Electrokinetic Enhanced Oil Recovery (EK-EOR) Prediction Using Machine Learning

🧠 Project Overview

This project applies supervised machine learning algorithms to predict the Recovery Factor (RF) in Electrokinetic Enhanced Oil Recovery (EK-EOR) processes. Using a synthetically generated dataset based on key reservoir and electrokinetic parameters, we evaluate the predictive performance of three models:

• Random Forest Regressor (RFR)
• LightGBM (LGBM)
• Polynomial Support Vector Machine (SVM)

Each model's performance is benchmarked using standard regression metrics and compared through visualized outputs.

🎯 Problem Statement

Enhanced Oil Recovery (EOR) through electrokinetics introduces electric fields to improve oil displacement efficiency. However, predicting Recovery Factor under such complex multiphysics interactions is challenging. This project explores the feasibility of ML models in capturing these nonlinear dependencies for robust and accurate RF prediction.

🛠️ Tools & Technologies

• Language: Python 3.8+
• Libraries: scikit-learn, LightGBM, matplotlib, pandas, numpy
• Modeling: Random Forest, LightGBM, Polynomial SVM
• Visualization: matplotlib, seaborn

📁 Project Structure

EK-EOR-Prediction/
├── LGBM_Model.py                 # LightGBM model training and evaluation
├── Polynomial_SVM_Model.py      # Polynomial SVM model training and evaluation
├── model_comparison_plots.py    # Performance visualization for model comparison
├── synthetic_data_generation.py # Synthetic dataset generation
├── requirements.txt             # Required dependencies
└── README.md                    # Project documentation

🧪 Methodology

1️⃣ Synthetic Data Generation

Features generated include:

• Zeta Potential
• Oil Viscosity
• Core Geometry (Length, Diameter, Area)
• Electroosmotic Permeability
• Darcy Flow Properties

This data mimics realistic reservoir conditions and is used for training/testing.

2️⃣ Model Training & Evaluation

Each model script includes:

• Data Preprocessing
• Model Training
• Metric Evaluation (MSE, MAE, R²)
• Performance Logging

3️⃣ Model Comparison

model_comparison_plots.py plots the comparative performance of all models for a clear visualization of strengths and weaknesses.

📊 Results Summary

Model	MSE	MAE	R² Score
Random Forest	4.2150e-06	1.7073e-04	0.9999
LightGBM	1.1899e-05	5.5813e-04	0.9983
Polynomial SVM	2.7362e-03	4.5293e-02	0.6101

👉 Conclusion: Random Forest and LightGBM exhibit superior accuracy, making them the preferred models for RF prediction in EK-EOR applications.

⚙️ Installation & Usage

🔹 Prerequisites

Ensure Python 3.8+ and pip are installed.

Install dependencies:

pip install -r requirements.txt

🔹 Running the Scripts

# Step 1: Generate synthetic dataset
python synthetic_data_generation.py

# Step 2: Train LightGBM model
python LGBM_Model.py

# Step 3: Train Polynomial SVM model
python Polynomial_SVM_Model.py

# Step 4: Compare and visualize model performances
python model_comparison_plots.py

🚀 Future Enhancements

• Integrate more domain-specific parameters (e.g., salinity, pH, wettability)
• Add Bayesian and ensemble stacking methods
• Deploy using Streamlit or Flask for real-time predictions
• Incorporate deep learning models like MLPs or Transformers

👤 Contributor

Victor Ikechukwu
Petroleum Engineer & Machine Learning Enthusiast
📧 Email: vikechukwu@gmail.com
🎓 Federal University of Technology Owerri
🔗 GitHub Profile