Multi-objective deep reinforcement learning approach for adaptive traffic signal control system with concurrent...

Area of Science:

• Intelligent Transportation Systems
• Artificial Intelligence in Traffic Management
• Sustainable Urban Mobility

Background:

• Traditional adaptive traffic signal control (ATSC) systems often prioritize traffic efficiency, neglecting crucial aspects like safety and environmental impact.
• Existing ATSC methods struggle to dynamically adapt to real-time traffic fluctuations and complex urban environments.
• There is a need for advanced control strategies that can holistically optimize traffic flow, safety, and decarbonization.

Purpose of the Study:

• To introduce a novel multi-objective adaptive traffic signal control (ATSC) approach using deep reinforcement learning (DRL).
• To develop a DRL-based ATSC algorithm capable of simultaneously optimizing traffic safety, efficiency, and decarbonization.
• To evaluate the performance of the proposed DRL-ATSC system against traditional methods in a simulated urban environment.

Main Methods:

• Implementation of a DRL algorithm, specifically the Dueling Double Deep Q Network (D3QN) framework, for ATSC.
• Simulation of a traffic intersection in Changsha, China, to test the proposed DRL-ATSC algorithm.
• Comparative analysis of the DRL-ATSC system against traditional ATSC and efficiency-focused ATSC algorithms.

Main Results:

• The DRL-ATSC algorithm achieved over a 16% reduction in traffic conflicts and a 4% decrease in carbon emissions compared to baseline methods.
• A significant 18% reduction in waiting times was observed compared to traditional ATSC, with a minimal 0.64% increase compared to a DRL algorithm focused solely on efficiency.
• The proposed system demonstrated superior performance across all three objectives (safety, efficiency, decarbonization) under high traffic demand scenarios.

Conclusions:

• The novel DRL-based ATSC approach effectively balances multiple traffic control objectives, including safety, efficiency, and decarbonization.
• The D3QN framework provides a robust foundation for developing adaptive traffic signal control systems that outperform traditional methods.
• This research offers a practical and advanced solution for optimizing traffic signal control in real-world, dynamic traffic conditions, contributing to smarter and more sustainable urban transportation.