A Discussion of the Artificial Intelligence V2X Signal Control Model
- Date：2023-09-20
- Update：2023-09-20
- Department：IOT
Plan Overview
In recent years, artificial intelligence has progressed by leaps and bounds, with the rapid development of numerous software and hardware field applications. Future applications of artificial intelligence, image recognition, Information and Communication Technology (ICT), vehicle-to-vehicle (V2X), and 5G technology are especially helpful for alleviating the public’s transport pain point as a result of insufficiently intelligent traffic signal control. The Institute of Transportation conducted a study titled "Proposal for the efficiency of intelligent transportation service series and the safety of V2X." Prioritized for discussion are Taiwan's urban transportation signal control and V2X integration, which provide dynamic information of real-time signals to various vehicles to create a safe, priority at intersections, energy-efficient road traffic environment. The Ministry of Transportation and Communications' "Proposal for an Intelligent Transportation Open Test Field in Danhai New Town" is founded on the Ministry of Transportation's V2X research series. In addition to addressing signal control and cellular vehicle-to-everything (C-V2X) integration protocol needs, the SAE standards serve as a reference for the development of the "Standards for Signal Controller and V2X Roadside ICT." In addition, "Research on Applied Artificial Intelligence in Transportation Data Collection and Signal Control" planned to implement Imitation Learning (IL) and signal "cycle"-based artificial intelligence reinforced learning (RL) signal control by 2020. In Taoyuan City, there are also domestic academic research and implementation cases. This plan has imported artificial intelligence to strengthen learning depth and confirm the application of “model-free” and V2X technology in traffic signal control under Deep Deterministic Policy Gradient (DDPG) and Proximal Policy Optimization (PPO) in traffic signal control. In addition, through actual traffic flow data at the experimental fields, training learning and quantitative performance analysis have been carried out in a simulation environment.
Research Results
- A traffic survey, simulation model construction, calibration, and estimation were completed in collaboration with Taipei City Government, Tainan City Government, and Kaohsiung City Government for three experimental fields: "Jhongshan N. Road-Dexing E. Road" multi-intersection main line, "Provincial Highway 86-19 Jia" single-intersection, and "Tai 88 Fengshan Exit" (Guopi Road-Fongding Road). Furthermore, artificial intelligence has improved the learning signal control model's learning training, simulation test, and performance analysis.
- According to simulation test results, artificial intelligence strengthened learning (DDPG and PPO) based on different field characteristics can improve existing traffic performance to varying degrees, indicating DDPG's potential and feasibility in strengthening learning signal control.
- To extract V2X data characteristics, which serve as strong learning data input, the convolutional neural network (CNN) structure is used. The experimental results show that V2X data has the potential to improve learning signal control data.
Result Promotion and Benefits
The county and city governments, academia, and industry were invited to promote the research results in this plan through result sharing on November 18 and 21, 2022. On November 22, 2022, education training was held.
Summary of Research Results
Taipei City Experimental Field and Learning Signal Control Model Strengthened through Artificial Intelligence and Training
Research Result Report
- Study of Artificial Intelligence and Vehicle-to-Everything Traffic Signal Control Model in Taiwan (Published in May 2023).