A Physics-Informed Hybrid Framework with Adaptive Calibration for Accurate Pore Pressure and Fracture Gradient Prediction in Carbonate Reservoirs

Accurate prediction of formation pore pressure and fracture gradient remains a major challenge in offshore carbonate drilling, where lithological heterogeneity and diagenetic alteration often reduce the reliability of empirical correlations. This study introduces a physics-informed hybrid framework that integrates classical models (Eaton, Miller, and Zhang) with a newly developed Adaptive Calibration Layer (ACL) and an Uncertainty Quantification Module (UQM) embedded within a gradient-boosted machine learning architecture. The ACL dynamically learns depth-dependent correction coefficients between empirical model outputs and high-fidelity Modular Formation Dynamics Tester (MDT) and Extended Pressure Test (XPT) measurements, enabling adaptive depth-wise calibration across variable lithofacies. The UQM provides probabilistic envelopes of pressure prediction through Monte Carlo–based uncertainty propagation. Application to six wells from an Iranian offshore carbonate gas field improved prediction accuracy from R² = 0.63 (classical) and R² = 0.88 (calibrated) to R² = 0.96 (hybrid ACL), reducing RMSE from 3.2 MPa to 1.1 MPa (≈65% error reduction) and constraining the 95% confidence interval to ±0.4 MPa. The proposed approach enhances reliability and physical consistency across wells, offering a scalable, real-time methodology for safe mud weight management and wellbore stability optimization in heterogeneous carbonate formations.