Application of Machine Learning for Permeability Prediction from Well Logs in Sandstone Reservoir: A Comparative Study of Model Performance

Abstract Accurate prediction of permeability and porosity in sandstone reservoirs is crucial for effective reservoir management and optimizing oil and gas recovery. This project explores the potential of machine learning algorithms, including K-Nearest Neighbors (KNN), Artificial Neural Networks (ANN), and Random Forest Regression (RFR), for predicting these reservoir properties using well log and core data from the decommissioned Volve field in the North Sea. Data collection and preprocessing steps involved the integration, cleaning, and scaling of diverse well log parameters to prepare the data for robust analysis. Two predictive scenarios were tested: Scenario 1 used a comprehensive set of log data, while Scenario 2 used a limited subset of influential parameters to reduce costs. Each model was trained and evaluated on both the training and testing datasets, with model performance assessed based on key metrics, including the coefficient of determination (R2) and Mean Squared Error (MSE). Results showed that the Random Forest Regression model achieved the highest prediction accuracy, with an R2 of 0.9893 and a low MSE in Scenario 1, indicating its robustness and precision in permeability prediction. While Scenario 2 demonstrated cost-efficiency by reducing input data, it did not match the accuracy of Scenario 1 across all models, highlighting the trade-off between data complexity and prediction accuracy. This study confirms the applicability of machine learning approaches for permeability prediction in sandstone reservoirs, with Random Forest Regression emerging as the most effective algorithm among those tested.