Deep Q-network-based traffic signal control models.

Area of Science:

• Artificial Intelligence
• Transportation Engineering
• Urban Planning

Background:

• Widespread urban traffic congestion necessitates innovative solutions.
• Artificial intelligence (AI) offers advanced capabilities for complex problems like traffic signal control.
• Reinforcement learning (RL) is a key AI technique showing promise in dynamic system optimization.

Purpose of the Study:

• To develop and evaluate AI-based traffic signal control models for urban intersections.
• To compare the performance of reinforcement learning models against traditional traffic signal optimization methods.
• To assess the effectiveness of AI in managing both isolated and coordinated traffic intersections.

Main Methods:

• Developed two traffic signal control models utilizing a deep Q-network (DQN), a type of reinforcement learning algorithm.
• Employed microscopic simulation for performance evaluation of the developed AI models.
• Benchmarked AI models against a fixed-time signal control strategy optimized using the Synchro model.

Main Results:

• The developed deep Q-network (DQN) traffic signal control model for an isolated intersection demonstrated successful validation.
• AI-based traffic signal control for coordinated intersections significantly outperformed the fixed-time signal control method.
• Reinforcement learning models showed superior performance in optimizing traffic flow compared to conventional methods.

Conclusions:

• AI, particularly reinforcement learning with deep Q-networks, provides an effective solution for traffic signal control.
• Coordinated intersection management using AI enhances traffic flow efficiency beyond fixed-time strategies.
• The study validates the potential of AI in mitigating urban traffic congestion.