Paper FR-LA-T39.3
Guo, Zhiyuan (Nanyang Technological University), Yao, Jiarong (Nanyang Technological University), Su, Rong (Nanyang Technological University)
A Centralized Adaptive Traffic Signal Control Method for Large-Scale Network Based on SVD-Enhanced Deep Q-Network
Scheduled for presentation during the Regular Session "S39c-Data-Driven Optimization in Intelligent Transportation Systems" (FR-LA-T39), Friday, November 21, 2025,
16:40−17:00, Coolangata 3
2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia
This information is tentative and subject to change. Compiled on October 18, 2025
|
|
Keywords AI, Machine Learning for Dynamic Traffic Signal Control and Optimization, Cyber-Physical Systems for Real-time Traffic Monitoring and Control
Abstract
Current signal control methods based on deep reinforcement learning (DRL) have become an important research branch of signal control methods yet most studies are targeted at single intersection or use a distributed control framework for network scenario. Aimed at adaptive traffic signal control in large-scale urban road networks, this paper proposed a centralized control method based on a deep Q-network (DQN) considering the whole network as a control agent. The generalization ability of deep learning in high-dimensional features is further enhanced by a singular value decomposition (SVD) algorithm together with reduction in computational complexity. Simulation evaluation was conducted based on the Singapore Jurong network, with six baseline methods for horizontal comparison. Results show that the centralized DQN controller outperforms all six benchmarks, especially with a significant improvement of nearly 56% in total waiting time and 11% in average speed, respectively, as compared with the fixed time signal control scheme. traffic efficiency, achieving higher average speed, shorter waiting time, and less time loss. Even when compared with a centralized DQN benchmark, the proposed method performs better with a reduction of 20.84% in computation time, meanwhile outperforming in control performance, demonstrating the effectiveness and application potential for real-time, large-scale network signal control optimization.
|
|