Data-driven Anomaly Detection and Forewarning Based on Grey Prediction Model

Abstract

Abstract: Among the endeavors towards high-quality development of manufacturing enterprises, the most fundamental bottom line is to ensure safety and prevent risks, and the data generated in manufacturing processes reflects potential risks in real time. Therefore, in this paper, through the case analysis of the time series data recorded by the equipment in the factory, the possible types of risks are obtained. The extent of deviation of exceptional value is acquired by fitting the normal data to indicate the degree of anomaly of the equipment. The paper proceeds to the building of a grey prediction model based on the model to predict the situation in the next hour, and then residual diagnostics and class ratio dispersion diagnostics are carried out to test the accuracy of that prediction, and the sensitivity analysis and overall evaluation on the prediction are made.