Improving traffic management efficiency through reinforcement learning-based traffic signal control and citywide transit simulation

Traffic congestion reduces productivity and harms the environment. Enhancing traffic signal control and public transportation are effective solutions. However, prior research has limitations stemming from the absence of real-time reliable data. Recent computer vision systems have made collecting traffic data easier. This thesis explores leveraging these data sources to enhance existing traffic signal controls (TSCs) and citywide transit simulations. For TSC, a comprehensive framework that facilitates rapid prototyping of reinforcement learning (RL) and an automatic feature engineering method are proposed. Additionally, RL techniques are implemented to a digital twin of Chattanooga smart corridor. Regarding transit simulations, a toolkit for calibrating large-scale simulations and an efficient solution for simulating changes in transit system settings are developed. Finally, we delve into a fundamental question of optimization for training neural networks and demonstrate that a novel approach using Neuroevolution outperforms Gradient Descent methods.

Grantor dc:publisher

University of Tennessee at Chattanooga

Author dc:creator

• Tran, Viet Toan

Contributors dc:contributor

• Sartipi, Mina
• Liang, Yu; Wu, Dalei
• College of Engineering and Computer Science

• Electronic traffic controls
• Reinforcement learning
• Urban transportation--Computer simulation

Licence

• http://rightsstatements.org/vocab/InC/1.0/

Language dc:language

English, eng

Repository record dc:identifier

https://scholar.utc.edu/theses/823

OAI identifier oai:identifier

oai:scholar.utc.edu:theses-2003