Connected vehicles' dynamic route planning based on reinforcement learning

Real-time route planning is a challenging work for the Internet of Vehicles (IoV) systems. Some people have used swarm intelligence algorithms to schedule routes for connected vehicles. These algorithms are always sensitive to the setting of initial parameters. These days, reinforcement learning algorithms to some extent overcome this drawback. However, previous research based on reinforcement learning has not taken into account the real-time variation in the number of vehicles on each lane. In this work, a new reinforcement learning algorithm, named dynamical-weighted Q-learning (DWQL), is proposed. This approach plans the routes for connected vehicles by introducing dynamic weighting factors into Q-learning algorithm. By employing dynamic weighting factors, the algorithm selects the optimal route for connected vehicles based on their current positions as well as congestion status of the accessible roads. The experimental results show that compared to existing algorithms, the DWQL algorithm proposed in this paper can effectively reduce the duration, time loss, waiting time, economic cost, and improve the traffic efficiency of the IoV system.