针对航天器集群在轨自主导航时全球卫星导航系统（Global Navigation Satellite System, GNSS）短期拒止情况下的导航发散问题，引入航天器间基于数据链（Data Link, DL）的相对测距系统及神经网络辅助系统，提升拒止期间导航品质。考虑到INS、GNSS及DL系统采样频率不同，本文采用具有即插即用特点的因子图（Factor Graph, FG）架构实现INS/GNSS/DL的多源测量信息的动态融合。为避免数据链测距信息解算时的误差放大，设计基于旋转矩阵的相对测距信息解算方法；设计基于惯性积分状态及历史测量残差的复合故障诊断环节，结合因子图即插即用特点，实现GNSS故障及DL故障的检测与隔离；此外，针对GNSS拒止情况，设计广义回归神经网络（General Regression Neural Network，GRNN）与Elman神经网络（Elman Neural Network，ENN）组合的辅助系统，对历史时刻GNSS与INS导航数据之间的潜在关系进行拟合，并对缺失GNSS导航数据进行在线预测与补偿。将GNSS预测数据与INS/DL数据进一步融合，获得GNSS拒止情况下的导航结果。仿真结果表明，组合使用基于数据链的相对测距信息及GRNN-ENN神经网络辅助系统能有效缓解GNSS拒止情况下的导航发散情况。

Aiming at the problems of navigation dispersion in the Global Navigation Satellite System (GNSS) short-term denial scenario during the on-orbit autonomous navigation of spacecraft clusters, we introduce a data-link-based (DL) relative ranging between spacecraft and a neural network auxiliary system to improve the navigation effect. Considering the different sampling frequencies of INS, GNSS, and DL systems, this paper adopts the factor graph (FG) architecture, which has the feature of plug-and-play, to realize the dynamic fusion of multi-source measurement information of INS/GNSS/DL. In order to avoid the error amplification when solving the ranging information of the data links, a relative ranging information solving method based on the rotation matrix is designed. A composite fault diagnosis mechanism based on the inertial integral state and the historical measurement residuals is designed, which combines with the plug-and-play characteristics of the FG to realize the detection and isolation of the GNSS faults and DL faults. Moreover, for the GNSS denial case, an auxiliary system combining General Regression Neural Network (GRNN) and Elman Neural Network (ENN) is designed to fit the potential relationship between GNSS and INS navigation data at historical moments and provide an online prediction and compensation of missed GNSS navigation data. The predicted GNSS data are further fused with the INS/DL data to obtain the navigation results under the GNSS denial case. The simulation results show that the introduction of datalink-based relative ranging information and the GRNN-ENN neural network-assisted system can effectively mitigate the navigation divergence in the GNSS denial case.