Aplicação de aprendizado de máquina para estimativa da permeabilidade utilizando dados de RMN de rochas análogas a reservatório

Abstract

This master's thesis explores the relevance of studies on rocks analogous to reservoirs, with a primary focus on the petrophysical permeability parameter. Traditional petrophysical techniques provide absolute values for this parameter but offer limited insights into the underlying reasons for these values. The absence of precise information necessitates the use of more advanced techniques for a systematic reservoir analysis. While Nuclear Magnetic Resonance proves effective in obtaining advanced reservoir information, its analytical models for permeability lack accuracy, particularly in heterogeneous reservoirs, revealing certain limitations. Filling this gap, this study aims to improve permeability predictions by employing machine learning models such as Random Forest, Gradient Boosting, and Multi-Layer Perceptron. The goal is to predict permeability based on patterns obtained through Nuclear Magnetic Resonance, compare them with analytical models, and identify the model that best suits the reservoir. It was observed that machine learning models provided more accurate predictions compared to analytical models.