Adaptive Hybrid Spatial-Temporal Graph Neural Network for Cellular Traffic Prediction

This is the original pytorch implementation of AHSTGNN in the following paper: Adaptive Hybrid Spatial-Temporal Graph Neural Network for Cellular Traffic Prediction, ICC 2023.

Requirements

python 3.9
numpy == 1.20.3
scipy == 1.7.3
pandas == 1.5.3
torch == 1.11.0

Dependencies can be installed using the following command:

pip install -r requirements.txt

Data

Step1: The Milan dataset used in the paper can be downloaded from Google Driver or Baidu Pan, password p9gx.

Step2: Process raw data

# Create data directories
mkdir -p data/{Milan}

# Milan
python generate_training_data.py --output_dir=data/Milan --traffic_df_filename=data/data_mi_min.npy

Train Commands

python train.py --gcn_bool --adjtype doubletransition --addaptadj  --randomadj

Main results

We compare our model with typical cellular traffic prediction methods including HA (Historical Average), LSTM, MVSTGN and AMF-STGCN, as well as the generic advanced spatial-temporal sequence prediction methods including Graph WaveNet, MTGNN and AGCRN.

Citation

if you find this repository useful, please cite our paper.

@article{wang2023adaptive,
  title={Adaptive Hybrid Spatial-Temporal Graph Neural Network for Cellular Traffic Prediction},
  author={Wang, Xing and Yang, Kexin and Wang, Zhendong and Feng, Junlan and Zhu, Lin and Zhao, Juan and Deng, Chao},
  journal={arXiv preprint arXiv:2303.00498},
  year={2023}
}

Acknowledgement

We appreciate the Graph WaveNet a lot for the valuable code base: https://github.com/nnzhan/Graph-WaveNet