A Reliable Model for Predicting Methane Solubility in Brine: Toward Effective Methane Emission Mitigation

Methane solubility in brine is a critical parameter in various natural and industrial processes, including gas production, storage, chemical engineering, and environmental studies. This study develops and validates explicit predictive correlations for methane solubility in brine using two robust white-box machine learning techniques, including the multigene genetic programming (MGGP) and the group method of data handling (GMDH). A comprehensive database of 684 experimental data points, covering extensive operational conditions of pressure (0.433–140 MPa), temperature (273.15–493.15 K), and NaCl molality (0–5.4 mol/kg), was employed to train and test the proposed correlations. The accuracy of the correlations was assessed using statistical metrics, error distribution analyses, and graphical techniques. Results revealed that both correlations yielded excellent predictive accuracy, with the MGGP-based scheme consistently outperforming the GMDH model by achieving a lower total RMSE value of 0.000243. Furthermore, the trend analysis and SHAP (SHapley Additive exPlanations) method demonstrated that the MGGP-based correlation successfully captured the physical trends of methane solubility in methane with respect to the input features. Additionally, a comparison with some prior paradigms showed the superior accuracy and computational efficiency of the MGGP-based correlation. Lastly, the applicability domain of the MGGP-based correlation was defined using the Leverage approach, confirming its robustness, with 98.538% of the data points falling within the trust region. The explicit nature of the MGGP-based correlation allows for its integration into environmental management and sustainability frameworks, facilitating improved assessments of methane behavior in natural systems. This integration enables the development of more efficient and effective strategies for methane emission mitigation and climate change mitigation efforts. Furthermore, these models enhance sustainability by supporting the design of environmentally responsible industrial operations. Beyond these applications, the correlation is also well-suited for phase equilibrium modeling, reservoir simulations, and optimization tasks, offering a versatile tool for both environmental and engineering purposes.