一种基于强化学习的低轨卫星网络智能路由算法  

An Intelligent Routing Algorithm for LEO Satellite Network Based on Reinforcement Learning

作　　者：左珮良 王晨 蒋华[1,2] ZUO Peiliang;WANG Chen;JIANG Hua(Beijing Electronic Science and Technology Institute,Bejing 100070,P.R.China;Xidian University,Xi'an 710071,P.R.China)

机构地区：[1]北京电子科技学院,北京市100070 [2]西安电子科技大学,西安市710071

出　　处：《北京电子科技学院学报》2022年第2期35-43,共9页Journal of Beijing Electronic Science And Technology Institute

基　　金：国家自然科学基金(项目编号:62001251,62001252)。

摘　　要：具备低延迟、广覆盖、快速机动等优点的低轨卫星网络是第六代移动通信技术(6G)和空天地一体化愿景的重要支撑,然而随着各国投入的增加,低轨卫星呈现出数量类型众多、轨道纷杂交错、载荷用频和处理能力相异的现象,使得传统卫星路由算法在通信开销和更新效率上出现瓶颈。鉴于此,探索性地提出一种适用于动态低轨卫星网络的卫星节点智能选径算法,节点能够依靠深度强化学习训练得到的模型,依据周围节点的空间位置、相互间距和可用带宽信息,自适应地选择回传链路。仿真和分析表明,所提方法具有较好的收敛性和泛化能力,相比于典型的分布式路由算法,具备更好的时延特性。With the advantages of low delay,wide coverage and fast mobility,LEO satellite network(SN)becomes an important support for the sixth generation mobile communication technology(6 G)and the vision of the space-air-ground integrated network(SAGIN).However,with the increase of continuous investment,LEO satellites suffers from problems of enormous types,complicated orbits,inconsistent frequency and processing capacity,which induces an bottleneck of communication overhead and update efficiency to the traditional routing algorithms.To address the problems,an intelligent path selection algorithm for the dynamic LEO SN is proposed.With the model obtained by using the deep reinforcement learning,satellite nodes are able to choose the return link adaptively according to the spatial location,mutual distance and available bandwidth of surrounding nodes.Simulation and analysis indicate that the proposed algorithm exhibits good convergence and generalization ability,and has better delay characteristics than the typical distributed routing algorithms.

关 键 词：低轨卫星网络 深度强化学习 分布式路由 智能路由 空天地一体化 

分 类 号：TM344.1[电气工程—电机]