基于深度强化学习的无人机自组网路由算法  

Reinforcement learning based routing algorithm for UAV ad-hoc networks

作　　者：乔冠华 吴麒[1] 王翔 潘俊男 张易新 丁建 QIAO Guanhua;WU Qi;WANG Xiang;PAN Junnan;ZHANG Yixin;DING Jian(China Southwest Electronic Technology Research Institute,Chengdu 610036,P.R.China;School of Communication and Information Engineering,Chongqing University of Posts and Telecommunications,Chongqing 400065,P.R.China)

机构地区：[1]中国西南电子技术研究所,成都610036 [2]重庆邮电大学通信与信息工程学院,重庆400065

出　　处：《重庆邮电大学学报（自然科学版）》2023年第2期335-342,共8页Journal of Chongqing University of Posts and Telecommunications(Natural Science Edition)

基　　金：中国电子科技集团公司第十研究所基金(SSJ-2125-XY-002)。

摘　　要：针对无人机自组网节点密度大、拓扑变换频繁,导致移动自组网复杂的问题,提出了一种基于深度强化学习(deep-reinforcement learning,DRL)的分布式无人机自组网路由算法。利用DRL感知学习无人机特征,使节点不断与环境交互、探索学习最优行动(路由)策略;通过存储经验知识,维护端到端路由,赋予无人机网络智能化重构和快速修复的能力,从而提高路径的稳定性,降低路由建立和维护开销,增强网络的鲁棒性能。仿真结果表明,提出的算法具有较好的收敛性能;在路由修复时间、端到端时延,以及网络适应性、扩展性方面都优于传统的路由算法。Aiming at the problem that the density of nodes in UAV ad hoc network is large and the topology transformation is frequent,which leads to the complexity of mobile ad hoc network,a distributed UAV ad hoc network routing algorithm based on deep reinforcement learning(DRL)is proposed.Using DRL perception to learn the characteristics of UAV,the node can continuously interact with the environment and explore and learn the optimal action(routing)strategy.By storing experience knowledge and maintaining end-to-end routes,the UAV network is endowed with the ability of intelligent reconfiguration and rapid repair,thus improving the stability of the path,reducing the cost of route establishment and maintenance,and enhancing the robustness of the network.Simulation results show that the proposed algorithm has good convergence performance.It is superior to traditional routing algorithms in routing repair time,end-to-end delay,network adaptability and scalability.

关 键 词：无人机自组网 深度强化学习 路由算法 路由修复 

分 类 号：TN929[电子电信—通信与信息系统]