A proxy-based workflow for screening and optimizing cyclic CO2 injection in shale reservoirs

Abstract

Optimizing cyclic CO₂ injection (CO₂ HnP) in shale reservoirs is challenging due to the numerous variables in the system, which exhibit complex coupling effects on the final hydrocarbon recovery. A workflow combining a multicomponent species transport model and a proxy model is proposed to identify suitable target blocks and optimize CO₂ HnP operational parameters for maximizing cumulative oil production. A single-well CO₂ HnP compositional simulation is developed based on a multiphase, multicomponent species transport model. This model accounts for key transport mechanisms such as viscous flow, molecular diffusion, and Knudsen diffusion in shale reservoirs. Least-squares support vector machine (LS-SVM) is used as a proxy for the simulation model to reduce computational costs in subsequent optimization processes. The optimal combination of operational parameters, as well as reservoir rock and fluid properties, is investigated to maximize oil recovery. Finally, the LS-SVM proxy model is integrated with a genetic algorithm to perform robust optimization. The Results and Discussion section presents three optimization scenarios derived from baseline parameters of the Eagle Ford shale reservoir, progressively incorporating more variables. The LS-SVM proxy model demonstrates its high predictive accuracy with a small training dataset, outperforming three alternative approaches: Long Short-Term Memory (LSTM), Genetic Algorithm-optimized Back Propagation (GA-BP) neural networks, and Extreme Gradient Boosting (XGBoost). A thorough optimization process is crucial to achieve higher oil recovery, potentially increasing CO₂ HnP recovery from 11.64 % to 19.13 % through the design of operational parameters. The findings also indicate that a larger volume of injected CO₂ leads to greater enhanced oil recovery by enabling deeper penetration into the reservoir and more effective mixing with crude oil. Furthermore, deep reservoirs containing low gas–oil ratio black oil are especially favorable for cyclic CO₂ HnP, as the injected CO₂ substantially enhances oil swelling and improves production potential.