Design of reinforcement learning for perimeter control using network transmission model based macroscopic traffic...

Area of Science:

• Traffic Engineering
• Urban Planning
• Artificial Intelligence

Background:

• Perimeter control is an emerging traffic signal control strategy for road networks.
• Existing model-based approaches using macroscopic fundamental diagrams (MFDs) face scalability and MFD shape variation challenges in large urban areas.

Purpose of the Study:

• To propose a scalable, model-free, and data-driven perimeter control approach for urban traffic networks.
• To combine reinforcement learning (RL) with network transmission models for traffic simulation and control.

Main Methods:

• Developed four perimeter control models using different macroscopic traffic variables and parametrizations.
• Integrated reinforcement learning with a network transmission model for macroscopic traffic simulation.
• Validated models using diverse test demand scenarios.

Main Results:

• The proposed model-free, data-driven approach addresses limitations of traditional model-based methods.
• Models incorporating travel demand information demonstrated superior adaptability to new demand scenarios compared to density-focused models.
• Reinforcement learning integration proved effective for optimizing perimeter control.

Conclusions:

• A data-driven, reinforcement learning-based perimeter control strategy offers a scalable and adaptive solution for urban traffic management.
• Integrating travel demand into control models enhances network resilience to fluctuating traffic conditions.
• This approach provides a promising alternative to traditional MFD-based perimeter control methods.