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.
Dependencies can be installed using the following command:
pip install -r requirements.txt
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
python train.py --gcn_bool --adjtype doubletransition --addaptadj --randomadj
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.
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}
}
We appreciate the Graph WaveNet a lot for the valuable code base: https://github.com/nnzhan/Graph-WaveNet