Traffic signal active control method for short-distance intersections.

Area of Science:

• Traffic Engineering
• Artificial Intelligence
• Transportation Systems Analysis

Background:

• Optimizing traffic signal control for short-distance intersections presents challenges in balancing overflow prevention and overall traffic efficiency.
• Existing methods struggle to simultaneously address these competing objectives.

Purpose of the Study:

• To propose an active traffic signal control method for short-distance intersections based on key state prediction.
• To enhance traffic flow and mitigate overflow risks in complex intersection scenarios.

Main Methods:

• Introduced the concept of an 'overflow index' for short-distance road sections and developed a prediction method for it.
• Utilized deep reinforcement learning for fast computation and solution of the active control scheme.
• Optimized the deep reinforcement learning algorithm to address reward sparsity, enhancing state space and reward function capabilities.

Main Results:

• The proposed method effectively ensures overall traffic efficiency and reduces travel delay at short-distance intersections.
• The system actively senses changes in overflow states, improving prevention and control capabilities.
• Demonstrated a significant reduction in overflow risk within the target scenarios.

Conclusions:

• The active control method based on key state prediction offers a viable solution for complex traffic management at short-distance intersections.
• This approach successfully balances traffic efficiency with overflow prevention, leading to safer and smoother traffic flow.
• The integration of a novel overflow index and advanced deep reinforcement learning provides a robust framework for intelligent traffic signal control.