Scholarlink

<p>Permeability is a fundamental property governing fluid flow in subsurface formations and plays a crucial role in reservoir characterization, hydrocarbon recovery, and production planning. Direct measurement of permeability, however, is difficult, costly, and limited in availability particularly in gas hydrate–bearing sediments making reliable prediction a significant challenge. This study develops a machine-learning-based workflow for permeability prediction in area B of the Krishna–Godavari (KG) basin. The workflow was developed using well-log data from the geologically distinct wells NGHP-02–19&#xa0;A and NGHP-02–20&#xa0;A. The data were subjected to rigorous preprocessing before model training. This included outlier detection using the Isolation Forest algorithm and embedded feature selection. Four ensemble regression models namely Decision Tree Regressor (DTR), Random Forest Regressor (RFR), Bagging Regressor (BR), and Extra Trees Regressor (ETR) were trained subsequently. Equivalent Circulation Density (ECD) and depth (mbsf) were identified as the most influential features. Among the four tested models, RFR and BR achieved the highest performance (R² = 0.88), with RFR selected for permeability prediction in ten blind wells. The predicted permeability in wells NGHP-02-16, NGHP-02-17, NGHP-02-22, and NGHP-02-23 shows good agreement with pressure-core-derived permeability. Additionally, in-situ permeability measured using the Modular Dynamic Formation Tester (MDT) tool in NGHP-02-23 also shows good agreement with model predictions. These results demonstrate the potential of the RFR-based workflow for permeability prediction in hydrate-bearing sediments in area B of KG basin. However, because the training data are restricted to two wells, the proposed framework should be regarded as localized, and its broader regional generalizability requires further validation before wider implementation.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Permeability prediction through machine learning regression in gas hydrate bearing sediments of Krishna-Godavari basin, using NGHP-02 well log data

  • 期刊论文

  • 发表日期：

    2026年7月7日

查看全文
  • Pradeep Kumar Yadav,
  • Uma Shankar,
  • Girish Chandra Rahi

摘要

Permeability is a fundamental property governing fluid flow in subsurface formations and plays a crucial role in reservoir characterization, hydrocarbon recovery, and production planning. Direct measurement of permeability, however, is difficult, costly, and limited in availability particularly in gas hydrate–bearing sediments making reliable prediction a significant challenge. This study develops a machine-learning-based workflow for permeability prediction in area B of the Krishna–Godavari (KG) basin. The workflow was developed using well-log data from the geologically distinct wells NGHP-02–19 A and NGHP-02–20 A. The data were subjected to rigorous preprocessing before model training. This included outlier detection using the Isolation Forest algorithm and embedded feature selection. Four ensemble regression models namely Decision Tree Regressor (DTR), Random Forest Regressor (RFR), Bagging Regressor (BR), and Extra Trees Regressor (ETR) were trained subsequently. Equivalent Circulation Density (ECD) and depth (mbsf) were identified as the most influential features. Among the four tested models, RFR and BR achieved the highest performance (R² = 0.88), with RFR selected for permeability prediction in ten blind wells. The predicted permeability in wells NGHP-02-16, NGHP-02-17, NGHP-02-22, and NGHP-02-23 shows good agreement with pressure-core-derived permeability. Additionally, in-situ permeability measured using the Modular Dynamic Formation Tester (MDT) tool in NGHP-02-23 also shows good agreement with model predictions. These results demonstrate the potential of the RFR-based workflow for permeability prediction in hydrate-bearing sediments in area B of KG basin. However, because the training data are restricted to two wells, the proposed framework should be regarded as localized, and its broader regional generalizability requires further validation before wider implementation.