Cooperative Traffic Signal Control with Traffic Flow Prediction in Multi-Intersection.

Area of Science:

• Intelligent Transportation Systems
• Artificial Intelligence
• Traffic Engineering

Background:

• Urban traffic congestion is a growing problem.
• Deep Reinforcement Learning (DRL), particularly Deep Q-Network (DQN), shows promise for traffic signal control.
• Existing DRL methods often focus on single intersections and lack real-world variable consideration due to simulator limitations.

Purpose of the Study:

• To propose a novel cooperative traffic signal control system for multi-intersection environments.
• To integrate traffic flow prediction (TFP) into cooperative control to account for real-world variables.
• To enhance traffic signal control efficiency in complex urban settings.

Main Methods:

• Developed a traffic flow prediction model to forecast future traffic states, incorporating real-world variables.
• Modeled each intersection as an agent within a multi-agent reinforcement learning framework.
• Implemented information sharing between adjacent intersection agents for efficient cooperative control.

Main Results:

• The proposed Traffic Flow Prediction-Cooperative Traffic Signal Control (TFP-CTSC) method was evaluated in a 4x4 intersection environment.
• Experimental results demonstrated the effectiveness of both the traffic flow prediction model and the cooperative control strategy.
• TFP-CTSC showed improved performance compared to existing traffic signal control algorithms.

Conclusions:

• The TFP-CTSC approach offers a viable solution for intelligent traffic signal control in multi-intersection scenarios.
• Integrating traffic flow prediction and cooperative multi-agent learning enhances traffic management efficiency.
• The study validates the potential of DRL for addressing complex, real-world traffic congestion challenges.