Abstract

The global energy transition demands innovative, data-driven methods in the evaluation of subsurface formations for the sustainable uses of carbon dioxide (CO₂) storage and hydraulic fracturing. This study presents a machine learning based geomechanical methodology that ties together standard well log analysis with artificial intelligence to evaluate reservoir intervals for energy transition purposes. Using well log data from two wells in the Cambay Basin, Random Forest model was trained on GR, NPHI, RHOB, and DT logs to predict key geomechanical properties including Young’s Modulus (E), Unconfined Compressive Strength (UCS), Brittleness Index (BI), and Poisson’s Ratio (PR). The models trained exceptionally well with coefficients of determination (R²) of over 0.99. When applied to the second well, predictions were extremely good with R² approaching 0.98 for E, UCS, and BI. Prediction of Poisson's Ratio was not possible due to very low variability and nearly constant training data values that did not allow good model generalization. A combined classification scheme was developed by utilizing both petrophysical (PHIT, PHIE) and geomechanical (E ≤ 30 GPa, UCS ≤ 35 MPa, BI ≤ 6.5 on the 0-10 BI scale) cut-offs to define intervals for CO₂ storage. Unconventional, fracturable zones were classified with an inverse set of criteria. The analysis revealed that only 0.14 percent (approximately 0.61 meters) of the 440-meter reservoir interval in Well-2 is suitable for CO₂ injection. Although the present research uses only two wells, the methodology as shown is versatile and can be used for other basins and reservoirs for extensive geomechanical analysis.