Scholarlink

<p>Shale oil development in China is still at an early stage, with limited sample size and highly nonlinear relationships between production and multiple geological and engineering parameters. Under these conditions, quantifying parameter sensitivity, elucidating their mechanisms of influence on production, and integrating sensitive parameter combinations to achieve accurate sweet spot prediction have become critical issues in shale oil exploration and development. To address this, a Variance-Adaptive Random Forest (VA-RF) method is proposed, in which node-splitting strategies are dynamically adjusted based on variance. This approach balances decision tree correlation and strength in small-sample scenarios, thereby reducing generalization error and improving prediction accuracy. Experimental results show that VA-RF reduces prediction error by 13.8% compared with the traditional random forest model and decreases the average relative error of single-well production prediction by 11.13%. Moreover, VA-RF produces more accurate sweet spot predictions and effectively highlights reservoir heterogeneity. Analysis of 27 geological and engineering parameters indicates that TOC exhibits the highest sensitivity, with geological parameters exerting greater influence than engineering parameters. High TOC combined with larger effective stimulated reservoir volume synergistically enhances production, while higher porosity coupled with medium-to-low Poisson’s ratio is favorable for capacity improvement. Overall, VA-RF provides a reliable basis for shale oil sweet spot prediction and development optimization.</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

Sweet spot parameter optimization and production prediction method of shale oil using Variance-Adaptive Random Forest

  • 期刊论文

  • 发表日期：

    2026年2月17日

查看全文
  • Hu Zhao,
  • Huan Yu,
  • Hong-Wei Yang,
  • Shan-Shan Gai

摘要

Shale oil development in China is still at an early stage, with limited sample size and highly nonlinear relationships between production and multiple geological and engineering parameters. Under these conditions, quantifying parameter sensitivity, elucidating their mechanisms of influence on production, and integrating sensitive parameter combinations to achieve accurate sweet spot prediction have become critical issues in shale oil exploration and development. To address this, a Variance-Adaptive Random Forest (VA-RF) method is proposed, in which node-splitting strategies are dynamically adjusted based on variance. This approach balances decision tree correlation and strength in small-sample scenarios, thereby reducing generalization error and improving prediction accuracy. Experimental results show that VA-RF reduces prediction error by 13.8% compared with the traditional random forest model and decreases the average relative error of single-well production prediction by 11.13%. Moreover, VA-RF produces more accurate sweet spot predictions and effectively highlights reservoir heterogeneity. Analysis of 27 geological and engineering parameters indicates that TOC exhibits the highest sensitivity, with geological parameters exerting greater influence than engineering parameters. High TOC combined with larger effective stimulated reservoir volume synergistically enhances production, while higher porosity coupled with medium-to-low Poisson’s ratio is favorable for capacity improvement. Overall, VA-RF provides a reliable basis for shale oil sweet spot prediction and development optimization.