Related Experiment Video
Updated: Jun 25, 2025
12:18
Pore-scale Imaging and Characterization of Hydrocarbon Reservoir Rock Wettability at Subsurface Conditions Using X-ray Microtomography
Published on: October 21, 2018
13.9K
Machine learning-based classification of petrofacies in fine laminated limestones
Gallileu Genesis1, Igor F Gomes1, José Antonio Barbosa2
1Universidade Federal de Pernambuco, Departamento de Engenharia Civil, Laboratório de Métodos Computacionais em Geomecânica, Av. da Arquitetura, s/n, Cidade Universitária, 50740-540 Recife, PE, Brazil.
Anais Da Academia Brasileira De Ciencias
|May 22, 2024
Summary
Machine learning models were tested for classifying petrofacies in thin limestone layers. Support vector machine (SVM) showed better results than Gaussian naive Bayes (GNB) for detailed classification.
Area of Science:
- Geosciences and Petroleum Engineering
- Machine Learning Applications in Geology
Background:
- Accurate lithological classification of hydrocarbon reservoirs is crucial for characterization and development.
- Thin reservoir units with limited vertical data pose challenges for traditional lithofacies classification.
- High-frequency vertical variations in diagenetic heterogeneities complicate petrofacies identification.
Purpose of the Study:
- To evaluate machine learning models for petrofacies classification in thin sedimentary intervals with high-frequency variations.
- To assess the performance of Gaussian Naïve Bayes (GNB) and Support Vector Machine (SVM) in an extreme classification scenario.
- To analyze the impact of mm- to cm-scale heterogeneities on automatic lithofacies identification.
Main Methods:
- Utilized pseudo-well data generated from outcrop measurements (radiometric and unconfined compressive strength logs).
- Applied Gaussian Naïve Bayes (GNB) and Support Vector Machine (SVM) machine learning techniques.
- Classified eight petrofacies types, grouped into two main categories, within a thin limestone interval.
Main Results:
- The Support Vector Machine (SVM) achieved the best overall petrofacies classification performance with an F1 score of 0.47.
- Gaussian Naïve Bayes (GNB) showed lower effectiveness for detailed classification (F1 score of 0.29) but improved when distinguishing the two main petrofacies groups.
- High-frequency facies variations, driven by small-scale heterogeneities, present a significant challenge for automatic lithofacies identification.
Conclusions:
- Machine learning models, particularly SVM, can classify petrofacies in challenging thin reservoir intervals, albeit with limitations.
- The study highlights the difficulty in automatic lithofacies identification due to complex depositional and diagenetic processes at the mm- to cm-scale.
- Understanding these small-scale heterogeneities is vital for accurately predicting fluid flow in porous media within hydrocarbon reservoirs.
