Deep Learning Driven Reservoir Simulation for Mapping Performance of CO2-EOR and Storage in Tight Oil Reservoir

Abstract The increasing need to address climate change and ensure energy security has emphasized the importance of carbon capture, utilization, and storage (CCUS) technologies, particularly CO2-enhanced oil recovery (CO2-EOR), for achieving net-zero emissions. While CO2-EOR has demonstrated success in conventional reservoirs, its application in unconventional formations, such as the Bakken shale, presents challenges due to ultra-low permeability, complex fracture networks, and geological heterogeneity. Traditional numerical simulations are computationally intensive and often fail to adequately capture the complex dynamics of unconventional reservoirs. This study introduces a novel workflow that integrates compositional reservoir simulations with advanced deep learning (DL) architectures to predict CO2-EOR performance and storage potential in the Bakken formation. Key reservoir parameters such as porosity, permeability, and fluid viscosities are transformed into multi-dimensional input features to train and evaluate three DL architectures: Fully Connected Neural Networks (FCNN), Residual Neural Networks (ResNet), and Densely Connected Networks (DenseNet). These models predict oil saturation, CO2 saturation, and pressure distributions with high accuracy. ResNet consistently outperformed the other models, achieving the lowest Mean Absolute Error (MAE) values of 0.0115, 0.008, and 0.1471, and the highest coefficient of determination (R2) values of 0.9204, 0.9406, and 0.9969 for oil saturation, CO2 saturation, and pressure, respectively. The integration of compositional simulations with DL models addresses computational and geological uncertainties, providing a scalable framework for optimizing CO2 injection strategies, enhancing hydrocarbon recovery, and maximizing CO2 storage in unconventional reservoirs. This research contributes to advancing CCUS technologies and improving the economic and environmental outcomes of subsurface energy management.