Title Study of Artificial Intelligence Traffic Signal Control - Urban Main Roads and Highway Interchange Areas Implementation
Dept Transportation Technology and Information Systems Division
Year 2024
Month 7
Price 280
Summary        In recent years, the rapid development of artificial intelligence in software and hardware technologies and its widespread application across various domains has led to significant progress. We can anticipate that future utilization of technologies such as artificial
intelligence, image recognition, information and communication technology (ICT), vehicle-to-everything (V2X), and 5G will play a crucial role in alleviating the pain points caused by incapable intelligent traffic signal control. Building upon the research project titled
“Exploration of Artificial Intelligence Vehicle-to-Everything Signal Control Models in Taiwan” conducted in 2022, this project aims to collaborate with local governments for on-site experimental testing. During the implementation of these tests, the artificial intelligence
signal control model can be continuously refined based on practical experience. We explore and establish a robust environment for reinforcement learning, introduce multi-objective reinforcement learning algorithms, and investigate the versatility of traffic flow
simulation models across various scenarios. The ultimate goal is to enhance the operational efficiency of artificial intelligence signal control models. Additionally, we address the challenges related to artificial intelligence signal control for freeway ramps. We prioritize
the construction of a simulated environment for interchange area signal coordination and the preliminary development of a decentralized AI signal coordination model using a multi-agent mechanism. Through these efforts, we aim to gradually strengthen Taiwan’s capabilities in artificial intelligence signal control and contribute to improving traffic efficiency and safety in urban environments under the context of artificial intelligence and vehicular networking.
       This project focuses on optimizing and enhancing artificial intelligence (AI) reinforcement learning models for traffic signal control. It can calculate the signal timing for the next phase in real-time based on traffic changes at each intersection. For example, in real-time, at the ‘Tai 86-19 Jia’ intersection in Tainan City, it calculates the green light duration for the next phase. In Taipei City, at the ‘Zhongshan North Road - Dexing East Road’ corridor, a single agent controls three intersections, dynamically adjusting the common signal cycle length and phase ratios to ensure continuous traffic flow. Field tests in Tainan and Taipei have shown that decisions based on different preferences can be made, and for most time periods, a set of preferences can be found that performs better than fixed-time controls.
       Additionally, the multi-task AI reinforcement learning model may not outperform models trained separately, even when trained with data from different time periods, but its overall performance is close. In the development of ramp metering and surface intersection
reinforcement learning signal coordination models for interchange areas, Yangmei Interchange in Taoyuan was selected as the testing site. A multi-signal control agent mechanism is introduced. Performance evaluation considers additional indicators related to driver
perception, and the ‘Centralized Training’ approach centralizes input data from ramps and surface roads to a single central agent. The trained machine learning model is then applied to construct a simulated experimental field at the Yangmei Interchange in Taoyuan. This reinforcement learning-based coordination model for interchange areas in Taoyuan will be further trained and tested in 2024.
Post date 2024/07/29
View count: 176