Study on Minimum Miscibility Pressure of CO₂–Oil System in Deep High-Temperature and High-Pressure Reservoirs

Deep high-temperature and high-pressure (HTHP) oil reservoirs have limited experimental MMP data, large differences between reservoir and saturation pressures, low gas–oil ratios, and pressure-sensitive CO₂–oil phase behavior, which make both minimum miscibility pressure (MMP) prediction and miscibility-mechanism identification challenging. To address these gaps, this study determines the MMP of a CO₂–oil system by integrating slim-tube experiments, empirical formula methods, the Multiple Mixed-Cell (MMC) method, the Method of Characteristics (MOC), compositional numerical simulation, and three intelligent algorithm models (GWO-RBF, GWO-LSSVM, and GWO-SVM). The slim-tube MMP of 44.13 MPa at 140 °C is used as the experimental reference for comparing prediction errors, whereas PVTsim and literature data are used for consistency checks and model benchmarking. The results show that when the injected CO₂ mole fraction exceeds 0.88, the formation oil under original reservoir conditions cannot achieve first-contact miscibility with CO₂, and the maximum dissolved CO₂–oil molar ratio is 7.3:1. Supercritical CO₂ forms dual displacement mechanisms, including front-end vaporizing miscible drive and rear-end condensing miscible drive, but the dominant mechanism for this CO₂–oil system is vaporizing miscible drive. During the vaporizing gas drive, the CO₂ + N₂ + C₁ content in the liquid phase increases from less than 60% to nearly 90%, indicating significant CO₂ dissolution into oil and associated density and viscosity reduction; meanwhile, the C₇₊ content in the gas phase increases to nearly 10%, indicating extraction of heavy components. Relative to the slim-tube reference at 140 °C, the deviations of MMC, GWO-SVM, GWO-LSSVM, compositional numerical simulation, GWO-RBF, MOC, and empirical formula methods are 2.97%, 3.08%, 3.40%, 4.24%, 4.26%, 11.62%, and 19.74%, respectively. The MMC method is the most suitable approach for this specific HTHP oil system, while intelligent algorithms should be regarded as supplementary predictors whose reliability depends on training-domain coverage and independent validation.