Scholarlink

<p>Relative permeability curves, which represent the relationship between the relative permeability of oil/water phases and water saturation, are fundamental data in oilfield development engineering. They are crucial for determining the reserves and productivity of oil and gas reservoirs, predict reservoir pressure, evaluate development plans, and identify optimal oil recovery strategies. However, conventional methods for obtaining these curves are often complex and time-consuming, requiring specialized equipment and personnel in the laboratory. The scarcity of relative permeability curve samples in oilfields often leads to inadequate characterization of seepage relationships within the development area, thereby affecting the accuracy of reservoir development modeling. To address the limitations of the above problems, this study develops machine learning models for the rapid prediction of relative permeability curves and can obtain irreducible water saturation and residual oil saturation simultaneously by logging data. Leveraging four ensemble learning algorithms with increasing model complexity—Random Forest, Adaptive Boosting (AdaBoost), Gradient Boosting Decision Tree (GBDT), and XGBoost, the model simultaneously predicts endpoint saturations (Swi, Sor) and the full relative permeability curves. Then we use sensitivity analysis identified four key logs (GR, CAL, R25, and COND) as the most predictive inputs. Subsequently, predictions were made using various logging curves and log segment data corresponding to the core samples. These predictions were then compared against laboratory-measured relative permeability curves. The optimal model (AdaBoost) achieved the highest R<sup>2</sup> reached 0.995 for the oil-phase curve and 0.983 for the water-phase curve, with the corresponding RMSE values as low as 0.021 and 0.015, respectively. This study contributes to the fast and accurate prediction of relative permeability curves and irreducible water saturation and residual oil saturation. It provides a robust foundation for enhancing oilfield exploration and development to increase reserves and production.</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

Prediction Of Relative Permeability Curve Based On Machine Learning Driven By Logging Data

  • 期刊论文

  • 发表日期：

    2025年12月13日

查看全文
  • Jia-ru Zou,
  • Shou-dong Huo,
  • En-liang Liu,
  • Tong-sheng Zeng,
  • Lian-lian Hu,
  • Ya-ying Song,
  • Meng Li

摘要

Relative permeability curves, which represent the relationship between the relative permeability of oil/water phases and water saturation, are fundamental data in oilfield development engineering. They are crucial for determining the reserves and productivity of oil and gas reservoirs, predict reservoir pressure, evaluate development plans, and identify optimal oil recovery strategies. However, conventional methods for obtaining these curves are often complex and time-consuming, requiring specialized equipment and personnel in the laboratory. The scarcity of relative permeability curve samples in oilfields often leads to inadequate characterization of seepage relationships within the development area, thereby affecting the accuracy of reservoir development modeling. To address the limitations of the above problems, this study develops machine learning models for the rapid prediction of relative permeability curves and can obtain irreducible water saturation and residual oil saturation simultaneously by logging data. Leveraging four ensemble learning algorithms with increasing model complexity—Random Forest, Adaptive Boosting (AdaBoost), Gradient Boosting Decision Tree (GBDT), and XGBoost, the model simultaneously predicts endpoint saturations (Swi, Sor) and the full relative permeability curves. Then we use sensitivity analysis identified four key logs (GR, CAL, R25, and COND) as the most predictive inputs. Subsequently, predictions were made using various logging curves and log segment data corresponding to the core samples. These predictions were then compared against laboratory-measured relative permeability curves. The optimal model (AdaBoost) achieved the highest R2 reached 0.995 for the oil-phase curve and 0.983 for the water-phase curve, with the corresponding RMSE values as low as 0.021 and 0.015, respectively. This study contributes to the fast and accurate prediction of relative permeability curves and irreducible water saturation and residual oil saturation. It provides a robust foundation for enhancing oilfield exploration and development to increase reserves and production.