针对5G新空口-车联网(New Radio-Vehicle to Everything,NR-V2X)场景下车对基础设施(Vehicle to Infrastructure,V2I)和车对车(Vehicle to Vehicle,V2V)共享上行通信链路的频谱资源分配问题,提出了一种联邦-多智能体深度Q网络(Federated Learning-Multi-Agent Deep Q Network,FL-MADQN)算法.该分布式算法中,每个车辆用户作为一个智能体,根据获取的本地信道状态信息,以网络信道容量最佳为目标函数,采用DQN算法训练学习本地网络模型.采用联邦学习加快以及稳定各智能体网络模型训练的收敛速度,即将各智能体的本地模型上传至基站进行聚合形成全局模型,再将全局模型下发至各智能体更新本地模型.仿真结果表明:与传统分布式多智能体DQN算法相比,所提出的方案具有更快的模型收敛速度,并且当车辆用户数增大时仍然保证V2V链路的通信效率以及V2I链路的信道容量.
Spectrum resource allocation for NR-V2X in-vehicle communication based on FL-MADQN algorithm
To address the spectrum resource allocation problem of shared uplink between vehicle-to-infrastructure(V2I)and vehicle-to-vehicle(V2V)in 5G New Radio-Vehicle to Everything(NR-V2X)scenario.A Federated Learning-Multi-Agent Deep Q Network(FL-MADQN)algorithm is proposed.In the decentralized algorithm,each vehicle user is treated as an agent to learn the local network model using the DQN algorithm based on the obtained local channel state information and the optimal network channel capacity as the objective function.Federated learning is used to speed up and stabilize the convergence rate of each agent ′s model training.The local model of each agent is uploaded to the base station for aggregation to form the global model,and then the global model is distributed to each agent to update the local model.Simulation results show that this scheme has a faster model convergence speed compared with the traditional distributed multi-agent DQN algorithm,and the communication efficiency of the V2V link and the channel capacity of the V2I link are still guaranteed when the number of vehicle users increases.
国家科技期刊平台
NSTL
万方数据
维普
