Machine learning approach for core permeability prediction from well logs in Sandstone Reservoir, Mediterranean Sea, Egypt

Precise reservoir description is crucial for the proper evaluation of reservoirs. The ability to predict permeability is key to successful reservoir characterization. Although obtaining permeability values through a reservoir is crucial, this is not a simple task. This requires a significant amount of time and money. Numerous methods use the core correlations documented in the literature. Well-log information that offers ways to determine permeability. In this study, we describe a machine learning-based permeability prediction method. With log and core (data), Generalized Additive Models (GAMs) allow for the expansion of the forecasts for(uncored)wells. It also considers the characteristics of the field to ensure a better comprehension of reservoirs, reservoir rock characteristics, and geological variations. We demonstrate the utility of generalized additive models (GAMs), a non-parametric regression-based technique, to account for nonlinear trends in seven wells located in the (WDDM) concession where data analysis is performed on the collected information to evaluate the model's performance. Both the linear and nonlinear functions were used to train the data. The findings of the analysis demonstrate that GAMs outperform segmented linear regression models when the trend is nonlinear, but they also demonstrate their effectiveness when the trend is linear. The GAMs used five wells for training and two blind wells for testing. The GAMs with all five wells used as input, were found to perform best in predicting permeability for the shaly sandstone, with coefficients of determination (Pseudo-R2) of approximately 0.98 and 0.82 for the training and blind data sets, respectively.