Rapid Screening and Optimization of CO2 Enhanced Oil Recovery Process in Unconventional Reservoirs Using Interpretable Random Forests

Abstract Rapid screening and optimization of CO2 enhanced oil recovery (EOR) process in unconventional reservoirs is urgently needed but challenging due to the complexity of this process and poor understanding of seepage in multiscale porous media. To address this issue, an interpretable model with high interpretability was developed to evaluate the response of CO2-EOR processes using random forest (RF) to maximize performance. The feature importance and interaction analyses were conducted to enhance the interpretability of RF model from global to local levels. The genetic algorithm (GA) was used to optimize the CO2-EOR response, and select operational parameters with a minimum CO2 injection and maximum incremental oil recovery. To validate the reliability of the method, the results of the GA-RF model were compared with numerical simulation results under the same conditions, yielding an average relative error of 5.3%. The feature importance analyses show that geological parameters such as Φ and SO have a greater impact on the oil recovery compared to operational parameters. This work represents a pioneering advancement in the rapid optimization of the CO2-EOR process in unconventional reservoirs. The developed workflow offers an efficient process for optimizing the most suitable CO2-EOR process for various types of reservoirs and provides valuable time and resource savings in the optimization.