Application of Artificial Intelligence (AI) for Predicting Fracture Intervals and Fracture Density from Well Log Data

Abstract

Fractured granitoid basement plays a critical role in hydrocarbon production in Vietnam, where fracture interval, fracture density and their characteristics are key parameters for reservoir evaluation, reserve estimation, and field development. However, data obtained from core analysis and borehole image (BHI) logs are usually scarce and expensive to acquire and process, limiting their availability for reservoir characterization. This study develops supervised machine learning (ML) models to predict fracture zones and fracture density using well log data from the SA field, Cuu Long Basin. Input data include DCALI, GR, LLD, NPHI, RHOB, DTC, DTS…, combined with hydrocarbon indicator, BHI-derived fracture labels for model training. LightGBM, an efficient gradient boosting algorithm, was selected for its speed and performance. The model achieved recall values exceeding 90% across training, validation, and testing datasets. Blind tests on well XA-2X confirmed the model’s robustness, with prediction accuracy of 82% respectively, and strong agreement between predicted and BHI-derived fracture densities. Fracture porosity and permeability were also calculated from predicted fracture density and aperture (available from two BHI-analyzed wells), yielding values consistent with fractured  basement reservoir behavior. The resulting ML-based prediction modules can be applied to wells lacking BHI data, supporting real-time fracture identification, reservoir characterization, and reserve estimation in fractured basement reservoirs. This approach significantly reduces dependency on high-cost imaging tools and paves the way for broader artificial intelligence into digital reservoir characterization and management workflows.