Optimized Gradient Boosting Models for Adaptive Prediction of Uniaxial Compressive Strength in Carbonate Rocks Using...

Area of Science:

• Geomechanics
• Machine Learning
• Drilling Engineering

Background:

• Uniaxial compressive strength (UCS) is vital for subsurface formation assessment.
• Traditional UCS lab tests are accurate but time-consuming and costly.
• Machine learning offers efficient real-time UCS prediction.

Purpose of the Study:

• Investigate Gradient Boosting Machines (GBMs) for UCS prediction.
• Apply lagging techniques for dynamic, real-time data input.
• Evaluate XGBoost's predictive performance against other GBMs.

Main Methods:

• Utilized 2056 drilling data points (ROP, GPM, SPP, RPM, T, WOB).
• Applied lagging techniques using historical drilling depth data as input features.
• Performed hyperparameter optimization for GBM models, including XGBoost.

Main Results:

• XGBoost demonstrated superior accuracy in UCS prediction.
• Optimized GBMs significantly improved prediction performance.
• Lowest error metrics were achieved on test and validation datasets by XGBoost.

Conclusions:

• Real-time UCS prediction using XGBoost with lagging techniques is highly effective.
• This method enhances drilling efficiency and mitigates risks like wellbore instability.
• Significant potential for application in carbonate formations is highlighted.