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

Uniaxial compressive strength (UCS) is a critical geo-mechanical property used to assess the mechanical properties of subsurface formations. While the traditional laboratory tests for UCS estimation are accurate, they are time-consuming and costly. The advancements in machine learning offer a more efficient option for UCS prediction using real-time data. This work investigates the predictive ability of three types of Gradient Boosting Machines (GBMs): Standard Gradient Boosting, Stochastic Gradient Boosting, and eXtreme Gradient Boosting (XGBoost) for UCS prediction. Unlike conventional machine learning approaches, which depend on static model inputs, lagging techniques were applied where drilling depth data from earlier depths were used as input features, allowing for dynamic model changes and enhanced prediction accuracy as new data is acquired in real time. The data set included 2056 drilling data points, comprising rate of penetration (ROP), mud pumping rate (GPM), standpipe pressure (SPP), rotary speed (RPM), torque (T), and weight on bit (WOB), with an unseen validation data set of 870 points. Hyperparameter optimization significantly improved the prediction performance with XGBoost achieving superior accuracy, shown by the lowest error metrics across the test and validation data set. This technique offers significant potential for improving real-time UCS predictions in carbonate formations, enhancing drilling efficiency while reducing risks such as wellbore instability.