Machine Learning–Based Prediction of the Mechanical Response of Hydrate-Bearing Sediments under Multifield Coupling

Summary The extraction of natural gas from offshore hydrate reservoirs is governed by a complex multiphysics coupling process influenced by various environmental factors. Accurate prediction of the deformation behavior of hydrate-bearing sediments (HBS) under such coupled conditions is essential for safe and efficient resource development. For this study, a comprehensive geomechanical database was constructed, incorporating key parameters such as temperature, sodium chloride (NaCl) concentration, confining pressure, hydrate saturation, and corresponding stress-strain data under multifield coupling scenarios. Three machine learning (ML) techniques—deep neural networks (DNNs), random forest (RF), and extreme gradient boosting (XGBoost)—were used to develop surrogate models capable of predicting stress-strain responses under specific operating conditions. The results show that all three models effectively captured the nonlinear stress-strain relationships, with the RF model demonstrating superior performance in nonlinear regression. These surrogate models accurately and efficiently reproduced stress-strain curves derived from laboratory experiments and numerical simulations across a range of scenarios. With this study, we propose a novel data-driven approach for evaluating the mechanical stability of HBS and optimizing extraction strategies, offering a robust and rapid assessment tool for hydrate reservoir development.