Scholarlink

<p>Accurate prediction of Total Organic Carbon (TOC) and Brittleness Index (BI) in shale reservoirs is essential for identifying organic-rich and mechanically favorable intervals in unconventional oil and gas exploration. An integrated framework combining a Deep Feedforward Neural Network (DFNN) and the non-dominated sorting genetic algorithm-II (NSGA-II) was developed using multi-type logging data from the Eheye 3 well with a hole depth of 2326&#xa0;m. The framework integrates 3σ outlier replacement, magnetic-positioning-assisted depth correction, min–max normalization, DFNN training with the conjugate gradient method, and NSGA-II-based optimization of feature subsets and network parameters. In the 400–450&#xa0;m interval, the TOC prediction error was controlled within ± 0.02 for representative samples, and the BI relative error reached 0.62%. Compared with the empirical ΔlogR method, the proposed model reduced the TOC mean absolute error from 0.1252 to 0.0093, indicating stronger capability in describing nonlinear coupling among logging responses, organic enrichment, and mechanical properties. The negative Spearman correlation between GR and BI ranged from − 0.25 to − 0.35, confirming geological consistency, while MC Dropout uncertainty analysis produced narrow 95% confidence intervals covering most measured values. These results indicate that DFNN-NSGA-II can improve quantitative shale reservoir evaluation and provide a practical intelligent tool for unconventional shale oil and gas exploration and development in Chinese sedimentary basins.</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

Total organic carbon and brittleness index prediction in shale reservoirs based on DFNN and NSGA-II algorithms

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月30日

查看全文
  • Xu Dong,
  • Yan Li,
  • Jie Wang,
  • Xifang Song,
  • Qiuwen Xu

摘要

Accurate prediction of Total Organic Carbon (TOC) and Brittleness Index (BI) in shale reservoirs is essential for identifying organic-rich and mechanically favorable intervals in unconventional oil and gas exploration. An integrated framework combining a Deep Feedforward Neural Network (DFNN) and the non-dominated sorting genetic algorithm-II (NSGA-II) was developed using multi-type logging data from the Eheye 3 well with a hole depth of 2326 m. The framework integrates 3σ outlier replacement, magnetic-positioning-assisted depth correction, min–max normalization, DFNN training with the conjugate gradient method, and NSGA-II-based optimization of feature subsets and network parameters. In the 400–450 m interval, the TOC prediction error was controlled within ± 0.02 for representative samples, and the BI relative error reached 0.62%. Compared with the empirical ΔlogR method, the proposed model reduced the TOC mean absolute error from 0.1252 to 0.0093, indicating stronger capability in describing nonlinear coupling among logging responses, organic enrichment, and mechanical properties. The negative Spearman correlation between GR and BI ranged from − 0.25 to − 0.35, confirming geological consistency, while MC Dropout uncertainty analysis produced narrow 95% confidence intervals covering most measured values. These results indicate that DFNN-NSGA-II can improve quantitative shale reservoir evaluation and provide a practical intelligent tool for unconventional shale oil and gas exploration and development in Chinese sedimentary basins.