Advancing Urban Gas Safety: Machine Learning-Driven Detection of Subterranean Methane Leaks

To enhance the safety of urban gas pipelines in China, this research presents a robust method for detecting subterranean gas leaks, addressing challenges like biogas interference and environmental variability. The methodology begins by identifying anomalous methane patterns based on expert insights. These patterns undergo systematic data processing, including cleaning, reducing, and deriving significant upward trends. Feature engineering then organizes these trends by analyzing their time-based dynamics, volatility, and temperature influences. Several classification techniques, such as K-nearest neighbor, decision tree, and support vector machine, were evaluated, with the decision tree algorithm yielding the highest performance, achieving a 92.86% recall rate in methane leak detection. Moreover, the decision tree model emphasizes the importance of certain features, particularly the peak values in the processed methane concentration data and the slopes of the upward trends, as key indicators for detecting methane leaks.