PhICNet

Repository for our work Physics-Incorporated Convolutional Recurrent Neural Networks for Source Identification and Forecasting of Dynamical Systems

If you find our work useful, please don't forget to cite.

@article{saha2021physics,
  title={Physics-incorporated convolutional recurrent neural networks for source identification and forecasting of dynamical systems},
  author={Saha, Priyabrata and Dash, Saurabh and Mukhopadhyay, Saibal},
  journal={Neural Networks},
  year={2021},
  publisher={Elsevier}
}

Installation

Compatible with Python 3.5 and Pytorch 1.1.0

1. Create a virtual environment by python3 -m venv env
2. Source the virtual environment by source env/bin/activate
3. Install requirements by pip install -r ./requirements.txt

Usage

Dataset and pre-trained models can be downloaded from these two links: dataset and models.

Training

To train heat system, run python scripts/train_heat_system.py --dataset ./dataset/train_heat_maps.npy

To train wave system, run python scripts/train_wave_system.py --dataset ./dataset/train_wave_maps.npy

Evaluation

To evaluate heat system, run python scripts/test_heat_system.py --dataset ./dataset/test_heat_maps.npy --model_path ./saved_models/heat_system/model.ckpt --param_path saved_models/heat_system/parameters.ckpt

To evaluate wave system, run python scripts/test_wave_system.py --dataset ./dataset/test_wave_maps.npy --model_path ./saved_models/wave_system/model.ckpt --param_path ./saved_models/wave_system/parameters.ckpt