A Real-Time Rig Control ROP Optimization Framework Using Machine Learning and Predictive Vibration Modeling

Abstract This paper presents a novel real-time rig control framework for optimizing the rate of penetration (ROP) in horizontal drilling operations by integrating machine learning (ML), transfer learning, and predictive vibration modeling within a constraint-aware optimization loop. A general ROP prediction model is first trained on a large, multi-well dataset using ensemble learning methods, with XGBoost emerging as the best-performing algorithm. To improve prediction accuracy in new wells, a transfer learning strategy fine-tunes the model using real-time and offset well data, ensuring adaptability to local geological conditions. The fine-tuned model is embedded within an optimization algorithm that recommends drilling parameters to maximize ROP while respecting operational constraints. These include rotary speed limits derived from a predictive vibration model that identifies resonance zones to prevent harmful drilling vibrations. Additional constraints ensure parameter feasibility, operational continuity, and adherence to target depth of cut specifications. The framework is validated in both simulated and real-world environments. In simulation, a physics-based drilling model is used to generate ground-truth data for assessing convergence and performance. Results show that the optimization algorithm reliably steers drilling parameters toward the region of maximum ROP. Field validation on real rig data further confirms the robustness and practicality of the proposed approach, demonstrating significant improvements in ROP prediction accuracy and drilling performance. This hybrid data-physics optimization framework enables safer, more efficient drilling by providing actionable, real-time parameter recommendations that account for both data-driven insights and physical system dynamics.