RESEARCH ON QUANTIFICATION OF HAZOP DEVIATION BASED ON A DYNAMIC SIMULATION AND NEURAL NETWORK

Hazard and operability (HAZOP) analysis has become more significant as the complexity of process technology has increased. However, traditional HAZOP analysis has limitations in quantifying the deviations. This work introduces artificial neural networks (ANNs) and Aspen HYSYS to explore the feasibility of HAZOP deviation quantification. With the proposed HAZOP automatic hazard analyzer (HAZOP-AHA) method, the conventional HAZOP analysis of the target process is first carried out. Second, the HYSYS dynamic model of the relevant process is established to reflect the influence of process parameters on target parameters. Third, to solve the problem of deviation identification based on multi-attribute and a large dataset, we use the ANN to process the input data. Finally, HAZOP deviation can be quantified and predicted. The method is verified by the industrial alkylation of benzene with propene to cumene. The results show that the predicted deviation severity can be close to the actual deviation severity, and the accuracy of prediction can reach nearly 100%. Thus, the method can diminish the probability of conflagration, burst, and liquid leakage.