Granger Causality Models with Gaussian Kernel Functions for Industrial Alarm Correlation Analysis

The alarm analysis between different variables is recognized as a complex and important subject in the process industry. However, the discrete type of alarm data and the unclear time-delay between alarm data of different variables discourage the use of traditional causality analysis methods. In this paper, a Granger causality model based on Gaussian kernel functions is used to analyze the correlation of industrial alarms. Firstly, aiming at the reduction of false alarms caused by chattering alarms and binary operating data associated with switching valves, the Gaussian kernel function method is used to convert binary data into continuous sequences. Secondly, Dynamic Time Warping approaches are used to perform similarity matching on continuous sequences, screening and correlating alarms. Finally, the Granger causality model is used to obtain the correlation between different variable alarms.