Full text loading...
-
Integrating MRGC and Neural Networks for Rock Typing and Log Permeability Prediction in Fractured Carbonates
- Publisher: European Association of Geoscientists & Engineers
- Source: Conference Proceedings, EAGE/FESM Conference on Petrophysics meets Geoscience: Unlocking Reservoir Potential in a Dynamic Energy Landscape, Nov 2025, Volume 2025, p.1 - 6
Abstract
Summary
This study introduces a machine learning-based workflow for rock typing and log permeability prediction in the ML reservoir, a fractured carbonate system in Iraq. Leveraging newly acquired core data from four wells, particularly enriching the previously under-characterized AVD formation, the research integrates Multi-Resolution Graph-Based Clustering (MRGC) and neural networks to enhance electrofacies classification and permeability estimation. The workflow began with thorough data hygiene, CT scan normalization, and conditioning in AspenTech Geolog software. Lithofacies were initially simplified using RHOB-NPHI plots to address the complexity of 41 core-described lithofacies. A training model was then constructed using selected representative core intervals with reliable log quality.
MRGC and neural networks enabled unsupervised classification of electrofacies. Electrofacies were validated using blind tests and log-core permeability comparisons. Results showed improved facies differentiation such as in AVP formation alone revealed 10 distinct electrofacies across two wells with limited overlap, indicating spatial heterogeneity and improved subsurface understanding.
Permeability estimates from refined electrofacies matched core data and helped exclude poor-quality log intervals affected by badhole conditions. The methodology significantly enhances static model accuracy and demonstrates the value of combining MRGC and neural networks for petrophysical analysis in complex carbonate reservoirs.
© EAGE Publications BV