Developing a Robust Intelligent Model for Predicting Minimum Miscibility Pressure in Hydrocarbon Gas - Oil Systems: Implication for Miscible Gas Enhanced Oil Recovery Process

Abstract Miscible gas-based enhanced oil recovery (EOR) methods are widely regarded as effective techniques for improving oil recovery when natural reservoir drive mechanisms become insufficient. However, the successful application of these methods relies on accurate prediction of key parameters, particularly the minimum miscibility pressure (MMP) in hydrocarbon gas–oil systems. The experimental determination of MMP is costly, time-consuming, and resource-intensive, making the need for reliable and rapid predictive tools more pressing. This study introduces two advanced machine learning (ML) models, namely artificial neural network (ANN) and extra trees (ET), to predict MMP in hydrocarbon gas–oil systems. A comprehensive experimental dataset consisting of over 100 datapoints of MMP, was utilized to train and validate the models. Various evaluation techniques, including statistical metrics, graphical analysis, trend evaluations, and the Shapley Additive exPlanations (SHAP) method, were employed to assess the robustness, accuracy, and interpretability of the models. The results demonstrated that the ANN model outperformed the ET model, achieving a root mean square error (RMSE) of only 1.1883 MPa. Graphical error analysis and trend comparisons further confirmed that the ANN model closely aligned with experimental MMP values across a range of operational conditions. Additionally, SHAP analysis highlighted the transparency and interpretability of the ANN model, providing insights into the influence of input variables on MMP predictions. The findings underline the advantages of using ML-based models for MMP prediction, offering significant improvements in terms of efficiency, cost-effectiveness, and operational optimization. These models can aid in making appropriate decisions about the timing and volume of hydrocarbon gas injection in EOR processes, contributing to more effective planning and execution of gas injection strategies.