Low porosity and ultra-low permeability are common characteristics of shale reservoirs. Traditional imbibition theory is unable to adequately describe fluid transport behavior in nanopores or capture microscopic mechanisms. In this study, imbibition efficiency was defined as the proportion of oil molecules displaced outside the initial oil phase region relative to the initial oil quantity. This study investigates shale oil spontaneous imbibition mechanisms by integrating molecular dynamics (MD) simulations with machine learning (ML) approaches. MD simulations were performed under baseline conditions of 353 K and 10 MPa, with additional simulations at temperatures ranging from 323 to 393 K, across quartz, calcite and dolomite, and at surfactant concentrations of 0.1% to 0.4% to analyze the influencing factors. Wettability differences among minerals were assessed indirectly through analysis of water density distributions, hydrogen bonding, and water–surface interaction energies, which consistently indicated that dolomite exhibits the strongest hydrophilic character, followed by calcite, with quartz showing the weakest water affinity. Results show that increased temperature, enhanced mineral hydrophilicity, and an optimal surfactant concentration of 0.3% significantly improve imbibition efficiency. Using four algorithms—Support Vector Regression trained, Gradient Boosting Regression Tree, XGBoost, and Random Forest—on the 36 MD-derived datasets, we built an ML model as a proof of concept. The Random Forest model performed the best after cross-validation and hyperparameter adjustment, with a validation R2 of 0.81. The novelty of this study therefore is a proof of concept demonstrating the feasibility of MD with ML integration for imbibition prediction, while clearly identifying limitations and directions for future improvement. This provides theoretical foundations for optimizing shale reservoir development and field-scale recovery enhancement.
Yun et al. (Wed,) studied this question.