摘要
针对巨型星座中卫星数量众多容易引发局部拥塞的问题,提出基于协作多智能体深度强化学习的巨型星座负载均衡算法。首先对巨型星座中的卫星进行分簇设计,实现巨型星座的分布式管理,降低网络管理开销。然后,利用Q-混合多智能体神经网络深度强化学习设计各卫星自主决策的路由规划方案,实现多传输任务的簇内协同。此外,提出基于自动编码器的簇状态压缩机制,提高多智能体深度强化学习的效率。仿真结果表明,所提算法相比于传统的单任务路由算法,传输成功率可提升40%以上,证明所提算法能够避免局部拥塞的发生,提高巨型星座的传输效率。
To overcome the local traffic congestion caused by the huge number of satellites in mega-constellation,a load balancing algorithm based on multi-agent deep reinforcement learning was proposed.Firstly,the satellites in the mega constellation were di-vided into clusters to perform the distributed management of the mega constellation,which could reduce the overhead of whole net-work.Then,based on the coordinated multi-agent deep reinforcement learning model,routing planning,which could be individually operated by satellites in the mega constellation,was designed to achieve the intra-cluster coordination.Additionally,a cluster state compression mechanism with autoencoder was proposed to compress the state space and improve the efficiency of multi-agent deep reinforcement learning.Finally,simulation results showed that compared with the traditional single-task routing algorithm,the pro-posed algorithm could increase the transmission success rate by more than 40%and the proposed algorithm could efficiently avoid local traffic congestion.
作者
罗树欣
张超
肖勇
刘建平
LUO Shuxin;ZHANG Chao;XIAO Yong;LIU Jianping(School of Information and Communication Engineering,Faculty of Electronic and Information Engineering,Xi’an Jiaotong University,Xi’an 710049,China;State Key Laboratory of Astronautic Dynamics,Xi’an Satellite Control Center,Xi’an 710043,China)
出处
《天地一体化信息网络》
2023年第4期49-60,共12页
Space-Integrated-Ground Information Networks
基金
国家重点研发计划资助项目(No.2020YFB1806102)
陕西省重点研发计划资助项目(No.2023-YBGY-251)。
关键词
巨型星座
多智能体深度强化学习
负载均衡
路由算法
mega constellation
multi-agent deep reinforcement learning
load balancing
routing algorithm