Reliable estimation via hybrid gradient boosting machine for mud loss volume in drilling operations

Area of Science:

• Petroleum Engineering
• Data Science
• Machine Learning

Background:

• Drilling mud loss significantly increases operational costs and risks in the oil and gas industry.
• Accurate prediction of mud loss volume is crucial for improving drilling efficiency and minimizing non-productive time.

Purpose of the Study:

• To develop a reliable predictive model for drilling mud loss volume using machine learning techniques.
• To enhance drilling efficiency and reduce non-productive time by accurately forecasting mud loss.

Main Methods:

• Utilized a dataset of 949 field records from Middle Eastern drilling sites.
• Applied data preprocessing including statistical evaluation, outlier detection, and normalization.
• Employed a Gradient Boosting Machine (GBM) with hyperparameter tuning via Evolution Strategies (ES), Batch Bayesian Optimization (BBO), Bayesian Probability Improvement (BBI), and Gaussian Process Optimization (GPO).

Main Results:

• The GBM-BPI model achieved the highest test performance with R² = 0.926, MSE = 1208.77, and AARE% = 26.73.
• SHAP analysis identified hole size, formation type, and pressure differential as key predictors of mud loss.
• Solid content was found to have a minimal impact on mud loss volume.

Conclusions:

• The GBM-BPI model provides an accurate and stable prediction of drilling mud loss.
• Understanding key influencing factors like hole size and pressure differential can aid in mitigating mud loss during drilling operations.