Development of a data-driven screening model for miscible carbon dioxide flooding

Carbon dioxide enhanced oil recovery methods (CO2-EOR) offer both enhanced oil recovery and carbon sequestration benefits, however their profitability depends on reservoir properties and CO2 injection design which necessitates a robust screening tool to evaluate its feasibility. This study presents a novel data-driven screening tool to automate the screening process for miscible CO2 flooding. A compositional numerical simulator (CMG GEM) was used to develop a database modelling heavy, black and volatile oils to understand the impact of oil gravity on miscibility. Artificial neural networks (ANNs) were trained using this database to predict EOR performance under various conditions. Our approach addresses the limitations of traditional EOR screening criteria, which often rely on oversimplification and expert judgement. We employed a rigorous validation process, including a univariate sensitivity analysis and comparison with CMG GEM simulation results, to ensure the tool's reliability. The ANN showed strong predictive capability, with performance indicators indicating high accuracy. This tool not only facilitates rapid and accurate CO2-EOR screening but also enhances decision making by integrating a wide range of reservoir rock and fluid characteristics. This study presents a significant advancement in the automation of EOR screening processes, providing a user-friendly and reliable solution for the oil and gas industry. [Received: 21 May 2024; Accepted: 27 July 2024]