A machine learning-based predictive model for estimating the potential impact radius of hydrogen-blended natural gas pipelines

The significant safety hazard posed by jet fires induced by hydrogen-blended natural gas (HBNG) pipelines primarily stems from the intense and sustained thermal radiation they emit. Accurately estimating the potential impact radius is critical for effective risk assessment and safety planning. This study first develops a machine learning-based predictive model to predict potential impact radius by combining a validated CFD-based jet fire simulation model, radiation intensity thresholds in high-consequence areas, and a machine learning model. By a thorough comparative analysis of backpropagation neural network (BPNN), convolutional neural network (CNN), and random forest (RF), a BPNN with the best prediction performance is used to learn nonlinear relationships between hydrogen blending ratio, pipeline diameter, pressure, and potential impact radius. The results show that in the case of full-size fracture, the potential impact radius of the HBNG pipeline ranges from 10 m to 275 m. The potential impact radius increases with increasing pipeline pressure and diameter while decreases with increasing the hydrogen blending ratio. Compared with the original empirical formula for the potential impact radius of natural gas pipelines, the prediction error is reduced from 14.9 % to 2.8 %, filling the gap in calculating the potential impact radius of HBNG pipelines.