Identification of Porous Systems in Complex Carbonates Using Artificial Intelligence

Abstract Reliable data is fundamental for volumetric hydrocarbon assessments and the development of optimal production scenarios. Understanding and interpreting reservoir behavior through porous system distribution reduces uncertainty in volumetric calculations, particularly for water saturation (Sw). Derived from Archie’s equation, Sw is a key reservoir quality indicator that is heavily influenced by the cementation exponent (m), which reflects pore system distribution and connectivity. This study introduces an innovative Artificial Intelligence (AI)-driven workflow that integrates geological, petrophysical, and production data to enhance the characterization and exploitation of Naturally Fractured Reservoirs (NFRs). By leveraging artificial intelligence, the workflow models heterogeneities, predicts reservoir performance, and reduces uncertainty, supporting decision-making in complex reservoir systems. Training, validation, and testing of the model with data from wells in different basins enabled a detailed probabilistic evaluation of key indicators, including porous system type (fractures and vugs), cementation exponent (m), Pore Aspect Ratio (AR), fluid type, reservoir quality, and classification. In Mexico, 80% of hydrocarbon production originates from NFRs; hence, this workflow builds on state-of-the-art algorithms developed exclusively for NFRs. When applied, it has been observed that the information provided helped to mitigate production decline in mature fields through the identification of remaining hydrocarbon zones and reducing risks in volumetric calculations. The study commenced with rigorous data collection and validation. Thin section analysis utilized advanced color segmentation in the Hue-Saturation Value (HSV) color space and geometric operators to discretize elements like vugs and fractures. Mathematical morphology operations, such as dilation and erosion, refined segmented areas and removed noise. Fractures and vugs were identified and segmented using the Hough Transform to detect linear and elliptical or circular contours, depending on their morphology. These were further refined with polynomial, elliptical, and circular functions, enabling precise geometric representations for detailed analysis. Segmented data were correlated with Mercury Injection Capillary Pressure (MICP) studies and Bayesian analysis of pore aspect ratios (AR) to validate and characterize fractures and vugs. Additionally, dynamic reservoir data were incorporated to align key properties with NFR classification, enhancing the fidelity of dynamic behavior reproduction. The workflow’s reliance on training data (Well-A), validation data, and test data (Wells B and C) ensured a robust machine learning framework. Training data optimized internal parameters for learning complex relationships, validation data refined hyperparameters and mitigated overfitting, and test data independently evaluated the model’s performance. This comprehensive approach provides a reliable, scalable methodology for NFR classification and reservoir management.