Real-time gas explosion prediction at urban scale by GIS and graph neural network

• Published in: Applied energy Vol. 377; p. 124614
• Main Authors: Shi, Jihao, Li, Junjie, Zhang, Haoran, Xie, Bin, Xie, Zonghao, Yu, Qing, Yan, Jinyue
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.01.2025
• Subjects: GIS; Graph neural network; Liquified natural gas; Urban energy transition; Urban scale gas explosion; GIS; Urban scale gas explosion; Urban energy transition; Graph neural network; Liquified natural gas
• ISSN: 0306-2619
• DOI: 10.1016/j.apenergy.2024.124614

Liquified gases are expected to play the significant roles in the context of urban energy transition. However, the accidental release of liquified gases induces a flammable vapor cloud, almost 10 times larger than that of non-liquified gases. Its ignition in a congested urban block will result in even greater magnitudes of blast load, catastrophic for both structures and people at urban scale. Machine learning approaches have been developed to efficiently assess the consequence of gas explosion, while still having poor 。performance to capture the interaction between congested buildings and blast wave propagation. This study aims to develop an integrated approach of Geographic information systems (GIS) and graph neural network (GNN), in which GIS provide 3D building geometries and grid information at urban scale. Computational fluid dynamics (CFD) simulation will be conducted to construct the benchmark dataset, by using which, the autoregressive GNN approach is developed to solve the interaction between congested buildings and blast wave propagation. Comparison with the existing machine learning-based prediction approaches is conducted. The results demonstrate our proposed approach has the higher R2 value of 0.946 and lower MSE value of 5.36E-4, indicating its more accuracy to gas explosion evolution prediction especially at the congested urban areas. Our approach also exhibits 1000× computational speedup improvement compared to CFD approach at urban scale. This proposed approach has the potential to efficiently and accurately analyze the worst-case destructive effects of gas explosion at urban scale, supporting the decision-makings to improve urban resilience in the context of urban energy transition. •A novel approach merges GIS with GNN for accurate urban-scale gas explosion prediction.•Outperforms current machine learning and CFD methods in predicting gas explosion impacts.•Introduces a real-time prediction model that considers complex urban geometries.•Utilizes a comprehensive CFD-simulated dataset for robust GNN model training.•Provides a decision-making tool for improving safety in the face of urban gas explosions.