基于树突神经网络的低轨卫星智能感知路由算法

LEO satellite intelligent-sensing routing algorithm based on a dendrite network

在低轨卫星网络中,卫星运行速度快、运行周期较短,星间链路动态变化.为了及时感知星间链路状态并选择正确的路由,提出一种基于树突神经网络的低轨卫星智能感知路由算法,通过卫星之间的可视性约束分析星间建链情况,实现星间链路态势感知;通过实时构造训练集,利用树突神经网络自动调整全局卫星网络链路的权值,进而优化传统迪杰斯特拉(Dijkstra)算法,实现星间链路质量感知,给出智能路由决策;通过周期性监测卫星网络拓扑,实时修正初始路由路径.仿真结果表明,基于树突神经网络的路由算法复杂度低,路径时延、时延抖动及丢包率均低于传统启发式路由算法和Dijkstra路由算法.

In a low-Earth orbit(LEO)satellite network,the satellite operation speed is high,the operation cycle is short,and intersatellite links change dynamically.To sense the intersatellite link state in time and select the correct route for an intelligent routing decision,a dendritic network-based intelligent-aware routing algorithm for LEO satellites is proposed in this paper.This algorithm divides the intersatellite link routing of an LEO satellite network into situation-aware,quality-aware,and routing-decision stages and establishes a routing policy framework with real-time correction capability from the source node to the destination.This approach overcomes the problems of the limited selection of routing paths from fixed labels of existing deep learning-based routing algorithms and the long convergence time of reinforcement learning-based routing algorithms.In the intersatellite link situational awareness stage,the intersatellite visibility of the entire LEO satellite network is periodically obtained by analyzing the constraint conditions of the intersatellite link establishment.In the intersatellite link quality perception stage,the final output of the probabilistic forwarding matrix based on the ant colony algorithm is used as the label of the training set,and the corresponding intersatellite link quality is evaluated using the probability value of the current node by selecting the next hop node.By changing the weight coefficients in the path cost function under different load states,more effective training set label data can be collected,which can be consequently used to improve the performance of the trained dendritic network.Moreover,the training set can be optimized in real-time through semi-supervised learning.The trained dendritic network is used to analyze and process the link state parameters,perceive the comprehensive service quality of the link,and output the evaluation value matrix of the next hop routing.It is also used to automatically adjust the weight of the global satellite network link.Meanwhile,the traditional Dijkstra algorithm is optimized to realize the quality perception of the intersatellite link.In the routing decision stage,the reciprocal of the evaluation value matrix is used as the adjacency matrix to pass the shortest-path algorithm.Then,the initial routing path between the source and destination nodes is obtained.Finally,the initial path is corrected via periodic monitoring to cope with the failure of the satellite node.The simulation results show that the routing algorithm based on the dendritic network has low computational complexity and fast convergence.The algorithm can determine the status of the intersatellite link establishment in time,assess the quality of the intersatellite link in real-time,and automatically avoid congested satellite nodes.Accordingly,its end-to-end path delay,delay jitter,and packet loss rate are lower than those of the traditional heuristic routing algorithm and Dijkstra routing algorithm.