Traffic signal control plays an important role in regulating road network traffic. Now the dominant signal control models are based on reinforcement learning. The state of the model is generally defined as the position and speed of the vehicle in the road grid and the state of the signal light and the action space (traffic signal plans) of the model is generally defined artificially in advance, but this action space definition is relatively rough and cannot Leveraging the full capabilities of the intersection, so this paper considers the combination of deep learning and reinforcement learning to improve the performance of the traffic signal control system. First, we use DTSE method to obtain the state of the road traffic. Based on this, we construct a recurrent neural network to produce the traffic signal’s action. Then we optimize the traditional Q-learning method. The feasibility of our model can be proven through experiments. To sum up, the research in this paper introduces new ideas in the field of signal control. By combining the recurrent neural network with reinforcement learning methods, we introduce new composite models for the field of signal control, which is expected to make a major breakthrough in practical applications.
Deep Reinforcement Learning-Based Traffic Signal Control
24th COTA International Conference of Transportation Professionals ; 2024 ; Shenzhen, China
CICTP 2024 ; 1675-1685
2024-12-11
Conference paper
Electronic Resource
English
Deep Reinforcement Learning-based Traffic Signal Control
IEEE | 2023
|Traffic signal control method based on deep reinforcement learning
European Patent Office | 2021
|Reinforcement learning-based traffic signal control
IEEE | 2024
|