A fusing NS with NN model for the consequence prediction of vapor cloud explosion

Vapor cloud explosions (VCEs) have been considered as a major hazard in petrochemical industry, accompanying with wide-ranging impact and huge destruction. The existing methods are incapable to make a rapid and accurate estimation when considering multi-factor coupling effects. Therefore, this study proposed a novel methodology of fusing numerical simulation (NS) with neural network (NN) technique for the prediction of explosion consequences. First, 6 parameters of VCEs influencing overpressure are selected as variables of a database. A CFD method is employed for simulating VCEs in a chemical site, by which sufficient blast data are generated. After the architecture of a NN model is determined, data on three generic VCEs are extracted for further model training process. A progressive training method is adopted to develop a general prediction model. Furthermore, data derived from ongoing simulation results are imported into the model for its constant self-improvement. The output of the well-trained model is subsequently transformed into a probabilistic function to assess the domino effect. The integrating NS with NN approach provides an accurate and efficient way to predict the blast effects, which can support more scientific rescue decision-making. Finally, the proposed model is applied to a case study for illustration.