A method for real-time estimation of gas leakage flow from leakage source based on point detection data

Abstract

Chemical leakage accidents may lead to significant hazards such as fire, explosions and the release of toxic gases. And the estimation of real-time accident hazards is of critical importance for emergency response. In this study, a method was proposed to predict the gas flow in real-time based on the simulated database. Based on datasets generated by numerical computations, a machine-learning algorithm was trained and used to then be validated by data from real-world scenarios. Compared with the full-scale experimental data, for the scenarios in which the relative error of the simulation was within the range of 0%–25%, the relative error of the developed model prediction was 9.8%. When the error of the simulation was within the range of 0%–25%, the machine learning model trained by simulated data can predict the gas leakage in real world with high accuracy.