Considering the intricate traffic flow dynamics between the mainline and entrance ramp areas of highways, a ramp control model based on deep reinforcement learning (DRL) is proposed. Leveraging the characteristics of traffic flow, this study builds upon existing reinforcement learning (RL) ramp control models and optimizes the design of the state space, action space, and reward function. Utilizing the PyCharm platform, SUMO simulation software is adopted for secondary development. Simulation experiments are conducted in various scenarios involving different congestion durations and demand distributions. Under identical road network conditions, four distinct control methods are devised: the no-control (NC) scenario, the ALINEA ramp control method, the RL control method, and the proposed DRL method. The experimental outcomes indicate that, compared to the NC method, all three alternative methods substantially reduce the average travel time. The proposed DRL method exhibits an average travel time reduction of 12.77% compared to the ALINEA method and a 5.22% reduction compared to the RL method. Furthermore, the proposed DRL method achieves a reduction of approximately 39.01% in the ramp queue length compared to the ALINEA method.
A Highway Entrance Ramp Control Method Based on Deep Reinforcement Learning
2023-12-14
320872 byte
Conference paper
Electronic Resource
English
Vehicle ramp entrance confluence control method based on deep reinforcement learning
European Patent Office | 2023
|Highway On-Ramp Truck Platooning Based on Deep Reinforcement Learning
Transportation Research Record | 2024
|AUTONOMOUS VEHICLE SOCIAL BEHAVIOR FOR HIGHWAY ENTRANCE RAMP MANAGEMENT
British Library Conference Proceedings | 2013
|Reinforcement Learning With Model Predictive Control for Highway Ramp Metering
IEEE | 2025
|