Machine Learning-Based Electrofacies Classification in Brazilian Pre-Salt Carbonates

This study applies unsupervised machine learning techniques for electrofacies classification in the heterogeneous carbonate reservoirs of the Brazilian Pre-salt. By employing hierarchical clustering and K-means clustering, the research analyzes well-log data from six wells in the Campos Basin, including gamma ray, bulk density, sonic, and NMR data. The hierarchical clustering method identified four electrofacies, with H1 and H2 as non-reservoir facies, H3 as predominantly non-reservoir, and H4 as the main reservoir facies. K-means clustering also identified four electrofacies, with K1 and K2 as non-reservoir facies, K3 as a mixed reservoir facies, and K4 as the best reservoir facies. The results indicate that K-means clustering provided a more refined delineation of reservoir facies, particularly for the highest-quality intervals, compared to hierarchical clustering. Both methods effectively identified lithological variations and delineated reservoir and non-reservoir zones, with K-means showing higher sensitivity to subtle variations in petrophysical properties. The study concludes that unsupervised machine learning methods are effective in classifying electrofacies within pre-salt carbonate reservoirs, providing valuable insights into reservoir characterization and quality.