中文摘要：

导航卫星的自主定轨是提高卫星导航系统生存能力的一个重要手段,在解决导航星座自主定轨中涉及到高精度的轨道预报,提高轨道预报精度对于自主定轨精度有着重要意义。针对利用动力学模型得到的预报轨道随时间推移精度衰减较快的问题,本文提出了一种改进北斗导航卫星中长期轨道预报精度的新方法。利用神经网络作为建立预报模型的工具,在动力学模型的基础上建立神经网络模型,通过对历史时刻预报误差的学习及训练,掌握其变化规律,再用于补偿和改进当前时刻的预报轨道,以达到提高预报精度的目的。本文制定了导航卫星轨道中长期预报方案,并利用实测数据进行了实验分析,结果表明,采用神经网络模型补偿预报轨道误差时,不同卫星在不同初始时刻下的改进效果是不同的。预报15d导航卫星的轨道精度由318m提高至19m,预报30d轨道精度由1 757m提高至49m。预报15d、30d轨道改进幅度分别为41%-80%、32%-88%。

英文摘要：

Autonomous orbit determination of satellites is important to improve the availability of a satellite navigation system, and depends on high-precision orbit predictions. The orbit predicted with a dynamics model has the problem of high dilution. To solve this problem, a method is proposed to improve long-term orbit predictions for BeiDou satellites Based on an artificial neural network （ANN） model. We developed an ANN model based on the dynamics model, in order to determine the variation characteristics in the orbit prediction errors by learning and training historical orbit prediction errors. We used this ANN model to improve the accuracy of orbit predictions by estimating and correcting prediction errors. For medium-term and long-term orbit predictions, experimental results showed that orbit prediction errors after the application of the proposed ANN model were less than those based on the dynamic model. The effectiveness of the improvements varies with different satellites and initial epochs. The error of orbit predictions for the 15-day prediction was reduced to 19 m from 318 m; and, for the 30-day prediction, was reduced to 49 m from 1 757 m. The improvement ratios for the 15-day and 30-day predictions were 41%-80% and 32%-88%, respectively.