A Note on Predicting Rate of Penetration Using Machine Learning Models

Abstract

Rate of Penetration (ROP) is a critical parameter influencing drilling efficiency and accelerating field development. Although conventional ROP models (e.g., Bourgoyne, Hareland) provided reasonable results, they often struggle with limited accuracy and adaptability across different well conditions. Recent advances in hybrid machine learning (ML)-physics or deep ML models improve ROP prediction; however, these methods typically require complex programming, limiting their practical adoption. This study addresses these gaps by introducing a prompt, simple, and strong ROP prediction for directional wells, eliminating the need for hybrid modeling through the platform Dataiku Data Science Studio (DSS). To evaluate the impact of domain-specific parameters, two calculated metrics, D-exponent and Mechanical Specific Energy (MSE) were integrated into the dataset. Three ML algorithms (Gradient Boosted Trees, XGBoost, and Support Vector Machines (SVM)) were trained and tested using R², Mean Absolute Error, and Root Mean Square Error (RMSE) across three directional offshore wells from the same field. XGBoost showed best performance and significant improvement R² scores for all wells after incorporating MSE and D-exponent: from 0.373 to 0.974 (Well-1), 0.216 to 0.945 (Well-2), and 0.862 to 0.983 (Well-3). Features importance and SHAP values analyses further quantified the contributions of MSE and D-exponent to all models’ accuracy, demonstrating their role in enhancing predictions. This work provides a practical, programming-free solution for ROP optimization in directional drilling, achieving high performance without using advanced ML technologies.