Abstract: Hydrogen is one of the most flammable and explosive fuels. Once leaked, it can easily diffuse and potentially cause a fire. Real-time and accurate prediction of hydrogen diffusion is essential for predicting spatial concentration, which enables fire prevention of hydrogen facilities. In this study, a physics-informed deep learning model was proposed to effectively and accurately predict hydrogen concentration and velocity using sparse sensor data. The dependency between sensor data was learned by the graph neural network, and the physical differential equations of hydrogen diffusion were solved by graph nodes. The computed residuals were then used to optimize the parameters of the deep learning model. Public experimental data was applied to validate the performance of our proposed model. The results show that compared with the existing methods, the proposed method not only has real-time capability, but also predicts hydrogen concentration and velocity more accurately. This study provides accurate and real-time concentration and velocity prediction for hydrogen diffusion, facilitating hydrogen fire prevention.

Key words: new energy, fire prevention, hydrogen, deep learning, graph neural network