Multi-Agent Reinforcement Learning- Based Resource Management for V2X Communication

Cellular vehicle-to-everything (V2X) communication is essential to support future diverse vehicular applications. However, due to the dynamic characteristics of vehicles, resource management faces huge challenges in V2X communication. In this paper, the optimization problem of the comprehensive efficiency for V2X communication network is established. Considering the non-convexity of the optimization problem, this paper ulitizes the markov decision process (MDP) to solve the optimization problem. The MDP is formulated with the design of state, action, and reward function for vehicle -to-vehicle links. Then, a multiagent deep Q network (MADQN) method is proposed to improve the comprehensive efficiency of V2X communication network. Simulation results show that the MADQN method outperforms other methods on performance with the higher comprehensive efficiency of V2X communication network.