A dynamic routing algorithm for VANET with graph neural networks and deep reinforcement learning

In vehicular ad hoc networks (VANET), the direct vehicle-to-vehicle communication provides a necessary supplement to the data transmission of intelligent transportation systems. However, the transient and volatility of VANET topology bring challenges to establishing an efficient and reliable end-to-end routing. To this end, a dynamic routing algorithm for VANET integrating graph neural networks and deep reinforcement learning is proposed. Firstly, the network topology is delineated according to the routing request, and graph features are extracted based on the routing establishment objective. Then, the routing relay selection problem is modelled as a Markov decision process, and the network topology information is learned using graph neural networks and solved using a deep Q-learning algorithm framework. Specifically, in order to better evaluate actions, a new fuzzy logic-based reward function is present. Simulation results show that the algorithm has better performance in terms of average end-to-end delay, hop count and packet delivery rate compared to other algorithms.