As conventional oil and gas reserves decline, enhanced oil recovery (EOR) becomes vital for improving oil recovery efficiency. Foam flooding reduces gas mobility and improves sweep efficiency, but its performance depends heavily on reservoir conditions and injection strategies. Most existing machine learning (ML) research on foam flooding has primarily focused on pure CO2 foam; the potential of N2–CO2 hybrid foam has not been sufficiently investigated. The hybrid foam flooding offers synergistic benefits by simultaneously reducing oil viscosity and stabilizing the foam, making it an effective EOR method. This study utilizes a data set of 420 laboratory data points gathered from literature sources. This study presents a data-driven ML framework that integrates three algorithms and two improved optimization methods, the mapping hiking optimization algorithm (MHOA) and mapping mountain gazelle optimizer (MMGO), to develop six hybrid predictive models. Among the six models, MHOA-XGBoost (extreme gradient boosting) delivered the highest test accuracy, with a coefficient of determination (R2) of 0.9917 and a mean absolute error (MAE) of 0.72. Among all the input variables, SHAP analysis identified injected volume as the most influential factor. Compared with the injection of a single foam, the injection of hybrid foam can recover more oil. Adjusting the injection ratio of the hybrid foam resulted in greater improvements, with the maximum recovery (>38%) observed at a ratio of 20:80 (N2/CO2). In summary, by integrating advanced ML techniques with optimization algorithms, this study has established a reliable and efficient framework for predicting the performance of hybrid foam flooding and optimizing its injection strategies.
Wang et al. (Fri,) studied this question.