Artificial Intelligence-Based Prediction of Flow Zone Indicator Using Mud Logging Data

Abstract

Reservoir characterization is a crucial step in evaluating petroleum reservoirs, especially in heterogeneous formations that are difficult to accurately represent using limited core data. This work aims to predict the real-time Flow Zone Indicator (FZI) using mud logging data, specifically drilling data (DD). The study employed supervised machine learning models, Random Forest, Gradient Boosting, Gradient Boosting Regressor, SVM, and, to predict FZI values. SHAP analysis was also used to identify the most significant variables affecting the model predictions. Using key statistical metrics including the coefficient of determination (R²) and root mean square error (RMSE), the results demonstrated that the Random Forest and Gradient Boosting gave the best predictive performance, while SVM showed weaker results. The best performance was obtained with the reduced input combination in RF Scenario 4 with an R² of 0.809884, and RMSE of 3.789239. SHAP analysis identified ROP, RPM, Q, and Torque as the most influential variables for FZI prediction. These findings confirm the effectiveness of integrating real-time drilling data with machine learning algorithms to provide a reliable and rapid alternative for estimating reservoir quality, thereby reducing reliance on costly laboratory data and supporting better operational decision-making during drilling.