Abstract
Fluid saturations are essential parameters in reservoir analysis due to their impact on reserve estimation and petroleum economics. Traditional methods for determining these properties are often limited by high costs, extensive time requirements, or inadequate accuracy. This study aims to develop a robust, machine learning-based prediction model to overcome these challenges, specifically applied to the Gharif Reservoir in North of Sultanate of Oman.
This study employed advanced data analytics and machine learning to develop a predictive model for fluid saturation. Logs from the Gharif Reservoir were pre-processed and analyzed through descriptive statistics. The Extreme Gradient Boosting (XGBoost) regression model served as the primary prediction tool. To refine model inputs, feature selection techniques, including SelectKBest, Recursive Feature Elimination (RFE), Random Forest, and Principal Component Analysis, were applied. A unique ensemble model combining Random Forest with Recursive Feature Elimination and enhanced feature engineering was then developed to further improve prediction accuracy and manage data dimensionality.
The custom ensemble model showed substantial improvements over the standalone XGBoost model, with enhanced accuracy in predicting water saturation. Key findings included a reduction in prediction errors, achieving a Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) of 0.023 for water saturation. Comparatively, the ensemble model's performance outpaced traditional XGBoost by leveraging optimized feature selection, demonstrating the advantages of integrating various machine learning techniques for complex reservoir characterization tasks.
This study presents a pioneering approach to reservoir analysis, offering an accurate, cost-effective solution with immediate application potential in oil field operations. By integrating machine learning, this model enhances reservoir characterization accuracy, which could significantly optimize reserve estimation and improve decision-making in petroleum economics.
