A Machine Learning-Based Co-Optimization Framework Improves CO2 Sequestration and Oil Recovery

Abstract This work combines CO2-enhanced oil recovery (EOR) methods with subsurface containment strategies to permanently store CO2 while simultaneously increasing cost-effective oil production from reservoirs. This study develops a novel workflow with multi-objective optimization techniques to assess the integration of pressure management methodologies for permanent geological CO2 storage in saline aquifers. We used a homogeneous reservoir model, reflecting a typical configuration found in oil and aquifer settings, complete with a sealing layer between them to predict the effects of CO2 injection for a 10-year simulation. Within this framework, we determined crucial subsurface and decision variables such as heterogeneity, injection/production rates, fluid types, and injection method. Using Latin Hypercube Sampling, we generated 1000 unique simulation data sets to train, validate, and test a neural network, forming a smart proxy for optimization. This process supports our machine learning-driven proxy model, aimed at optimizing CO2 injection strategies to enhance sequestration, oil recovery, and Net Present Value. We fine-tuned operational parameters through a gradient-based optimization, focusing on these critical objectives. Our "Smart Proxy Model" demonstrated remarkable accuracy in predicting optimization objectives, with an R2 (R-squared) value of 0.993 and a relative squared error of 2.08. The results reveal that the Water Alternating Gas cycle, fluid properties, and operational parameters such as CO₂ injection and oil production rates significantly influence NPV and project feasibility without adversely affecting the integrity of the cap rock due to changes in pressure. The optimization results indicate that the moderate fluid type attained the highest Net Present Value (NPV) of $67.51 billion, surpassing the undersaturated case, the light fluid case, and the base case. Simulation results further validate the optimization findings. The pressure and saturation behavior over a 10-year development period confirmed the effectiveness of the optimized strategy, with the moderate fluid exhibiting the most efficient pressure drop and saturation distribution. Our innovative approach to optimizing CO2 injection and storage presents an effective EOR strategy and emphasizes the dual benefit of mitigating greenhouse gas emissions through effective sequestration. Our methodology displays the potential for substantial environmental benefits alongside notable economic gains by optimizing the balance between CO2 storage in aquifers and its use in EOR, highlighting a sustainable path forward in addressing climate change and maximizing resource efficiency.