Scholarlink

<p>Accurately prediction of recovery factor(RF) is essential for optimizing reservoir development, especially with the sustainability goal of maximizing fossil fuel efficiency. Conventional methods, such as analogies and numerical simulations, face challenges in dealing with complex and uncertain reservoir conditions. In this study, the conversion ratios of horizontal and vertical wells were calculated using digital twin technology and the study sample was supplemented by virtual development to enhance RF prediction. The study created a dataset of 128 actual and 190 virtual reservoirs using Latin Hypercube Sampling to fully enhance the data. Of the 11 evaluated machine learning models, AdaBoost performed the best (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\hbox {R}^{2}\)</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mtext>R</mtext> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation> = 0.9482, MAPE = 1.76%). Feature importance analysis using SHAP identified key factors such as oil saturation and well pattern density, providing transparency for sustainable decision-making in reservoir management. In addition, a user-friendly web application was developed for real-time forecasting, leading to more efficient and convenient resource management.</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

Predicting Recovery Factor with Digital Twins and Interpretable Machine Learning: A Case Study of South China Sea Reservoirs

  • 期刊论文

  • 发表日期：

    2025年11月17日

查看全文
  • Yandong Hu,
  • Zhijie Wei,
  • Yunhong Xie,
  • Yun Liu,
  • Xiankang Xin,
  • Gaoming Yu

摘要

Accurately prediction of recovery factor(RF) is essential for optimizing reservoir development, especially with the sustainability goal of maximizing fossil fuel efficiency. Conventional methods, such as analogies and numerical simulations, face challenges in dealing with complex and uncertain reservoir conditions. In this study, the conversion ratios of horizontal and vertical wells were calculated using digital twin technology and the study sample was supplemented by virtual development to enhance RF prediction. The study created a dataset of 128 actual and 190 virtual reservoirs using Latin Hypercube Sampling to fully enhance the data. Of the 11 evaluated machine learning models, AdaBoost performed the best ( \(\hbox {R}^{2}\) R 2 = 0.9482, MAPE = 1.76%). Feature importance analysis using SHAP identified key factors such as oil saturation and well pattern density, providing transparency for sustainable decision-making in reservoir management. In addition, a user-friendly web application was developed for real-time forecasting, leading to more efficient and convenient resource management.