Optimization of the Drilling Rate of Penetration (ROP) in Iraqi Kurdistan Fields Using Artificial Neural Networks

Abstract

The primary aim in the oil and gas industry is to optimize drilling operations to reduce costs and time. Accurate Rate of Penetration (ROP) prediction is essential for optimizing operations and directly influences the cost spent and non-productive time on the rig. Multiple mechanical and geological factors complicate the prediction of ROP, and standard empirical models often fail to capture the complex dynamics between the factors that affect it. This study developed machine learning models to optimize ROP in the complex geological formations of the Kurdistan region of Iraq. Using a dataset of 10,887 field-measured points, the research primarily employs Artificial Neural Networks (ANN) optimized via WV-curves and Backpropagation as the core predictive engine. To further enhance accuracy, the framework integrates Random Forest (RF) and Extreme Gradient Boosting (XGBoost) through a comparative optimization analysis. This process includes evaluating model performance on both raw and optimized datasets, revealing that data optimization significantly improves predictive reliability. To ensure a realistic assessment, a well-based data splitting strategy (70% training, 30% testing) was implemented to prevent data leakage. Quantitative results from the testing phase showed that XGBoost achieved the highest performance, marginally outperforming RF. Specifically, the optimized ANN model proved reliable, whereas the conventional Maurer's model failed to provide a reliable estimate of the rate of penetration. By integrating SHAP-based interpretability, this research bridges the gap between machine learning theory and drilling physics. The study contributes novel insights into the impacts of reservoir-specific parameters, offering a validated, transparent tool for real-time ROP optimization that is methodologically superior to standard random-splitting approaches. This study is considered to be one of the first in terms of type and location, and also demonstrates the vital role of machine learning approaches, with an emphasis on ANN, to enhance the petroleum industry by predicting ROP.