HydroNet

Physics-informed machine learning framework for solving 2D shallow water equations using deep learning.

Overview

HydroNet is a PyTorch-based framework that implements physics-informed neural networks (PINN) and physics-informed DeepONet (PI-DeepONet) for solving 2D shallow water equations. The framework enables operator learning and provides tools for training, testing, and applying models to hydrodynamics and hydraulics problems.

Features

• Physics-Informed Neural Networks (PINN): Direct solution of PDEs using neural networks with physics constraints
• Physics-Informed DeepONet (PI-DeepONet): Operator learning for parameterized PDE solutions
• 2D Shallow Water Equations: Specialized implementation for hydrodynamics modeling
• Flexible Configuration: YAML-based configuration system for easy model customization
• Data Processing: Utilities for handling GMSH meshes, VTK files, and SRH-2D simulation data
• Visualization: Tools for plotting solutions and training history

Installation

Requirements

• Python >= 3.8
• PyTorch >= 2.0.0
• NumPy, SciPy, Pandas
• VTK >= 9.2.0 (for mesh processing)
• pyHMT2D (for SRH-2D simulation data; https://github.com/psu-efd/pyHMT2D)
• Additional dependencies listed in requirements.txt

Install from Source

git clone https://github.com/psu-efd/HydroNet.git
cd HydroNet
pip install -e .

For development dependencies:

pip install -e ".[dev]"

Quick Start

See the examples/ directory for complete working examples:

• examples/PINN/: PINN model examples
• examples/PI_DeepONet/: PI-DeepONet model examples

Project Structure

HydroNet/
├── HydroNet/
│   ├── models/
│   │   ├── PINN/          # PINN model implementation
│   │   └── PI_DeepONet/   # PI-DeepONet model implementation
│   ├── utils/             # Data processing and utility functions
│   └── configs/           # Configuration files
├── examples/              # Example scripts and data
├── requirements.txt       # Python dependencies
└── setup.py              # Package setup

Examples

See the examples/ directory for complete working examples:

• examples/PINN/: PINN model examples
• examples/PI_DeepONet/: PI-DeepONet model examples

Configuration

Models are configured using YAML files. Key configuration sections include:

• Model architecture: Network layers, activation functions, dropout
• Physics parameters: Gravity, scales, steady/transient settings
• Training parameters: Learning rate, batch size, loss weights, schedulers
• Data paths: Training, validation, and test data locations

See examples for example configurations.

License

MIT License - see LICENSE file for details.

Author

Xiaofeng Liu (xiaofengliu19@gmail.com)

Citation

If you use HydroNet in your research, please cite:

@misc{liu2026,
      title={Physics-Informed Deep Operator Learning for Computational Hydraulics Modeling}, 
      author={Xiaofeng Liu and Yong G. Lai},
      year={2026},
      eprint={2601.08086},
      archivePrefix={arXiv},
      primaryClass={physics.flu-dyn},
      url={https://arxiv.org/abs/2601.08086}, 
}

@software{hydronet2025,
  title={HydroNet: Physics-informed machine learning for hydrodynamics modeling},
  author={Liu, Xiaofeng},
  year={2026},
  url={https://github.com/psu-efd/HydroNet}
}