Maximizing Conventional Oil Recovery and Carbon Mitigation: An Artificial Intelligent Driven Assessment and Optimization of Carbon Dioxide Enhanced Oil Recovery with Physics-Based Dimensionless Type Curves

Raghda Emera Amirmasoud Kalantari Dahaghi ^*

• University of Kansas, Lawrence, United States

Carbon Dioxide Enhanced Oil Recovery (CO₂-EOR) is a well-established technology that has been deployed for over two decades, primarily to boost oil recovery rates. Recently, however, CO₂-EOR has gained attention as a potential carbon mitigation strategy, given its ability to both enhance oil recovery without requiring extensive new drilling and store CO₂ in subsurface formations. This dual function aligns with net-zero carbon goals, as CO₂ is partly trapped in the reservoir through solubility and hysteresis effects on relative permeability.The performance of CO₂-EOR, in terms of both oil recovery and CO₂ storage potential, depends on numerous factors, including reservoir properties such as porosity, permeability, thickness, fluid composition, and operating conditions like bottom-hole pressure (BHP) and injection rates. Traditional screening for CO₂-EOR candidate reservoirs typically relies on experimental work, simulation studies, and field analogs, all of which require significant time and resources. However, a large dataset exists from prior CO₂-EOR projects, which could enable more efficient screening.To leverage this data and capitalize on recent advancements in artificial intelligence, we developed an integrated methodology to predict CO₂-EOR production profiles rapidly and accurately. Using Artificial Neural Networks (ANN), we trained a proxy model (PM) with over 2,000 simulation cases based on real-world CO₂-EOR projects. The model's novelty lies in its ability to generate dimensionless type curves and their derivatives, which can be matched with production data to estimate average reservoir characteristics at later project stages.Our results demonstrate that the proxy model achieves a high level of accuracy, with a maximum Mean Absolute Error (MAE) of 0.012 and a correlation coefficient of 0.99 between predicted and simulated results across three output variables. Additionally, a sensitivity analysis revealed the significant influence of parameters such as fluid composition, rock-fluid interaction, porosity, permeability, and initial reservoir pressure on CO₂-EOR production profiles. This approach provides a rapid, cost-effective alternative to conventional methods, allowing for quicker and more informed decision-making in CO₂-EOR projects.