Real-time reconstruction of hydrogen leakage concentration field based on transient sparse monitoring data in hydrogen refueling stations

This study proposes a model for real-time reconstruction of hydrogen leakage concentration field in hydrogen refueling stations (HRS) using transient sparse monitoring data. The model compresses high-dimensional hydrogen concentration features into low-dimensional representations using the encoder of vector quantized variational autoencoder (VQVAE). A multilayer perceptron (MLP) maps the sparse data to these representations, and a decoder is subsequently used to reconstruct the concentration field. The effect of monitoring point sparsity on the reconstruction accuracy is examined using a genetic algorithm (GA). The results show that the proposed VQVAE-MLP model outperforms other models, proving its effectiveness in compressing high-dimensional data. The relationship between monitoring point sparsity and reconstruction accuracy is explored, which can be used to optimize the sensor layout of real HRS. The reconstruction accuracies of different risk areas were compared by structural similarity index measure (SSIM) metrics, and the effects of wind speed and direction on the reconstruction results were analyzed. In conclusion, the proposed model effectively reconstructs hydrogen leakage risk areas in real time, enabling rapid identification of high-risk zones and enhancing the safety and emergency response capabilities of HRS.