Methodological improvements in the risk analysis of an urban hydrogen fueling station

Abstract The Fuel Cell Vehicles are going to be introduced in domestic cities of China, which will require urban Hydrogen Refueling Stations (HRS). Such urban refueling center would be a public concern given their location in the congested area and the potential hydrogen release and fire risk. Risk analysis of possible fire scenarios is an efficient approach to identify, evaluate, and mitigate the risk from such hydrogen fire accidents. However, due to lack of availability of data, there still exists high parametric uncertainty in the scenario envisaging risk analysis. It is crucial to minimize this parametric uncertainty to have robust hydrogen fire risk analysis. If minimizing the corresponding uncertainty, traditional procedure could bring much computational intensity. This paper proposes a robustly integrated procedure of the fire risk analysis for the urban HRS to reduce the uncertainty and the computational intensity. This newly proposed procedure integrates the Bayesian Regularization Artificial Neural Network (BRANN)-based non-intrusive method with the consequence modeling and statistical approaches. A case study is conducted to demonstrate the scenario-related parametric uncertainty effect and the feasibility of the procedure. The results indicate 300 simulation inputs are the optimal trade-off between the BRANN model’s generalization capacity and computational intensity. With such number of simulations, the BRANN model could achieve the R2 = 0.98. In addition, the proposed procedure could reduce the parametric uncertainty and computation cost by 97% and 99%, respectively.