1. Introduction

Official code for article Leveraging Queue Length and Attention Mechanisms for Enhanced Traffic Signal Control Optimization.

This article has been received by ECML PKDD 2023.

If you use our method, please cite our article.

@inproceedings{attentionlight,
  title={Leveraging Queue Length and Attention Mechanisms for Enhanced Traffic Signal Control Optimization},
  author={Zhang, Liang and Xie, Shubin and Deng, Jianming},
  booktitle={Joint European Conference on Machine Learning and Knowledge Discovery in Databases},
  pages={141--156},
  year={2023},
  organization={Springer}
}

2. Requirements

python3.6,tensorflow=2.4, cityflow, pandas, numpy

cityflow needs a Linux environment, and we run the code on Manjaro Linux.

3. Quick start

For the method in our article, run the following:

python run_attention_light.py

python run_max_ql.py

python run_ql_dqn.py

python run_ql_frap.py

python run_ql_gat.py

For the baseline methods,

• Fixed-Time

  python run_fixedtime.py

• Max-Pressure

  python run_maxpressure.py

• PressLight

  python run_presslight.py

• MPLight

  python run_mplight.py

• FRAP

  python run_frap.py

• Colight

  python run_colight.py

  3.1 Evaluate the results

  Change the folder name in summary.py to yours, and run:

  python summary.py

  4. Code details

  4.1、structure

• models: contains all the models used in our article.
• utils: contains all the methods to simulate and train the models.

4.2、Reference

The code is modified from Efficient_XLight. The Max-Pressure is created by ourselves, based on MaxPressure.

• PressLight: Based on LIT model, which comes from Colight.
• Colight : Based on Colight.
• Fixed-Time: From MPLight.
• MPLight: From MPLight.

License

This project is licensed under the GNU General Public License version 3 (GPLv3) - see the LICENSE file for details.