Traffic congestion remains a major issue in urban areas, leading to environmental deterioration through increased noise and air pollution. Existing traffic management solutions often fall short in adapting to the dynamic nature of urban traffic patterns. Adaptive traffic control systems utilizing Reinforcement Learning (RL) offer a promising alternative by adjusting traffic signals based on real-time traffic conditions in each lane to alleviate congestion. In this paper, we propose an RL-based approach leveraging Deep Q-Networks (DQN) to minimize the cumulative waiting time of vehicles at intersections. Our method involves training a deep neural network to predict Q-values for various traffic signal actions, which are then used to optimize traffic light changes. The simulation environment is set up using SUMO (Simulation of Urban Mobility) for network and traffic generation, with TraCI (Traffic Control Interface) providing real-time traffic data. Our approach demonstrates significant improvements over the traditional methods. Additionally, we include visualizations of the simulation results to illustrate the effectiveness of our RL-based traffic control system.
Adaptive Traffic Signal Control System Using Deep Reinforcement Learning
07.12.2024
431412 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Wiley | 2020
|IET | 2021
|