Risk Prediction and Control Study of a Multitower Separation Process Based on DQRA and Bi-LSTM

Due to the complexity of the multitower separation process (the ethylene separation process as an example) and the numerous variables, traditional risk analysis methods cannot meet the needs of modern enterprises. In this paper, an intelligent dynamic quantitative risk assessment combining bidirectional long short-term memory network (DQRA-Bi-LSTM) is proposed for multitower separation process risk early warning. First, a dynamic simulation of the ethylene separation process is carried out to obtain working condition data. Based on the data from the process simulation and the actual conditions of the plant, a preliminary risk assessment of the ethylene separation process is performed using the Dow Chemical Fire and Explosion Index (F&EI) method. Then, the dynamic simulation data are quantitatively converted to risk values using risk definitions and predicted by a bidirectional long short-term memory network (Bi-LSTM). Finally, genetic algorithms (GAs) are introduced to control process risk. The dynamic quantitative risk analysis method is applied to the demethanization system and the ethylene distillation system to predict risk values. The application results of the two cases show that the proposed method is able to predict the system risk threshold 0.2 h in advance and successfully control the risk value below the threshold.