HiSTGNN

This is a Pytorch implementation of the paper: HiSTGNN: Hierarchical Graph Neural Networks for Weather Forecasting.

Running requirements

The basic dependencies are Python 3 and Torch 1.2.0. The others are specified in requirements.txt.

You can run the following commands to prepare the environment. We recommend the virtual conda environment.

conda create --name [myenv]
source activate [myenv]
pip install -r ./HiSTGNN/requirements.txt

Data

We place the used data in Google driver. You can download them and then put them in the following corresponding paths.

1. "./HiSTGNN/data/wfd_BJ"
2. "./HiSTGNN/data/wfd_Israel"
3. "./HiSTGNN/data/wfd_USA"

Model Training

wfd_BJ

bash BJ_train.sh

wfd_Israel

bash ISR_train.sh

wfd_USA

bash USA_train.sh

If you find our work helpful for your research, please consider citing our paper. :-)

@article{ma2023histgnn,
  title={HiSTGNN: Hierarchical spatio-temporal graph neural network for weather forecasting},
  author={Ma, Minbo and Xie, Peng and Teng, Fei and Wang, Bin and Ji, Shenggong and Zhang, Junbo and Li, Tianrui},
  journal={Information Sciences},
  volume={648},
  pages={119580},
  year={2023}
}