基于深度强化学习的无人机通信速率优化

UAV Communication Rate Optimization Based on Deep Reinforcement Learning

针对城市空对地模型中无人机与地面用户通信视线连接受阻的问题,提出了基于深度强化学习的无人机通信速率优化方案。利用智能反射面(reconfigurable intelligent surface,RIS)辅助无人机通信,采用双深度Q网络(double deep Q-Learning,DDQN)算法联合RIS相移和无人机的3D轨迹优化无人机的通信速率,在自建仿真平台上对该方案进行验证。结果表明:与RIS随机相移的DDQN方案、未部署RIS的DDQN方案及RIS相移优化的决斗深度Q网络方案相比,该方案在无人机飞行周期内的平均吞吐量,分别提高了38.61%、30.03%、53.97%。

Since the communication line of sight between unmanned aerial vehicles(UAVs)and ground users is blocked in the urban air-to-ground model,an optimization scheme of UAV communication rate based on deep reinforcement learning was proposed.The reconfigurable intelligent surface(RIS)was used to assist the UAV communication,and the double deep Q-learning(DDQN)algorithm was used to combine the RIS phase shift and the 3D trajectory of the UAV to optimize the communication rate of the UAV.The scheme was verified on the self-built simulation platform.The experimental results show that the average throughput of the proposed scheme in the UAV flight cycle is 38.61%,30.03%,and 53.97%higher than that of the DDQN scheme with RIS random phase shift,the DDQN scheme without RIS deployment,and the dueling DQN scheme optimized by RIS phase shift,respectively.