Predicting Gas Hydrate Saturation in Fine-Grained Sediments Using Machine Learning: A Case Study of the Shenhu Area in the Northern South China Sea

In fine-grained marine sediments dominated by clayey silt and silt, gas hydrate saturation has been shown to have a highly nonlinear relationship with well-logging data and reservoir petrophysical properties. This complexity arises from such factors as strong reservoir heterogeneity, high clay content, and low permeability. Therefore, accurately predicting hydrate saturation has remained a significant challenge. On the basis of conventional geophysical well-logging techniques, in this study, we applied five machine learning (ML) algorithms to estimate hydrate saturation. We used measured saturation data and well-log records from three sites in the Shenhu Area to develop predictive models and applied them to estimate hydrate saturation at unmeasured locations. According to our results, resistivity and Delta-T compressional wave from a monopole source (an acoustic logging parameter, DTCO) had the strongest correlation with hydrate saturation. Using either parameter alone or their simple combination, however, resulted in limited predictive accuracy. The optimal feature set to predict hydrate saturation typically includes 3–4 types of logging data, and it must contain at least either resistivity or DTCO. Additionally, because gamma ray (GR) logging has low correlation with other parameters, it offers complementary information that is independent of core features. This enhances the predictive accuracy of ML models under complex lithological conditions. Among the five ML algorithms evaluated, extreme gradient boosting (XGBoost) achieved the best predictive performance. It attained a high coefficient of determination (R²) of 0.9242 on the test set, and its predicted saturation curve closely aligned with the measured data. In this study, we demonstrated the accuracy and reliability of ML algorithms to predict hydrate saturation. These results offer a valuable technical approach to quantitatively evaluate hydrate resources in the Shenhu Area and provide theoretical support for the industrial development of gas hydrates.