Multi-Agent Adaptive Traffic Signal Control Based on Q-Learning and Speed Transition Matrices

Area of Science:

• Transportation Engineering
• Artificial Intelligence
• Traffic Management Systems

Background:

• Connected vehicles (CVs) offer real-time microscopic traffic data, enhancing road capacity and safety.
• Vehicle-to-Everything (V2X) communication enables CVs to act as mobile sensors.
• Speed Transition Matrices (STMs) can process CV data while preserving spatio-temporal features.

Purpose of the Study:

• To propose a novel adaptive traffic signal control strategy for connected vehicle environments.
• To leverage STMs and cooperative multi-agent learning for intelligent traffic management.
• To evaluate the proposed system's performance under varying CV penetration rates and cooperation levels.

Main Methods:

• Development of an adaptive traffic signal control system integrating STMs and cooperative multi-agent learning.
• Simulation of an intersection network environment to test the proposed control strategy.
• Comparative analysis of system performance across different CV penetration rates and agent cooperation coefficients.

Main Results:

• The proposed adaptive traffic signal control system demonstrates effectiveness in simulated intersection networks.
• Performance improvements were observed with increased CV penetration rates and higher agent cooperation.
• The integration of STMs and multi-agent learning provides a robust approach to traffic signal optimization.

Conclusions:

• The novel adaptive traffic signal control approach utilizing STMs and cooperative multi-agent learning is effective for CV environments.
• This research highlights the potential of CVs as mobile sensors for advanced traffic management.
• Future work can explore real-world deployment and scalability of the proposed system.