In the heart of China, researchers at the State Key Laboratory of Oil and Gas Reservoir Geology and Exploitation, Southwest Petroleum University, have developed a groundbreaking approach to enhance oil recovery and optimize CO2 utilization in unconventional reservoirs. Led by Shuqin Wen, the team has harnessed the power of interpretable machine learning to tackle the complexities of CO2 enhanced oil recovery (EOR) operations, potentially revolutionizing the energy sector.
Unconventional reservoirs, such as shale and tight oil formations, have long posed challenges due to their intricate porous structures and the multitude of factors influencing fluid flow. Traditional methods of CO2-EOR, while effective, have struggled to keep pace with the need for rapid screening and optimization. This is where Wen and her team’s innovative framework comes into play.
The researchers employed three different machine learning methods—random forest (RF), support vector regression (SVR), and artificial neural network (ANN)—to create proxy models using data from a specific unconventional reservoir. The random forest model emerged as the top performer, offering a robust foundation for further analysis. But the team didn’t stop at model performance; they went a step further to enhance interpretability.
“Interpretability is crucial for stakeholders to trust and act upon the model’s predictions,” Wen explained. To achieve this, the team proposed multiway feature importance analysis and Shapley Additive Explanations (SHAP) to quantify the contribution of individual features to the model output. This level of transparency is a game-changer, allowing energy companies to understand not just what the model predicts, but why it makes those predictions.
With a clear understanding of feature importance, the team coupled the random forest model with a genetic algorithm (GA) to optimize the CO2-EOR process. The results were impressive, with the GA-RF model yielding a minimum relative error of just 0.34% and an average relative error of 5.3% when compared to simulation results under various reservoir conditions.
The implications for the energy sector are significant. This interpretable machine learning framework enables rapid screening of suitable CO2-EOR strategies based on specific reservoir conditions. It provides a practical tool for field applications, helping energy companies to maximize oil recovery while minimizing costs and environmental impact.
As the world continues to seek sustainable energy solutions, this research offers a promising path forward. By optimizing CO2 utilization and storage, energy companies can not only enhance oil recovery but also contribute to reducing carbon emissions. The work, published in Petroleum, sets a new standard for the industry, paving the way for future developments in CO2-EOR operations in unconventional reservoirs.
The energy sector is on the cusp of a new era, where data-driven decisions and interpretability go hand in hand. With pioneers like Wen and her team leading the charge, the future of CO2-EOR operations looks brighter—and more sustainable—than ever before. As the industry continues to evolve, the insights gained from this research will undoubtedly shape the strategies and technologies of tomorrow.
