A deep reinforcement learning-based intelligent intervention framework for real-time proactive road safety management

Area of Science:

• Traffic Engineering
• Artificial Intelligence
• Road Safety

Background:

• Existing variable speed limit (VSL) systems face limitations in applicability due to varying traffic detector spacing.
• Real-time safety interventions for urban expressways require more adaptable and effective prediction and control mechanisms.

Purpose of the Study:

• To develop and evaluate a novel real-time variable speed limit (VSL) system for improving urban expressway safety.
• To address the limitations of existing VSL systems by incorporating adaptable traffic simulation and advanced control algorithms.

Main Methods:

• Utilized a cell transmission model (CTM) to simulate traffic states adaptable to different detector spacings.
• Developed a real-time crash prediction model (RTCPM) using a dynamic Bayesian network (DBN).
• Employed a deep Q-network, a reinforcement learning algorithm, for intelligent VSL control strategy selection.

Main Results:

• The proposed CTM was modified to integrate VSL control capabilities.
• The real-time system effectively assessed crash risk and triggered VSL adjustments.
• A significant reduction in crash risk by 19% was achieved in the study area.

Conclusions:

• The integrated VSL system demonstrates a promising approach to real-time traffic safety enhancement on urban expressways.
• The use of CTM, DBN, and deep Q-networks offers a robust framework for adaptable VSL control.
• This system provides a significant improvement over existing methods for managing traffic safety proactively.