Novel enhanced oil recovery screening methodologies by implementing improved stacking ensemble learning algorithms

Abstract

Precise and dependable screening for Enhanced Oil Recovery (EOR) is essential for the optimal planning and design of EOR projects. The adoption of machine learning-based models for EOR screening presents a promising solution to the challenges inherent in the process. Nonetheless, issues such as data imbalance and overfitting pose significant hurdles to enhancing the predictive accuracy of these models. The goal of this research is to develop innovative EOR screening models utilizing an improved Stacking ensemble learning approach to address these challenges, thereby offering reservoir engineers more precise guidance for the swift and effective selection of the most appropriate EOR methods. In this study, traditional models such as Random Forests (RF), Extreme Gradient Boosting (XGBoost), Artificial Neural Networks (ANN), Decision Tree (DT), Support Vector Machine (SVM) and Logistic Regression (LR) were employed to establish classical EOR screening frameworks. Then based on the improved Stacking ensemble learning, these models were integrated to form advanced EOR ensemble screening models. To comprehensively assess model performance in the context of data imbalance, in addition to conventional evaluation indicators, the Kappa coefficient and the Matthews Correlation Coefficient (MCC) were introduced as novel evaluation metrics. The findings indicate that the accuracy, Kappa coefficient and MCC value of the newly developed models can reach up to 96.87 %, 0.955 and 0.955, which are much higher than other EOR screening models. The new models can provide more accurate and faster EOR screening decision support for reservoir engineers.