Multi-Model Collaborative Inversion Method for Natural Gas Pipeline Leakage Sources in Underwater Environments

The identification of leakage sources in underwater natural gas pipelines (UNGPs) remains a critical challenge due to complex environmental conditions. In this study, we propose a novel simulation–optimization method, integrating numerical bubble plume dynamics models with surrogate models to enable accurate leakage parameter inversion. First, a bubble plume underwater motion simulation model was developed based on the actual conditions of the study area to predict the future spatial and temporal variation characteristics of the bubble plumes in certain wave fields. Then, the simulation–optimization method was applied to determine the leakage velocity and offset distance of the underwater gas pipeline leakage source via inversion. To reduce the computational load of the optimization model by repeatedly invoking the simulation model, the Kriging method and a backpropagation (BP) neural network were used to build surrogate models for the numerical model. Finally, the optimized surrogate model was solved using the simulated annealing method, and the inverse identification results were obtained. The experimental results show that both methods can achieve a high inversion accuracy. The relative error of the Kriging model is no more than 12%, and the running time is 13 min. Meanwhile, based on the BP neural network surrogate model, the relative error of the BP neural network model is about 14%, and the running time is 2.5 min.