Reservoir Characterization and Enhanced Oil Recovery Optimization Using Integrated Geomechanics and Machine Learning for Unconventional Tight Reservoirs

Project Abstract

Reservoir characterization and Enhanced Oil Recovery (EOR) optimization in unconventional tight reservoirs require a holistic approach that integrates geomechanics with advanced machine learning (ML) techniques to accurately predict reservoir behavior, optimize injection strategies, and maximize hydrocarbon recovery while mitigating stimulation risks. This study develops a multi-scale framework that combines rock-physics-informed geomechanical modeling, high-resolution imaging, and data-driven analytics to capture the coupled flow-deformation-thermal processes governing tight formations. The research leverages a robust dataset comprising core samples, wireline logs, X-ray computed tomography (CT) scans, microseismic events, production history, and lab-derived PVT and relative permeability measurements to calibrate a unified reservoir model. A novel aspect is the integration of geomechanical response with ML-powered surrogate models to accelerate scenario analysis across uncertain petrophysical parameters, fracture networks, and stress regimes typical of shale and tight carbonate systems. The methodology begins with detailed petrophysical characterization to delineate pore-scale features, nano- to micro-fracture networks, and mineralogy, followed by geomechanical modeling that accounts for in-situ stress, pore pressure changes, and rock mechanical properties under varying production schedules. This is coupled with high-fidelity flow simulations that incorporate shale gas/oil transport mechanisms, dual-porosity/dual-permeability concepts, and dynamic fracture conductivity. To overcome computational bottlenecks, a suite of machine learning algorithms, including deep neural networks, gradient boosting, and Gaussian processes, are trained on high-fidelity simulations and experimental data to develop fast surrogate models for permeability evolution, fracture aperture changes, and induced seismicity risk indicators. Active learning strategies are employed to iteratively refine model predictions with new field data, reducing epistemic uncertainty and enabling adaptive field development plans. A key objective is to optimize EOR strategies—such as polymer, surfactant, alkali, or CO2 injections—by evaluating their effects on sweep efficiency, pressure management, and fracture conductivity under coupled geomechanical constraints. The research introduces an optimization framework that integrates ML-predicted reservoir responses with economic metrics (net present value, break-even oil price, and operating costs) to identify optimal injection strategies, timing, and well placement in unconventional plays. Validation is performed against field-case studies from fractured shale and tight carbonate reservoirs, where the integrated approach demonstrates improved prediction accuracy for production forecasts, reduced uncertainty in critical parameters, and enhanced control over geomechanical risk factors. The project also assesses environmental and operational risks, including induced seismicity, fluid leakage pathways, and stimulation-induced damage, proposing mitigation measures grounded in probabilistic risk assessment. The expected outcome is a transferable, scalable toolkit that enables operators to perform rapid, data-driven optimization of EOR operations in tight reservoirs while maintaining reservoir integrity and economic viability.

Project Overview

What This Project Is About

---

A straightforward, easy-to-understand look at how scientists study oil reservoirs that are hard to extract from. The project combines simple geology with light computer ideas to understand how oil moves underground and how to get more of it out safely.

---

---

The Problem It Addresses

---

Unconventional tight reservoirs release oil slowly because rocks are very dense and connected pathways are limited. Traditional methods may waste time and money. The project aims to find better ways to predict where oil hides and how to unlock more oil efficiently.

---

---

Objectives of the Project

---

1. Explain the basic challenges of extracting oil from tight rocks in simple terms.
2. Introduce geomechanics and machine learning ideas with clear examples.
3. Show how combining these ideas can improve oil recovery predictions.
4. Develop a small, practical workflow students can simulate or test.
5. Discuss potential benefits and limits of the approach for industry.

---

---

What You Will Do Step by Step

---

1) Learn basic terms used in oil reservoirs and why tight rocks behave differently. 2) Review simple case studies to see existing methods. 3) Build a basic model that links rock behavior to oil flow. 4) Introduce a lightweight machine-learning idea to improve predictions. 5) Run small simulations with sample data. 6) Interpret results in plain language. 7) Discuss practical steps to apply the approach in real projects.

---

---

---

---

Expected Outcome

A clear, easy-to-follow framework that shows how geomechanics plus simple machine-learning ideas can help predict oil recovery from tight reservoirs. The project should provide a practical set of steps, potential gains in efficiency, and guidance on limitations for real-world use.