Predicting Oil Recovery Coefficient With Machine Learning Methods on Data Generated From the Buckley-Leverett Model

The pursuit of enhancing oil extraction processes is now embracing technological advancements, particularly in machine learning (ML), to address the inefficiencies and environmental challenges inherent in conventional methods. This research explores the use of ML algorithms for accurately predicting the Oil Recovery Coefficient (COR), an essential metric for evaluating the effectiveness of oil extraction strategies. Leveraging synthetic data generated through the Buckley-Leverett model, which incorporates variables such as viscosity, porosity, time, and production rate, this study rigorously assesses five ML models: Linear Regressors, Random Forest Regressors, Extreme Gradient Boosting Regressors, Support Vector Machines, and Artificial Neural Networks. Our findings demonstrate that Artificial Neural Networks and Random Forest Regressors notably excel, with R-squared values reaching 0.99981 and 0.99821, respectively, on test datasets. These achievements highlight the superior predictive precision and dependability of these models in estimating COR, marking a significant advancement over existing prediction techniques. In conclusion, this investigation not only evidences the transformative potential of machine learning in refining oil recovery forecasts.