Abstract The rapid expansion of underground natural gas pipeline networks in urban areas increases the chances of methane leaks that are difficult to detect and thus create safety and environmental issues. In this paper, an adaptable machine learning system is introduced to locate methane leaks quickly and to decrease the risk of methane leaks in underground gas pipelines. The Evolutionary Optimization (EVO) and Firefly Algorithm (FLA) are used to optimize CATB, ANFIS, and Histogram Gradient Boosting (HGB) predictive models so that their detection accuracy and computational performance can be enhanced. The merged dataset is a source of operational and meteorological parameters, while feature selection based on the eigenvalue is used to extract the most relevant variables for model training. The optimized models have higher predictive accuracy and lower computational costs, so the leak can be detected at an early stage and decision-making can be improved. The paper introduces the potential of the system as a vehicle for real-time leak detection, emission control, and regulatory compliance that, in turn, results in safer and more sustainable gas infrastructure management.
Liu et al. (Sun,) studied this question.