with rectangular granules. The one-way analysis of variance method is used to detect the difference for the similarity between the time series to be detected and the historical time series and the similarity between any two historical time series, thus achieving the anomaly detection of the time series. The experiment is conducted on real-world data from an enterprise. The result shows that the proposed method outperforms the probability density analysis method and can effectively detect abnormal time series.