Temporal-Aware Structure-Semantic-Coupled Graph Network for Traffic Forecasting

1. Environment

Our experiments are conducted based on Python 3.9 and Pytorch 1.21.1. You can install Pytorch following the instructions of Pytorch.

And you can use the following command to install other packages needed for experiments.

pip install -r requirements.txt

2. Data Preparation

2.1 Download Data

The datasets used in our paper are collected by the Caltrans Performance Measurement System(PeMS). The raw data can be downloaded from ASTGNN (IEEE TKDE2021)

After downloading, the raw data should be placed as follows.

./data/PEMS03/PEMS03.npz
./data/PEMS04/PEMS04.npz
./data/PEMS07/PEMS07.npz
./data/PEMS08/PEMS08.npz

2.2 Data Process

Taking dataset PEMS08 for example, you can use the following commands to generate time series data from raw data.

python generate_time_series_data.py --data_name=PEMS08

3. Training

3.1 Encoder

After generating the time series data, use the following command to train the encoder and save the encoded representation.

python train_encoder.py --data_name=PEMS08

3.2 Labeler

After training the encoder, use the following command to train the labeler to generate and save labels.

python train_labeler.py --data_name=PEMS08

3.3 Predictor

After label generation, use the following command to train the predictor to predict the label of future series from the history series.

python train_predictor.py --data_name=PEMS08

3.4 Self-Sampling

After generating the predicted label, use the following command to generate the self-sampling series index, which will be used to generate temporal-aware graphs in TASSGN.

python generate_self_sampling_index.py --data_name=PEMS08

3.5 TASSGN

Finally, use the following command to train TASSGN and get the final test result.

python train_TASSGN.py --data_name=PEMS08

Citing

@article{chen2024temporal,
  title={Temporal-aware structure-semantic-coupled graph network for traffic forecasting},
  author={Chen, Mao and Han, Liangzhe and Xu, Yi and Zhu, Tongyu and Wang, Jibin and Sun, Leilei},
  journal={Information Fusion},
  pages={102339},
  year={2024},
  publisher={Elsevier}
}

ChenMao20374042 / TASSGN