Hybrid neural network-based surrogate model for fast prediction of hydrogen leak consequences in hydrogen refueling station

Hydrogen refueling stations (HRSs) play a vital role in the hydrogen energy industry and are being built worldwide. However, the potential risk of hydrogen leakage poses a significant challenge. Accurately predicting the consequences of such accidents is crucial for effective mitigation. Traditional methods, like numerical simulation, provide accurate results but lack timely predictions. To address this, we propose a hybrid surrogate model that combines Generative Adversarial Network (GAN) and Long Short-Term Memory Network (LSTM), incorporating Deep Neural Network (DNN) for predicting hydrogen leakage consequences in HRSs based on source parameters. The surrogate model was trained using training samples generated through numerical simulation. Reduced-scale experiments were conducted to verify the results of numerical simulations and assess the surrogate model's generalization ability. Results demonstrate the model's ability to predict hydrogen distribution after an accident, indicating its potential for real-time decision making in emergency response plans.