Hydrogen jet fire accident prediction model for hydrogen refueling station based on hybrid neural network

Abstract

The burgeoning hydrogen energy sector, particularly in China, is a key part of the ongoing energy transition. However, the rapid expansion of hydrogen refueling stations (HRSs) introduces significant safety concerns due to potential high-pressure hydrogen leaks leading to fires or explosions. Accurate consequence prediction is crucial for effective risk mitigation. While numerical simulations offer precision, they are computationally intensive, hindering timely responses. Theoretical models provide faster results, but their applicability diminishes in complex scenarios, such as when jet fires interact with obstacles. To address these limitations, this study introduces a hybrid model-based and data-driven approach that combines the strengths of Generative Adversarial Network (GAN) and Long Short-Term Memory (LSTM) and incorporates a model-based flame parameters prediction method. This model aims to rapidly and accurately predict the flame shape and temperature distribution of hydrogen jet fires, even when obstacles are present. The numerical simulation data used for model training have been validated through literature experiment, ensuring the accuracy of the predictions. Similarly, the accuracy of the model's predicted flame width data was assessed using experimental data, with an average relative error of 7.96 %. The results demonstrate that the model surpasses traditional numerical methods in terms of efficiency and is capable of handling complex scenarios, providing rapid and relatively accurate predictions. It serves as a valuable tool for process safety and risk management, offering substantial support for safety management and emergency decision-making at hydrogen refueling stations.