Research on the Reliability Assessment Method of a Natural Gas Pipeline under Third-Party Excavation Loads Based on Machine-Learning Models

Third-party excavation activities are among the primary causes of pipeline failure. A scientific evaluation is required to prevent failures of natural gas pipelines under such activities. This study develops a comprehensive quantitative evaluation method based on a machine-learning model to assess the reliability of natural gas pipelines under excavation impact loads. First, a feature selection algorithm considering multidimensional evaluations is established to enhance the machine-learning model’s performance. Second, the Shapley additive explanations (SHAP) model interpretation method is applied to reveal the prediction mechanisms, enhancing the model’s reliability and credibility. Third, an explicit limit state function for pipeline failure under excavation impact loads is developed based on the machine-learning model, and a reliability calculation method for natural gas pipelines is constructed using the Monte Carlo simulation approach. Finally, the feasibility and effectiveness of the proposed method are validated through case analysis. The results show that (1) the proposed multimetric recursive feature elimination with cross validation algorithm considers both the mean absolute percentage error and coefficient of determination (R2) evaluation metrics, improving the model’s goodness of fit and prediction accuracy; (2) the SHAP model interpretation method explains the prediction process from three perspectives: local predictive mechanisms, global predictive mechanisms, and feature importance, validating the model’s reliability and credibility; and (3) in the case analysis of the reliability calculation method, the convergence time for a single iteration is less than 130 s, ensuring efficient computational performance.