An AI-powered solution that uses deep reinforcement learning to discover optimal wind turbine layouts, maximizing energy production by intelligently minimizing wake interference.
Wind turbines create "wakes" of slower, turbulent air behind them, reducing downstream turbine efficiency by 10-15%. Traditional layout design relies on manual iteration or simple grid patterns, leaving significant energy potential untapped.
Project ATLAS solves this using AI.
- 🤖 Deep Reinforcement Learning - DDPG agent learns optimal placement strategies
- 🌍 Real-World Data - Integrates actual terrain and wind climate data from Tamil Nadu, India
- ⚡ Physics-Based Simulation - Implements Jensen Wake Model for accurate AEP calculations
- 📊 Proven Results - Achieves 92.6-99.5% efficiency across different scenarios
- ⚙️ Highly Configurable - Simple YAML-based configuration for all parameters
- 📁 Reproducible - Automatic experiment tracking and result archiving
- Anaconda
- Git
# Clone the repository
git clone https://github.com/realyashagarwal/Project-ATLAS.git
cd Project-ATLAS
# Create and activate the conda environment
conda env create -f environment.yml
conda activate atlas_env# Run with default settings (10 turbines, 500 episodes)
python src/train.pyResults will be automatically saved in outputs/ with timestamp.
Edit config.yaml to adjust parameters:
turbine:
max_turbines: 20 # Number of turbines to optimize
training:
total_episodes: 1000 # Training duration
site:
area_size: 2000 # Site dimensions in meters| Configuration | Turbines Placed | AEP (MWh/year) | Efficiency |
|---|---|---|---|
| 10-Turbine | 10 | 216,741 | 99.5% |
| 20-Turbine | 19* | 403,459 | 92.6% |
*Agent learned that 19 turbines was more efficient than placing all 20
ATLAS/
├── config.yaml # Central configuration file
├── environment.yml # Conda environment specification
├── README.md # This file
├── REPORT.md # Detailed technical report
│
├── data/
│ ├── terrain.tif # SRTM elevation data
│ └── wind_data.lib # Global Wind Atlas wind data
│
├── notebooks/
│ └── 1-Data_Exploration.ipynb # Data parsing & visualization
│
├── outputs/ # Timestamped experiment results
│ └── 2025-10-01_21-24-00/
│ ├── config.yaml
│ ├── final_layout.png
│ └── training_log.txt
│
└── src/
├── agent.py # DDPG agent implementation
├── environment.py # Wind farm simulation environment
├── main.py # Component testing script
└── train.py # Main training pipeline
A digital twin of a wind farm site incorporating:
- Real wind data from Global Wind Atlas (12 directional sectors)
- Terrain elevation from NASA SRTM
- Jensen Wake Model for physics-based AEP calculation
A Deep Deterministic Policy Gradient (DDPG) agent that:
- Places turbines sequentially on the site
- Receives rewards for increasing AEP
- Gets penalties for constraint violations (spacing, boundaries)
- Learns optimal placement strategies through trial and error
Over hundreds of episodes, the agent:
- Explores different placement strategies
- Learns to avoid wake interference
- Discovers counter-intuitive but efficient layouts
- Converges to near-optimal solutions
Core Technologies:
- Python 3.10+
- TensorFlow 2.x (Deep Learning)
- Gymnasium (RL Environment)
Data & Computation:
- NumPy & Pandas (Data processing)
- Rasterio (Geospatial analysis)
- PyYAML (Configuration)
Visualization:
- Matplotlib (Plots)
- Tableau (Dashboard)
- REPORT.md - Comprehensive technical documentation
- Data Exploration Notebook - Data parsing and analysis
This project uses open data from:
- Terrain: NASA Shuttle Radar Topography Mission (SRTMGL1) via OpenTopography
- Wind Climate: Global Wind Atlas by DTU & World Bank Group
- Renewable Stats: Ministry of New and Renewable Energy (MNRE), Government of India
Developed as part of the 1M1B Green Internship, this project demonstrates how advanced AI can address real-world climate challenges. Tamil Nadu has significant untapped wind energy potential—ATLAS aims to help maximize that potential through intelligent design.
- Terrain-aware optimization (slope penalties)
- Interactive Streamlit web interface
- Additional RL algorithms (PPO, SAC)
- Docker containerization
- Multi-turbine type support
- Real-time wind data integration
Contributions are welcome! Whether it's bug fixes, feature additions, or documentation improvements, feel free to:
- Fork the repository
- Create a feature branch (
git checkout -b feature/AmazingFeature) - Commit your changes (
git commit -m 'Add some AmazingFeature') - Push to the branch (
git push origin feature/AmazingFeature) - Open a Pull Request
This project is licensed under the MIT License - see the LICENSE file for details.
Yash Agarwal
- GitHub: @realyashagarwal
Special thanks to:
- Technical University of Denmark for the Global Wind Atlas
- NASA and OpenTopography for terrain data
- The open-source community for the amazing tools and libraries
If you find this project useful, please consider giving it a ⭐️
Made with ❤️ for a sustainable future