Facies Classification and Porosity–Permeability Modeling Using Self-Organizing Maps in Hybrid Carbonate Rocks

Resumo

Detailed studies on the hybrid carbonate rocks of the Morro do Chaves Formation are essential, as these outcrops serve as analogues to the Brazilian pre-salt carbonate reservoirs, allowing for the validation and application of advanced petrophysical characterization methods. However, the heterogeneity resulting from the complex mixture of carbonate and siliciclastic minerals significantly challenges facies classification and the prediction of critical properties such as porosity and permeability. Unsupervised techniques based on Self-Organizing Maps (SOM) have gained prominence in petrophysics for their ability to project complex multidimensional data onto two-dimensional maps, preserving original topology and revealing clusters and nonlinear patterns not perceptible through traditional approaches. In this context, four distinct SOM models were developed in the Interactive Petrophysics (IP) software, and the best-performing model was selected. This spherical-topology network, comprising 1,442 nodes, was trained over 50,000 epochs (η₀=0.1). To train the SOM, we employed well logs, 57 core-plug measurements of porosity and permeability, and a lithologic profile based on core descriptions. Classification was validated using contingency tables, with the best model achieving 72.59% accuracy, and was complemented by qualitative analysis through Fit Flag. Subsequently, petrophysical properties were predicted and statistically evaluated using the mean absolute error (MAE), root-mean-square error (RMSE), and uncertainty bands. These bands correspond to a ±1 σ standard deviation, weighted at each node, and are derived from continuous calibration data, and plotted alongside the predicted curves. Finally, regression fits between predicted and observed values yielded correlation coefficients (R²) of 0.9398 for porosity and 0.8266 for permeability, corroborating the robustness and expanding the applicability of this approach to well-scale petrophysical characterization of hybrid carbonate systems.