Chemical Fault Diagnosis Method based on DAM-CGRU

In order to better capture the complex relationship of chemical process data and improve the ability to identify various faults, a fault diagnosis method based on dual attention mechanism (DAM) and convolutional gating recurrent unit (CGRU) is proposed. Firstly, the chemical fault data is standardized and sub-sequence samples are divided using sliding window, and then high-dimensional local features and time series features of the subsequence are extracted through one-dimensional convolutional neural network(1D-CNN) and gate recurrent unit(GRU), and the improved attention score function is used to empower the variable dimension and time step dimension of the fault data respectively to enhance the model's attention to key variables and time nodes, and finally, the softmax classifier is used to classify various chemical faults. Verified and compared on the Tennessee-Eastman dataset, the experimental results show that the DAM-CGRU method can extract chemical fault characteristics better than the traditional deep fault diagnosis method and obtain higher diagnostic accuracy.