A hydrogen concentration evolution prediction method for hydrogen refueling station leakage based on the Informer model

Abstract

The utilization of hydrogen energy contributes to the alleviation of energy crisis and environmental pollution. Hydrogen refueling stations are essential for hydrogen energy applications but operate under high-pressure storage, which increases the risk of hydrogen leakage. Safe use of hydrogen energy requires vigilance on leakage issues at refueling stations and implementation of practical monitoring and warning measures. This study focuses on predicting the evolution of hydrogen diffusion concentrations at refueling stations using deep learning, addressing the timeliness limitations of conventional gas concentration prediction methods. Therefore, we present H2-Informer, a model specifically designed for predicting the hydrogen diffusion process, built on the Informer architecture. Using sparse sensor concentration data along with wind speed, wind direction, and height information, H<inf>2</inf>-Informer predicts two-dimensional planar hydrogen diffusion distributions at multiple future time points, including the evolution of hydrogen diffusion at different heights. After hyperparameter tuning, the H<inf>2</inf>-Informer model achieves an R2 of 0.9775 and an MSE of 1.96 × 10−5, with an inference time of only 1.5 s. This significantly reduces prediction time compared to CFD simulation and meets real-time prediction requirements. Compared with the traditional Transformer model, the H<inf>2</inf>-Informer model predicts the hydrogen diffusion concentration distribution in the next 30 time steps with the R2 remaining above 0.9, which shows stronger fitting ability and shorter inference time in long series prediction. In summary, the H<inf>2</inf>-Informer prediction model is capable of quickly and accurately predicting the concentration evolution of hydrogen leak diffusion in hydrogen refueling stations. It helps the hydrogen refueling station to quickly take emergency measures in case of a leakage accident, and improves the safety management level of the station.