A Robust Mechanistic Model for Pore Pressure Prediction from Petrophysical Logs Aided by Machine Learning in the Gas Hydrate-Bearing Sediments over the Offshore Krishna–Godavari Basin, India

Pore pressure (PP) is the most significant and dynamic parameter in reservoir geomechanics, and it optimizes well drilling in the hydrocarbon industry. Improved error accuracy for PP prediction could reduce drilling risk and hazards, and improve wellbore stability and better casing seat selection. Choosing the appropriate mud weight design for an optimized wellbore drilling is another aspect of PP prediction. Initial estimates of the vertical stress (SV) are made in the petrophysical log (especially sonic, density, and resistivity). We attempted to predict PP using four empirical models: the Eaton, Bower, Miller, and Tau models. The magnitudes of SV and PP ranged 25.87–32.72 MPa and 25.31–31.82 MPa, respectively, in the depth interval of 2548.12–2980.02 m, respectively, for linked wells at site National Gas Hydrate Program (NGHP) Expedition-02. In contrast, logging while drilling (LWD) derived actual and predicted pressures were validated with coefficients of determination, R2, varying from 0.995 to 0.998, which were used to evaluate the most precise PP prediction. Further, robust machine learning (ML) techniques, namely artificial neural networks (ANN), decision trees (DT), and support vector regression (SVR), were employed for the prediction of PP using petrophysical log datasets. As a result, numerous datasets were collected from selected wells and applied for model training, testing, and validation. The DT (best-suited) techniques produced the most accurate prediction for PP, with R2 of 0.998. No overpressure generation, whereas normal pressure was monitored in the gas hydrate zone, and slightly higher pressure was experienced in the free gas zone.