TEAM: Topological Evolution-aware Framework for Traffic Forecasting

Code for TEAM:Topological Evolution-aware Framework for Traffic Forecasting（PVLDB 2025).

TEAM is a traffic forecasting framework that utilizes continual learning algorithm to handle the evolution of topology of traffic road network (expanding or removing)

Introduction

Data

Download raw data from PeMS3_Evolve and PeMS4_Evolve, unzip the file and put it in the data folder. We also provide another version of PeMS4 with different evolve frequency (evolve daily and weekly) PeMS4_Evolve_Diff

Usage

Create conda environment by

conda env create -f trafficStream.yaml

To run CAST (retrain for each year)

python main.py --conf conf/cast_pems04.json --gpuid 1

To run TEAM (incremental training)

python main.py --conf conf/team_pems04.json --gpuid 1

Citation

@article{kieu2024team,
  title={TEAM: Topological evolution-aware framework for traffic forecasting},
  author={Kieu, Duc and Kieu, Tung and Han, Peng and Yang, Bin and Jensen, Christian S and Le, Bac},
  journal={Proceedings of the VLDB Endowment},
  volume={18},
  number={2},
  pages={265--278},
  year={2024},
  publisher={VLDB Endowment}
}