Fault Root Cause Diagnosis Method Based on BILSTM to Reconstruct Granger Causality

The traditional Granger causality (GC) method has low accuracy in nonlinear fault root diagnosis applications, causality redundancy, and complex analysis. In this paper, a fault root diagnosis method for industrial processes based on kernel principal component analysis (KPCA) and gray wolf algorithm (GWO) optimization bidirectional long-short term memory neural network (BILSTM) is proposed. Firstly, the KPCA model is adopted to reduce redundant fault variables. Secondly, the GWO-BILSTM prediction model is used to predict the variables, and the prediction error of the endogenous input and the exogenous input is used to determine the causal relationship, and finally, the root of the fault is obtained according to the causal relationship. Experiments show that the KPCA and GWO-BILSTM method have superior performance in reducing redundant causality and improving the accuracy of traceability.