面向星座卫星高精度自主导航技术需求，设计了一种融合X射线脉冲星和星间定向观测信息的组合导航方法。通过X射线探测器获得脉冲到达时间观测量，照相观测星相机和星间链路设备获得星间相对位置矢量观测量，设计导航滤波器对观测量进行处理，估计参与导航的星座卫星的运动状态。针对地球星历误差影响组合导航性能的问题，将地球相对于太阳系质心的位置扩充为状态向量，设计了扩维扩展卡尔曼滤波器，利用敏感器观测量对导航所需的地球位置矢量进行实时估计，从而削弱地球星历误差的影响。进而，针对滤波器参数选取影响状态估计精度的问题，设计了一种Q学习扩展卡尔曼滤波器（QLEKF），主要思路是利用Q学习方法的决策能力，自适应地选择适当的滤波器参数以改善估计性能。数学仿真结果表明，所提方法能够有效减小地球星历误差对星座自主导航的影响，取得优于传统滤波算法的定位精度。

An integrated navigation method based on X-ray pulsars and inter-satellite measurements is designed for constellation satellite autonomous navigation with high-accuracy requirement. The X-ray detector is used to obtain the pulse arrival time. The relative position vector measurement between satellites is obtained from star cameras and inter-satellite links. A navigation filter is designed to process the obtained data， so as to estimate the kinematic states of the satellites that participate in the navigation. To cope with the problem that the ephemeris error of Earth affects the integrated navigation performance， the position vector of Earth relative to the solar system barycenter is augmented as a state vector， and an augmented extended Kalman filter is designed. The position vector of Earth for navigation is estimated in real time， such that the effect of the ephemeris error of Earth is suppressed. In addition， to cope with the problem that state estimation accuracy is affected by the choice of filtering parameters， a Q-Learning Extended Kalman Filter （QLEKF） is designed. The key idea is to select the appropriate filtering parameters adaptively with the Q-learning method， which has the decision-making ability. It is illustrated via numerical simulations that effect of the ephemeris error of Earth on the autonomous navigation performance of constellation is decreased efficiently using the proposed algorithm. The QLEKF is superior to traditional filtering algorithms in positioning accuracy.