Título: Integration of Laboratory Data and Geophysical Logs for Permeability Prediction in the Barra Velha Formation, Santos Basin
Autores: Filipe de Moura Antônio Cordeiro, Alexsandro Guerra Cerqueira
Resumo: Understanding rock permeability – a measure of how easily fluids like oil and gas can flow through underground formations – is crucial for efficient hydrocarbon recovery. This is particularly challenging in heterogeneous carbonate reservoirs, such as those in the Barra Velha Formation in the Santos Basin. In this sense, this study proposes a robust machine learning-based approach to estimate rock permeability by integrating conventional well log data (geophysical measurements taken down a borehole) and laboratory core measurements (direct analysis of rock samples). Four regression algorithms were employed and compared: K-Nearest Neighbors (KNN), Random Forest, XGBoost, and Support Vector Machines (SVM). The methodology included data preprocessing, with an emphasis on recovering missing well log data using XGBoost, followed by rigorous training and validation of the models using techniques such as holdout cross-validation. The results demonstrate that tree-based algorithms (Random Forest and XGBoost) provided the most accurate predictions, with good generalization in blind tests, a significant advancement considering the complexity of carbonate permeability. This research contributes to the enhancement of complex reservoir characterization, offering an efficient and economically feasible alternative to traditional permeability assessment methods.
Palavras-chave: Permeability prediction; Carbonate Rocks; Pre-Salt; Santos Basin; Barra Velha Formation.
Páginas: 8
Código DOI: 10.21528/CBIC2025-1191656
Artigo em PDF: CBIC_2025_paper1191656.pdf
Arquivo BibTeX:
CBIC_2025_1191656.bib