Enhancing Rate of Penetration Predictions in the Surat Basin Using Supervised Machine Learning: A Comparative Study of ANN and XGBoost Models

Abstract The Surat Basin (the Basin) in Australia is rich in coal seam gas resources, characterized by up to 2500 meters of relatively continuous sediment deposition. However, the Basin presents significant geological challenges, including extended drilling cycles and rising costs, necessitating precise rate-of-penetration (ROP) predictions. Current models often rely on engineers' experience and exhibit considerable errors (typically between 25%-50%) and unstable performance. This highlights the need for refined approaches to improve ROP prediction accuracy. This study employs supervised machine learning methods, specifically Artificial Neural Networks (ANN) and Extreme Gradient Boosting (XGBoost), to predict ROP. Data were collected and preprocessed from five wells in the Surat Basin, divided into three training/validation wells and two testing wells. The data underwent detailed cleaning and processing, including selecting long drilling periods and eliminating low rotations per minute (RPM) points during frequent drilling pipe extensions. For model training, hyperparameter tuning and cross-validation were employed to optimize the models. Evaluation metrics included Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and the Coefficient of Determination (R²). Both XGBoost and ANN models delivered good results, with XGBoost generally outperforming ANN, particularly after data filtering. The two testing wells (i.e., W2 and W6) showed enhanced performance through data filtering, which removed noise and irrelevant data points. For well W2, the results were R² of 0.59, MAE of 10.2 m/h, RMSE of 13.5 m/h, and MAPE of 14.3%. For well W6, the results were R² of 0.55, MAE of 11.7 m/h, RMSE of 15.5 m/h, and MAPE of 15.8%. The predicted ROP curves closely matched the actual drilling ROP curves. The application of ANN and XGBoost for ROP prediction demonstrates significant improvements in accuracy and predictability, with XGBoost showing superior performance. This study provides a robust and reliable tool for optimizing drilling operations in complex geological settings, contributing to more efficient coal seam gas extraction.