Speaker
Description
During waterflooding in reservoirs, complex pore structures and heterogeneous pore-throat distributions lead to the formation of substantial amounts of residual oil at the microscopic scale. Accurately predicting its spatial distribution remains a critical challenge for understanding pore-scale displacement mechanisms and improving oil recovery. Although conventional physical experiments and numerical simulations provide valuable insights into pore-scale processes, they are commonly limited by high experimental costs, intensive computational requirements, and insufficient adaptability to complex three-dimensional pore networks. Recent advances in digital core technology, together with high-resolution CT imaging, enable realistic representation of pore structures and create new opportunities for data-driven approaches. Here, we investigate the application of deep learning to the prediction of microscopic residual oil distribution during digital core-based waterflooding. Three-dimensional CT data are first used to construct digital core models that explicitly capture pore connectivity and pore-throat structural characteristics. Multiple three-dimensional deep learning architectures are then trained to predict the spatial distribution of residual oil under waterflooding conditions. The predictive accuracy, stability, and generalization performance of different models are systematically evaluated in reservoirs with complex pore structures. By quantitatively assessing the ability of deep learning models to characterize the relationships between pore structure and fluid distribution, this study elucidates their applicability and limitations in microscopic residual oil prediction. These results provide insights into the potential and constraints of deep learning-based approaches for investigating pore-scale displacement mechanisms and optimizing enhanced oil recovery strategies.
| Country | China |
|---|---|
| Acceptance of the Terms & Conditions | Click here to agree |
