Awesome

GTA.

A Graph-Based Temporal Attention Framework for Multi-Sensor Traffic Flow Forecasting

Dataset

We evaluate GTA using real traffic data collected from all the strategic road network in England. The traffic dataset includes average speed, traffic flow, time period, location, and date of 249 monitoring stations. The sensors are located from site A414 between M1 J7 and A405, which covers several cities that include Manchester, Liverpool, and Blackburn. Specifically, we use a whole year of traffic data ranging from January 1st, 2014 to December 31st, 2014 for the experiments. The total number of data entries is 8,724,960, the mean value of traffic volume is 466.

We process the collected traffic data by normalizing them to [0, 1] before we feed them into the algorithm. Because the dataset does not contain road network distance between two monitoring stations, we enhance the data with these topology information based on distances collected from Google Services.

The distribution of monitoring stations studied in our experiment.

Citation

If you find this repository useful in your research, please cite the following paper:

@article{zhang2021graph,
  title={A Graph-Based Temporal Attention Framework for Multi-Sensor Traffic Flow Forecasting},
  author={Zhang, Shaokun and Guo, Yao and Zhao, Peize and Zheng, Chuanpan and Chen, Xiangqun},
  journal={IEEE Transactions on Intelligent Transportation Systems},
  year={2021},
  publisher={IEEE}
}

Contact

For questions, please feel free to reach out via email at skzhang@pku.edu.cn.