StreamerNet is a cutting-edge deep learning framework designed to model and predict the complex dynamics of streamer discharges using state-of-the-art neural network architectures.
-
Multiple Neural Network Architectures
- U-Net: Specialized for spatial feature extraction
- Fourier Neural Operator (FNO2D): Leverages spectral methods for efficient computation
- Cylindrical Symmetric FNO (CSFNO2D): Custom architecture optimized for cylindrical symmetry
-
Advanced Data Processing
- Handles complex HDF5 datasets
- Supports multi-dimensional time series data
- Built-in data normalization and preprocessing
- Efficient batch processing for large-scale simulations
-
Comprehensive Analysis Tools
- Fourier transform analysis capabilities
- Real-time visualization of predictions
- Model performance comparison utilities
- Detailed logging and monitoring through Weights & Biases
Our models have demonstrated remarkable success in predicting streamer discharge evolution:
- Accurate prediction of electron density distributions
- Precise electric field calculations
- Efficient computation compared to traditional numerical methods
- Validated against high-fidelity physical simulations
- Input format: HDF5
- Supported features: electron density ('e'), electric field
- Configurable time steps for input and prediction
- Customizable network architectures
- Flexible hyperparameter tuning
- Support for various activation functions
- Adaptable for different physical domains
This project is licensed under the GNU General Public License v3.0 - see the LICENSE file for details.
This research was supported by Ramón Areces Foundation. Special thanks to Jannis Teunissen for his valuable feedback and support.