What question did this study set out to answer?

The study aims to improve standpipe pressure prediction using machine learning techniques for better monitoring during offshore drilling operations.

April 24, 2026Journal of Engineering and Applied Science0 citationsOpen Access

Machine learning-based prediction of standpipe pressure using operational parameters from offshore directional gas wells in the Nile Delta

AHA.A. Husseiny AAAttia M. Attia AHAhmed G. Hagag

Key Points

The study aims to improve standpipe pressure prediction using machine learning techniques for better monitoring during offshore drilling operations.
Utilized in-memory machine learning models in Dataiku Data Science Studio for SPP predictions.
Adapted four ML algorithms: LightGBM, Extra Trees, XGBoost, and Random Forest.
Evaluated model performance using R2 scores and RMSE on data from four offshore wells.
All models achieved R2 scores exceeding 0.968, indicating strong predictive performance.
The in-memory ML models provided accurate predictions under complex drilling conditions.
Feature importance analysis revealed insights into selected features' contributions to predictions.

Abstract

The Standpipe Pressure (SPP) is one of the most important drilling parameters as it reflects to wellbore stability and loss circulation. Monitoring SPP values are still challenging due to complex and uncertain downhole changes while drilling. Machine Learning (ML) algorithms are most promising techniques for forecasting SPP to optimize monitoring its values under downhole conditions of pressure fluctuations, fluids gains/losses. Hybrid technologies are used to combine advanced ML models and physical-based models to prove their effectiveness for predicting SPP. However, these technologies require both more surface and subsurface data with advanced programming languages for handling with complex data inputs to perform this task. This study aimed to conduct prompt SPP prediction leveraging Dataiku Data Science Studio (DSS) using in-memory ML models without using complex approaches with respect to reaching precise results. Four ML models were adapted successfully in this study for SPP prediction: LightGBM, Extra Trees, XGBoost, and Random Forest. Four offshore wells were being drilled into the West Delta Deep Marine Egypt, provided total 5921 datapoints to predict SPP. Models performance were being evaluated through metrics of R2 scores and RMSE. All models achieved very strong performance for both training and testing dataset with R2 exceed 0.968. The results confirmed applicability of in-memory ML models for delivering accurate SPP predictions in a complex deepwater drilling. The developed methodology presented a promising template for similar applications, though local retraining on new data would be required for other fields. Features importance analysis and SHAP values were employed in this study for reaching deeper insights of the selected features and their numerical values for enhancing prediction reliability.

KI fragen

Bookmark

View Full Paper