基于深度强化学习的低轨卫星互联网微代价拓扑控制与重构研究

This project takes the dynamics of low-orbit satellite Internet topology as the research object, uses the complex network theory to study the correlation between the low-orbit satellite Internet topology features and routing performance optimization, and builds the relational mapping for topology feature parameters and route optimization based on deep reinforcement learning，that is a theoretical basis to form a micro-cost topology reconstruction algorithm for routing optimization. At the same time, the local optimal solution heuristic algorithm which shortens the reconstruction delay is studied, and the k-core decomposition method based on complex network is used to search for the largest node set or edge set of topological feature parameters in the network, then the global topology optimization is achieved. we build a virtualized and scalable satellite Internet environment, simulate the physical and topological and link status of the integrated network, and verify the correlation between the dynamic link connection (topology) and route optimization and Logical topology reconstruction caused by link failure. The satellite Internet topology control is designed to realize the cooperation between perception and prediction. It demonstrates the advantages of using complex network theory for topology micro-cost reconstruction, and finally complements the low-orbit satellite Internet coverage and transmission performance.

本项目以低轨卫星互联网拓扑动态性为研究对象，利用复杂网络理论研究低轨卫星互联网拓扑特征与路由性能优化之间的关联性，构建基于深度强化学习的拓扑特征参数与路由优化之间的关系映射，并以此为理论依据形成面向路由优化的微代价拓扑重构算法设计。进而研究缩短重构延迟的局部最优解启发式算法，并探索基于复杂网络中的k-核分解方法以获取网络中拓扑特征参数变化幅度最大节点集或边集，实现全局拓扑最优化。在此基础上，构建支持虚拟化、可扩展的卫星互联网环境演示平台，模拟全天候、全空时的天地一体化网络物理拓扑和链路状况，验证动态链路连接关系（拓扑结构）和路由优化的关联性以及节点/链路失效所带来的逻辑拓扑重构。本项目旨在实现感知与预测协作的卫星互联网拓扑控制，论证利用复杂网络理论进行拓扑微代价重构的优势，最终实现低轨卫星互联网覆盖范围和传输性能之间的互补。

本项目以低轨卫星互联网拓扑动态性为研究对象，利用复杂网络理论研究低轨卫星互联网拓扑特征与路由性能优化之间的关联性，并以此为理论依据形成面向路由优化的微代价拓扑重构算法设计 。项目提出了空天地一体化融合网络弹性抗毁自愈技术，实现天地异构网络拓扑的弹性适变和稳定高效。提出了基于位置的天地一体化网络高效寻址机制，将用户地址与移动性相解耦，解决真实世界与虚拟网络基于统一时空基准的定位和寻址，而且也从根本上解决了路由收敛（稳定性）和空间网络拓扑动态时变和连接关系不确定之间的矛盾，实现卫星网络与地面网络的异构高效融合与全球无缝互联。设计卫星互联网试验验仿真平台，基于虚拟化技术，对天地融合网络中的海量节点进行仿真，同时支持网络协议栈的模拟，可动态加载真实的网络协议与网络流量，构建大规模数据驱动、真实可信、具备星座时空特性的融合网络试验环境。在学科顶级、重要学术期刊 IEEE JSAC、IEEE WCL、IEEE JIOT、通信学报和学科旗舰会议IEEE WCNC和ICCC等上发表 SCI/EI 学术论文 9篇。针对拓扑调整和路由优化理论和方法，申请国家发明专利 5 项，已授权 2项，完成预期目标。