| Summary |
Traffic jam problems have been a great challenge for the urban traffic management, from traffic-flow data collection to signal-control strategy development, which depletes many human and material resources. As the integration of artificial intelligence (AI) in systems grows, the utilization of AI deep learning neural networks to tackle traffic problems has become an important topic. Hence, this project
has collected important traffic data to control signals by means of AI. The Leye–Shijia intersection and Leye-Dongying intersection in Taichung City were chosen as the testing sites of the project. Specifically, this project collected traffic data through AI image recognition equipment and used it to establish a framework of AI signal control. The trained AI signal control model can select and implement an appropriate timing plan based on real-time traffic flow.
Regarding AI signal control models, reinforcement learning requires considerable trial-and-error time to find the best timing plan, which engenders few real case studies. Accordingly, this study uses traffic signal timing and phase optimization software, PaSO, to
generate the best timing plan of arterial road progression based on the scope of the test, as well as utilizes a time-of-day parameter to
carry out AI reinforcement learning. Finally, the trained AI signal control model is applied to the testing sites. Furthermore, this study
initially adopts micro traffic simulation software, SUMO, to create a simulation environment of the testing sites and to generate the
simulation performance of AI signal controls, and finally test the proposed method in the testing sites. The results of the tests reveal that
the proposed method can enhance the traffic flow performance by more than 10% during peak hours. |
