A Framework for Predictive Modeling of Enhanced Oil Recovery Performance

Thesis Abstract

Enhanced oil recovery (EOR) techniques are critical for maximizing hydrocarbon extraction from mature reservoirs; however, predicting EOR performance accurately remains a significant challenge due to the complex interplay of geological, petrophysical, and operational variables. This study aims to develop a comprehensive predictive modeling framework that integrates multivariate analytical approaches and theoretical principles to improve the forecasting accuracy of EOR outcomes. The primary objectives include identifying key reservoir and operational parameters influencing EOR performance, establishing relationships among these variables through data-driven models, and validating the framework using real-world data to facilitate optimal decision-making in EOR projects. The research adopts a mixed-methods approach, combining quantitative data analysis with qualitative assessments. The study population comprises 25 mature oil fields globally, selected based on the availability of detailed production and reservoir data spanning at least five years. A stratified random sampling technique was employed to select a representative sample of 10 fields, ensuring diversity in geological characteristics and EOR methods applied. Data collection instruments include a combination of proprietary operational datasets, well logs, core analysis reports, and production records from these fields. Additional data were gathered via structured interviews with reservoir engineers and EOR specialists, providing contextual insights and validation for quantitative findings. Analytical procedures involve extensive data preprocessing, including normalization, outlier detection, and missing data imputation. The core of the modeling framework employs multivariate regression analysis, specifically partial least squares regression (PLSR), to elucidate the relationships between reservoir parameters—such as porosity, permeability, and residual oil saturation—and EOR performance indicators like incremental recovery factor, sweep efficiency, and cumulative oil production. Principal component analysis (PCA) was utilized to reduce dimensionality and identify dominant variable clusters influencing recovery outcomes. Model validation involved cross-validation techniques, including k-fold validation with k=10, and performance assessment metrics such as root mean square error (RMSE) and coefficient of determination (R²). Expected findings suggest that the developed predictive model will significantly improve the accuracy of forecasting EOR performance by accounting for complex interactions among reservoir parameters and operational variables. The framework is anticipated to highlight critical factors such as injector and producer well placement, injection rates, and fluid properties as primary determinants of recovery efficiency. Additionally, the study expects to demonstrate that integrating domain-specific theories—such as the Buckley-Leverett displacement theory and the Darcy flow model—within statistical modeling enhances interpretability and predictive power. These theoretical underpinnings will serve as anchors for the empirical relationships uncovered, ensuring that the framework aligns with fundamental reservoir physics principles. This research contributes to existing knowledge by providing a validated, scalable modeling framework that can be implemented in reservoir management to optimize EOR strategies dynamically. It bridges the gap between complex reservoir heterogeneities and practical prediction tools, facilitating data-driven decision-making in fluid injection and production optimization. Furthermore, the integration of statistical and theoretical modeling approaches advances the methodological repertoire available to petroleum engineers. The main conclusion emphasizes that the proposed framework offers a robust, adaptable tool for enhancing the precision of EOR performance prediction across diverse reservoir types. Recommendations from the study include the incorporation of real-time monitoring systems to update models dynamically, further validation using larger datasets, and exploration of machine learning techniques—such as artificial neural networks (ANN)—to refine predictive accuracy. Future research directions involve expanding the model to incorporate thermal and chemical EOR methods, integrating uncertainty quantification, and developing user-friendly digital platforms for widespread industry application.