CO2 Flooding Optimization Using Artificial Neural Networks: Enhancing Oil Recovery and Carbon Sequestration

Abstract The rising global demand for hydrocarbons underscores the need to optimize oil recovery from existing reservoirs. Conventional methods, such as primary and secondary recovery, leave significant volumes of oil unrecovered, while Enhanced Oil Recovery (EOR) techniques—particularly CO2 flooding—offer the dual benefit of increased oil recovery and carbon sequestration. CO2 flooding improves oil displacement efficiency by reducing viscosity, with the Minimum Miscibility Pressure (MMP) being a critical parameter for achieving miscibility. However, accurate MMP determination remains challenging, as traditional experimental methods are costly and often limited in addressing impure CO2 streams. This research addresses these limitations by developing an Artificial Neural Network (ANN) model trained on a comprehensive dataset encompassing both pure and impure CO2 cases. The ANN demonstrated superior predictive performance compared to traditional methods, achieving an R2 of 0.95 during training and 0.94 during testing. Key factors influencing MMP, including CO2 purity, reservoir temperature, and oil composition, were systematically analyzed, revealing that higher temperatures and heavier hydrocarbons increase MMP, while intermediate hydrocarbons reduce it. CMG reservoir simulations further confirmed that injecting CO2 at pressures above the predicted MMP results in the highest oil recovery and CO2 retention, enhancing both EOR efficiency and carbon sequestration. To ensure practical field application, the ANN model was integrated into a mobile application, MMP Predictor, providing real-time, offline MMP predictions with minimal computational overhead. This AI-driven tool improves decision-making, reduces operational costs, and advances CO2-EOR practices by providing a scalable, accurate, and field-ready method for MMP prediction that supports sustainable oil recovery and climate change mitigation efforts.