Avaliação de Redes Neurais na Caracterização de Reservatórios com Dados De RMN: Predição de Parâmetros FZI, RQI e HFU, e Classificação de Tipos Poros

Evaluation of Neural Network in Reservoir Characterization with NMR Data: Prediction of Permeability, FZI, RQI and HFU Parameters and Classification of Porosity Types

Abstract

Machine learning has advanced scientific research by enabling the analysis of large and complex datasets, including uncorrelated variables. In reservoir characterization, petrophysical parameters such as porosity and permeability are essential for calculating indicators like Flow Zone Indicator (FZI), Reservoir Quality Index (RQI), and Hydraulic Flow Units (HFU), and that can support pore type classification. Nuclear Magnetic Resonance (NMR) is a powerful technique in this context, as it enables direct porosity measurements and permeability estimation through models. In this study, porosity and gas permeability data from 506 carbonate samples were used to evaluate four semi-empirical models (SDR, Timur-Coates, Rios, and Han) and a deep learning model, the Multi-Layer Perceptron (MLP). The MLP outperformed the semi-empirical models, achieving an R² of 0.79 and σ = 3.07 for training and 0.71 and σ = 3.92 for testing. It also effectively differentiated HFUs and closely matched laboratory results. In pore type classification, the MLP model showed superior performance. These results highlight the potential of integrating NMR data with deep learning to improve HFU, FZI, and RQI predictions and support more accurate pore type characterization.