REAL-TIME MACHINE LEARNING-BASED PREDICTION OF FORMATION PRESSURE: EVALUATING THE IMPACT OF ROP IN DRILLING OPERATIONS

ABSTRACT

Formation pressure is a critical factor in drilling operations, directly influencing well control and operational safety. Traditional estimation methods—relying on well logs and seismic surveys— often fall short in real-time applicability and incur high costs, especially in challenging drilling conditions. This study presents a novel machine learning framework that integrates dynamic drilling parameters, notably the Rate of Penetration (ROP), with static petrophysical data to predict formation pressure in real time. Data were collected from sandstone formations in the Niger Delta, encompassing both drilling parameters (e.g., ROP, mud weight) and well log measurements (e.g., gamma ray, resistivity, neutron porosity). After thorough preprocessing and feature engineering, several models were evaluated, with the Random Forest (RF) model emerging as the most robust, achieving an R² of 0.978—a significant improvement over models excluding ROP and traditional methods like the Eaton equation. The inclusion of ROP was found to be particularly beneficial, as it captures dynamic changes in wellbore conditions that static data cannot reflect, thereby enhancing the predictive accuracy and reducing RMSE values substantially. Feature importance analysis highlighted neutron porosity and ROP as key indicators, validating the model's interpretability and its alignment with domain knowledge. The integration of this model into real-time drilling operations promises enhanced hazard detection, improved wellbore stability, and reduced operational costs. While the current study is limited to the Niger Delta context, the findings lay a strong foundation for future research aimed at testing the model across diverse reservoir types and incorporating additional dynamic parameters for further optimization.