Scholarlink

Carbon dioxide (CO2) injection into petroleum reservoirs has been offering the dual benefits of greenhouse gas emission reduction and enhanced oil recovery (EOR). In the context of miscible flooding, CO2 solubility is a critical factor in designing efficient CO2 injection processes. Enhanced CO2 solubility is associated with reduced oil viscosity, increased crude oil swelling, and lower interfacial tension (IFT), ultimately leading to improved oil mobility and higher oil recovery rates. Machine Learning has emerged as an efficient and effective method to predict CO2 solubility recently. In this study, we have assembled a comprehensive dataset for both live oil and dead oil. Linear regression, Decision Tree (DT), Random Forest (RF), Artificial Neural Networks (ANN), Support Vector Regression (SVR), and Extreme Gradient Boosting (XGBoost) are employed to deliver highly accurate predictions of CO2 solubility. Key input parameters in our modeling effort include reservoir temperature, oil saturation pressure, oil molecular weight, and oil specific gravity in a dead oil model. For the live oil model, bubble point pressure is also included. The dataset is divided into training (80%) and testing (20%) subsets. We employed a range of statistical calculations and graphical representations to assess the tool's effectiveness. Our comprehensive error analysis reveals that the XGBoost model demonstrates exceptional accuracy, achieving an R2 value of 0.996 for training data and 0.875 for testing data of live oil and an R2 value of 0.999 for training data and 0.948 for testing data for dead oil systems. In summary, the XGBoost model stands as a valuable and accurate correlative model for this specific prediction, enabling rapid and precise estimation of CO2 solubility in crude oil systems. Our research offers a valuable contribution to the field, addressing the dual challenges of EOR and emission reduction. These insights hold significant value for the industry, fostering improved practices and outcomes in CO2 EOR and greenhouse gas mitigation.

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

Applying Machine Learning Algorithms to Predict CO2 Solubility in Oil During CO2 Flooding

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Xue-jia Du,
  • Shi-hui Gao,
  • Irvin Archer

摘要

Carbon dioxide (CO2) injection into petroleum reservoirs has been offering the dual benefits of greenhouse gas emission reduction and enhanced oil recovery (EOR). In the context of miscible flooding, CO2 solubility is a critical factor in designing efficient CO2 injection processes. Enhanced CO2 solubility is associated with reduced oil viscosity, increased crude oil swelling, and lower interfacial tension (IFT), ultimately leading to improved oil mobility and higher oil recovery rates. Machine Learning has emerged as an efficient and effective method to predict CO2 solubility recently. In this study, we have assembled a comprehensive dataset for both live oil and dead oil. Linear regression, Decision Tree (DT), Random Forest (RF), Artificial Neural Networks (ANN), Support Vector Regression (SVR), and Extreme Gradient Boosting (XGBoost) are employed to deliver highly accurate predictions of CO2 solubility. Key input parameters in our modeling effort include reservoir temperature, oil saturation pressure, oil molecular weight, and oil specific gravity in a dead oil model. For the live oil model, bubble point pressure is also included. The dataset is divided into training (80%) and testing (20%) subsets. We employed a range of statistical calculations and graphical representations to assess the tool's effectiveness. Our comprehensive error analysis reveals that the XGBoost model demonstrates exceptional accuracy, achieving an R2 value of 0.996 for training data and 0.875 for testing data of live oil and an R2 value of 0.999 for training data and 0.948 for testing data for dead oil systems. In summary, the XGBoost model stands as a valuable and accurate correlative model for this specific prediction, enabling rapid and precise estimation of CO2 solubility in crude oil systems. Our research offers a valuable contribution to the field, addressing the dual challenges of EOR and emission reduction. These insights hold significant value for the industry, fostering improved practices and outcomes in CO2 EOR and greenhouse gas mitigation.