Spatial distribution and machine learning-based prediction model of natural gas explosion loads in a utility tunnel

Abstract

A reasonable natural gas explosion load model is essential to evaluate the damage to the utility tunnel. This study systematically investigates critical parameters that affect gas explosion loads in the utility tunnel by the finite volume method, including ignition point location, methane volume, methane concentration, internal obstacles, and utility tunnel structure. An explosion load model in a typical utility tunnel is established using the artificial neural network (ANN), with peak overpressure, impulse, and arrival time as output parameters. The spatiotemporal distribution equation of gas explosion loads in a utility tunnel is given, and the importance analysis of parameters is also carried out. The results indicate that when the methane volume is larger than 300 m3, the detonation phenomenon generally occurs in the utility tunnel, and the effect of the obstacle on the explosion loads also increases. The ANN-based model can accurately describe the explosion overpressure and duration at each section in the utility tunnel. The methane volume and cross-sectional size of the utility tunnel have the most significant impact on the natural gas explosion. The important coefficient of the cross-sectional size of the utility tunnel for peak overpressures reaches 30.94 %, and the important coefficients of the methane volume for explosion impulses and arrival time are 47.31 % and 39.64 %. ANN-predicted model is efficient and quick to estimate the gas explosion loads for structural dynamics analysis.