Reinforcement learning-based route planning method for connected vehicles considering real-time road traffic information

The economy, environment, and resident travel are significantly impacted by traffic congestion, which is a prevalent issue in urban traffic systems. Route planning is considered an effective method to address traffic congestion and enhance travel efficiency. Existing route planning methods have overlooked the impact of real-time traffic information on the timeliness of optimal routes. In this paper, an improved bureau of public roads-Q learning (IBPR-QL) control model considering real-time road traffic information is proposed for route planning of connected vehicles in cooperative vehicle-road scenarios. Firstly, the improved bureau of public roads (IBPR) function is enhanced through the utilization of speed state staging methodology, enabling adaptive adjustment of road priorities across different speed scenarios. Secondly, the IBPR function is incorporated into the Q-learning (QL) action value evaluation process to enhance model learning efficiency. Thirdly, predictive travel coefficient is proposed to mitigate congestion drift-induced oscillation phenomenon in road networks and enhance overall travel efficiency. Finally, road network datasets from four economically developed cities in China are utilized to construct a simulated road scene. The simulation results indicate that, compared to the classical QL method, IBPR-QL has a wider range of applicability across scenarios, effectively reducing travel time by 7.96%. In terms of alleviating the impact of congestion drift, the overall road network efficiency can be improved by 9.35%.