Industrial Fault Prediction Based on ARIMA-GRU Hybrid Model

Nowadays, the production process of chemical industry presents the characteristics of high nonlinearity, susceptibility to disturbance and uncertainty, which makes the process parameters generated by chemical process become complex and difficult to handle. In this paper, a hybrid modeling approach of GRU and ARIMA neural networks is proposed to carry out risk analysis and prediction studies for chemical processes. The model incorporates the advantages of GRU and ARIMA to improve the accuracy and speed of prediction. Failure data from The Tennessee Eastman Process was used as a dataset to simulate real failures occurring in the plant, and the experimental results show that the prediction results of the GRU-ARIMA hybrid model are more effective than the comparison model because the hybrid model not only outperforms the comparison model in prediction accuracy, but also performs well in terms of speed. The method of this paper can help the safety management personnel to obtain the decision-making basis with reference significance, which can reduce the occurrence of safety accidents to a certain extent.