Risk-Aware Rate of Penetration Prediction and Optimization based on Machine Learning Algorithm

This thesis presents a comprehensive data-driven framework for Rate of Penetration (ROP) prediction and optimization in drilling operations using machine learning. The approach addresses key challenges in real-world drilling data, including noise, outliers, operational disturbances, and geological heterogeneity. By combining drilling parameters with logging data, the study develops optimized machine-learning models for accurate and robust ROP prediction. It further introduces methods for uncertainty quantification and risk assessment, enabling prediction intervals and risk levels to be considered alongside conventional point predictions. Building on these results, a risk-aware optimization strategy is developed to identify drilling parameter adjustments that can improve ROP while accounting for predictive uncertainty and engineering constraints. The proposed methods are integrated into ROPilot, an intelligent drilling decision-support system that combines data preprocessing, data fusion, ROP prediction, uncertainty analysis, and parameter optimization. The work provides an integrated approach for advancing drilling from conventional ROP prediction toward data-driven, uncertainty-aware, and intelligent decision support.