The minimum miscibility pressure (MMP) is a critical parameter for designing efficient miscible gas injection projects for both enhanced oil recovery (EOR) and carbon capture, utilization, and storage (CCUS). The conventional methods for MMP determination, including slim-tube experiments and equation-of-state (EOS) simulations, are time-consuming and expensive. While machine learning (ML) models have emerged as rapid and accurate alternatives, their black-box nature has hindered their adoption by the industry as they lack physical transparency and trustworthiness. In this study, a novel approach was employed to address the black-box nature of the problem by utilizing an Explainable Machine Learning (XAI) framework that combines Kendall rank correlation, feature, and SHAP analysis using a comprehensive data set of over 1,800 experimental MMP measurements. Different ML algorithms were used to predict the MMP in conditions of interest, which are CCUS and EOR applications. The combination of the outcomes of these analyses provides insight into the thermodynamic drivers of miscibility. The analysis quantitatively ranks feature importance, confirming the dominant role of reservoir temperature and C7+ molecular weight, and reveals complex, nonlinear interactions between injection gas impurities and oil composition that align with physical principles. The resulting Random Forest model achieved exceptional accuracy (R2 > 0.990) and was validated to ensure reproducibility across a wide range of reservoir and gas compositions. Critically, the XAI results not only quantify the nonlinear influence of parameters such as fluid intermediate composition and reservoir temperature but also reconcile the model’s predictions with fundamental thermodynamic principles, establishing it as a trustworthy and rapid screening alternative to time-intensive EOS simulations for CCUS and EOR applications.
Lidong et al. (Fri,) studied this question.