Research on Intelligent Control of Entrance Ramps for Networked Vehicles Based on Deep Reinforcement Learning

With the acceleration of urbanization, freeway traffic congestion is becoming increasingly serious, especially at entrance ramps, where the concentrated inflow of traffic often leads to increased traffic pressure on the mainline, affecting the overall access efficiency. In order to alleviate the ramp congestion problem, this paper proposes a deep reinforcement learning-based intelligent control method for entrance ramps of network-connected vehicles, which adopts Proximal Policy Optimization (PPO) algorithm to optimize the ramp vehicle flow and speed control strategy in real time by constructing a reinforcement learning control framework. In this paper, simulation experiments are conducted in different traffic density scenarios and compared with the traditional reinforcement learning algorithms DQN and A2C. The experimental results show that the PPO algorithm is able to converge quickly in low, medium and high traffic densities, significantly improve the cumulative reward value, and exhibit higher stability and superiority compared with other algorithms. The research in this paper not only provides an intelligent solution for ramp flow optimization, but also provides theoretical support and technical reference for the development of intelligent transportation system.