Geomechanical Impacts of Pore Types Identified by Machine Learning in Permian–Triassic Reservoirs of the Persian Gulf

Abstract

Understanding geomechanical behavior is critical in the petroleum industry for optimizing wellbore stability, drilling operations, production, and hydrocarbon recovery. Since pore types controls many reservoir properties, this study investigates their influence on geomechanical characteristics. Petrographic analysis, wireline logs, routine core data, and geomechanical measurements were used. Geomechanical data include Young’s modulus (E), Poisson’s ratio (ϑ), shear modulus (G), bulk modulus (K), Schmidt hammer rebound values, unconfined compressive strength (UCS), cohesion (C), and internal friction angle (Ф). The DataFrame was partitioned by porosity to address its effect on geomechanical properties. Hence, missing porosity values were constructed using the ExtraTreesRegressor model, achieving an R² of 0.82. Due to varying sampling depths across datasets, the radial basis function (RBF) method interpolated geomechanical properties from wireline logs to core depths. The effect of pore type on geomechanical characteristics was determined by the barplot method using machine learning. Consequently, it was shown that the samples containing non-fabric selective pore types have the highest geomechanical features. Conversely, rocks with fabric-selective pore types have the lowest geomechanical properties. The samples with depositional pore type have lower geomechanical properties compared to the rocks with non-fabric selective pore type. However, it displays higher geomechanical characteristics compared to the fabric-selective pore. By combining petrographic and petrophysical data with machine learning techniques, this study presents a practical workflow for assessing geomechanical heterogeneity. The approach enhances reservoir characterization and supports more informed development strategies in carbonate fields.