Publisher
University of Tennessee at Chattanooga
Place of Publication
Chattanooga (Tenn.)
Abstract
Machine learning (ML), especially reinforcement learning (RL), has garnered significant attention for optimizing traffic signal control in intelligent transportation systems. However, existing ML approaches face scalability and adaptability challenges, especially in large traffic networks. This paper proposes an innovative solution by integrating decentralized graph-based multi-agent reinforcement learning (DGMARL) with a Digital Twin (DT) to enhance traffic signal optimization, aiming to reduce traffic congestion and network-wide fuel consumption associated with vehicle stops and delays. DGMARL agents learn traffic state patterns and make informed decisions on traffic signal control, further facilitated by the Digital Twin module simulating real-time traffic behaviors. Evaluation utilized PTV-Vissim, a microscopic traffic simulation platform, focusing on the MLK Smart Corridor in Chattanooga, Tennessee. Comparative analysis against an actuated signal control baseline showed significant improvements, with a remarkable 55.38% reduction in Eco_PI over 24 hours. In a PM-peak-hour scenario, the average Eco_PI reduction reached 38.94%. These findings underscore the effectiveness of the proposed approach.
Document Type
presentations
Language
English
Rights
http://rightsstatements.org/vocab/InC/1.0/
License
http://creativecommons.org/licenses/by/4.0/
Recommended Citation
Kumarasamy, Vijayalakshmi; Jairam Saroj, Abhilasha; Liang, Yu; Wu, Dalei; P. Hunter, Michael; Guin, Angshuman; and Sartipi, Mina, "Integration of Decentralized Graph-Based Multi-Agent Reinforcement Learning with Digital Twin for Traffic Signal Optimization". ReSEARCH Dialogues Conference proceedings. https://scholar.utc.edu/research-dialogues/2024/Proceedings/16.
Integration of Decentralized Graph-Based Multi-Agent Reinforcement Learning with Digital Twin for Traffic Signal Optimization
Machine learning (ML), especially reinforcement learning (RL), has garnered significant attention for optimizing traffic signal control in intelligent transportation systems. However, existing ML approaches face scalability and adaptability challenges, especially in large traffic networks. This paper proposes an innovative solution by integrating decentralized graph-based multi-agent reinforcement learning (DGMARL) with a Digital Twin (DT) to enhance traffic signal optimization, aiming to reduce traffic congestion and network-wide fuel consumption associated with vehicle stops and delays. DGMARL agents learn traffic state patterns and make informed decisions on traffic signal control, further facilitated by the Digital Twin module simulating real-time traffic behaviors. Evaluation utilized PTV-Vissim, a microscopic traffic simulation platform, focusing on the MLK Smart Corridor in Chattanooga, Tennessee. Comparative analysis against an actuated signal control baseline showed significant improvements, with a remarkable 55.38% reduction in Eco_PI over 24 hours. In a PM-peak-hour scenario, the average Eco_PI reduction reached 38.94%. These findings underscore the effectiveness of the proposed approach.