ML-Driven Forecast and Optimization of Wettability Alteration and Oil Recovery Using Synthetic and Natural Surfactants for CEOR

Abstract

This thesis examines the use of ML models, specifically XGBoost, SVR, LightGBM, and CatBoost, for forecasting wettability alteration reduction and optimizing Enhanced Oil Recovery (EOR) processes through the application of different surfactants. The study focuses on understanding the impact of surfactant characteristics, environmental factors, and reservoir physical properties on the efficacy of oil recovery methods. The study used several performance metrics, such as R², RMSE, AIC, and BIC, to see how well the models could predict the future. The results show that SVR did the best among others, with the highest value of R² and the lowest RMSE. XGBoost and LightGBM also demonstrated good abilities in prediction. The analysis of permutation feature importance indicated that the surfactant concentration, water type, and rock type all affect the final results. The results indicate that the optimal combination of cationic surfactants, brine water, and carbonate rocks produces the most effective outcomes for oil recovery. Other than that, this research confirms that bio-surfactants could be a good eco-friendly option, but there are still problems with making them work on large-scale projects in the oil and gas industry. The thesis suggests some insights into improving surfactant-based enhanced oil recovery (EOR) methods and advocates for a more sustainable approach in recovering oil, reducing environmental side effects.