Self-learning adaptive traffic signal control for real-time safety optimization.

Area of Science:

• Intelligent Transportation Systems (ITS)
• Traffic Engineering
• Artificial Intelligence in Transportation

Background:

• Adaptive traffic signal control (ATSC) optimizes traffic flow but often neglects safety.
• Real-time safety models for signalized intersections are emerging, enabling safety-focused control.
• Connected vehicles (CVs) provide data crucial for advanced traffic management.

Purpose of the Study:

• To develop and evaluate a novel self-learning ATSC algorithm for real-time traffic safety optimization.
• To integrate real-time safety evaluation into ATSC strategies.
• To assess the algorithm's performance under different connected vehicle market penetration rates (MPRs).

Main Methods:

• Developed a novel ATSC algorithm using Reinforcement Learning (RL).
• Trained the RL algorithm using the VISSIM simulation platform.
• Validated the algorithm with real-world traffic data from two signalized intersections.

Main Results:

• The proposed ATSC algorithm reduced traffic conflicts by approximately 40% compared to traditional systems.
• Significant safety benefits were observed even at lower connected vehicle market penetration rates (e.g., 50% benefit at 30% MPR).
• The algorithm demonstrates the feasibility of real-time safety optimization in ATSC.

Conclusions:

• This study presents the first self-learning ATSC algorithm optimized for real-time traffic safety.
• The developed algorithm offers a substantial improvement in traffic safety at signalized intersections.
• The findings highlight the potential of integrating AI and real-time safety models for future traffic control systems.