Awesome

Spatio-Temporal Differential Equation Network

This is a Pytroch implementation of Spatio-temporal Differential Equation Network (STDEN) for physics-guided traffic flow prediction, as described in our paper: Jiahao Ji, Jingyuan Wang, Zhe Jiang, Jiawei Jiang, and Hu Zhang, STDEN: Towards Physics-guided Neural Networks for Traffic Flow Prediction, AAAI 2022.

The training framework of this project comes from chnsh. Thanks a lot :)

Requirement

• scipy>=1.5.2
• numpy>=1.19.1
• pandas>=1.1.5
• pyyaml>=5.3.1
• pytorch>=1.7.1
• future>=0.18.2
• torchdiffeq>=0.2.0

Dependency can be installed using the following command:

pip install -r requirements.txt

Model Traning and Evaluation

You can run the code by

# traning for dataset GT-221
python stden_train.py --config_filename=configs/stden_gt.yaml

# testing for dataset GT-221
python stden_eval.py --config_filename=configs/stden_gt.yaml

The configuration file of all datasets are as follows:

Note the data is not public and I am not allowed to distribute it.

Cite

If you find the paper usefule, please cite as following:

@inproceedings{ji2022stden,
  title={STDEN: Towards Physics-guided Neural Networks for Traffic Flow Prediction},
  author={Ji, Jiahao and Wang, Jingyuan and Jiang, Zhe and Jiang, Jiawei and Zhang, Hu},
  booktitle={2022 AAAI Conference on Artificial Intelligence (AAAI'22)},
  year={2022} 
}