Gas‐Hydrate Reservoir Identification in Permafrost Regions Using a GWO‐Optimized LightGBM Model

Abstract

Accurate identification of gas‐hydrate‐bearing layers in permafrost regions is critical for gas‐hydrate exploration and development. Conventional well‐logging methods are often affected by lithology, pore structure, and other interfering factors, leading to complex and ambiguous responses that hinder precise evaluation of gas‐hydrate saturation. Machine learning models offer an automated approach to well‐logging interpretation, improving efficiency and reducing dependence on expert knowledge. In this study, four models—Light Gradient Boosting Machine (LightGBM), Extreme Gradient Boosting (XGBoost), support vector machine (SVM), and K ‐nearest neighbors ( K NN)—are optimized using the grey wolf optimizer (GWO) algorithm and subsequently compared. The results demonstrate that the GWO–LightGBM model achieves an overall identification accuracy of 97.14%. For the gas‐hydrate class (the positive class), it yields an F 1‐score of 84.82%. The proposed model significantly outperforms the other three comparison models. Furthermore, validation using independent well data and SHAP (SHapley Additive exPlanations) analysis confirms the excellent generalization ability and interpretability of the proposed model.