Prediction of a Hydrogen Vapor Cloud Explosion with a Barrier Wall Using Various Machine Learning Methods

Hydrogen is considered the next energy to replace fossil fuels, but it must be handled with care given that it is a flammable gas. A barrier wall is an effective way to mitigate the effect of an explosion, and to build a safe barrier wall, research on hydrogen explosions is necessary. Experiments and CFD (computational fluid dynamics) are two commonly used methods, but both are costly to use under any condition. Machine learning can be used to enhance the data from experiments and CFD as the trained model can predict explosion pressure levels very rapidly under various conditions. We propose the prediction of a hydrogen VCE (vapor cloud explosion) with a barrier wall using various machine learning methods. This research uses CFD simulation data from KAERI (Korea Atomic Energy Research Institute) as training data. MLP (multi-layer perceptron), LSTM (long short-term memory), and the Transformer architectures are used to train the hydrogen VCE and are compared. In our research, MLP produces the best score among all learning processes, with an R2 value exceeding 0.97, outperforming both LSTM and Transformer in terms of accuracy and speed. The trained machine learning model can be used to build safe barrier walls in hydrogen refueling stations. Evaluating the safe distance from the barrier wall and evaluating the optimal position of the barrier wall are possible usages.