GeoPix Framework for Subsurface Reservoir Characterization

Overview

The GeoPix Framework leverages Generative Adversarial Networks (GANs), specifically the Pix2Pix model, to generate high-resolution property maps such as porosity and permeability from facies images. This model is designed to improve subsurface reservoir characterization by providing more accurate predictions of rock properties.

Network Architecture

Generator

• A U-Net-based architecture designed to generate property maps (e.g., porosity, permeability) from facies input images.
• The U-Net structure allows for high-quality image-to-image translation by combining convolutional layers with skip connections.

Discriminator

• A PatchGAN-based convolutional network is used to evaluate the realism of the generated property maps.
• It operates on image patches to ensure that both global structure and local texture are realistic.

---

Architecture of the GeoPix Framework

---

High-level Architecture of the Proposed GeoPix Framework for Subsurface Reservoir Characterization

---

Key Features

• Data-Driven Modeling: Utilizes machine learning models to predict rock properties directly from geological facies data.
• Robust Performance: The combination of U-Net and PatchGAN improves the generation of high-resolution, geologically plausible property maps.
• Application in Reservoir Characterization: The model facilitates a deeper understanding of subsurface formations, which is crucial for hydrocarbon exploration and production.

Repository Structure

GeoPix/
├── Figures/                 # Contains architecture and framework images
├── Examples_of_Results/     # Example output images
├── model/                   # GAN model architecture files
├── data/                    # Training and testing datasets
├── README.md                # Project overview and documentation
└── requirements.txt         # Python dependencies

How to Run

1. Clone the repository:

   git clone https://github.com/ARhaman/GeoPix.git
   cd GeoPix

2. Install the dependencies: bash Copy code pip install -r requirements.txt
3. Train the model: bash Copy code python train.py --dataset your_dataset --epochs 100
4. Generate predictions: bash Copy code python generate.py --input facies_image.png --output predicted_property.png Contact For any questions or collaboration inquiries, please contact:

Abdulrahman Al-Fakih Ph.D. Researcher | Geophysics King Fahd University of Petroleum and Minerals (KFUPM) 📞 +966 500916367 📧 alja2014ser@gmail.com