Trajectory reconstruction for nanosatellite in very low Earth orbit using machine learning
Trajectory reconstruction for nanosatellite in very low Earth orbit using machine learning
复制标题
DOI：
10.1016/j.actaastro.2022.02.010
复制
发表时间：
2022-03-03
影响因子：
3.5
通讯作者：
Yamada, Kazuhiko
中科院分区：
文献类型：
--
作者：
Takahashi, Yusuke;Saito, Masahiro;Yamada, Kazuhiko
微卫星和纳卫星被发射到极低地球轨道（VLEO）。然而，VLEO中的大气密度不明确，这使得预测卫星的行为和寿命变得困难。在一颗纳卫星的在轨停留和再入任务期间，利用铱星卫星网络在400至100千米的高度进行了基于全球定位系统（GPS）的定位。这可以为估算VLEO中的大气密度提供有用的见解。然而，GPS数据的间歇性使这些估算变得困难。我们通过将GPS数据用作训练数据进行高斯过程回归（GPR），并从稀疏的定位数据中重构连续数据。所提出的重构方法在核函数的选择上具有灵活性。利用获得的GPR结果、非惯性坐标系下三自由度质点系统的运动方程模拟以及贝叶斯优化，重构了卫星的速度剖面。验证了GPR的预测性能以及轨迹重构的特性。
Micro-and nanosatellites are launched into very low Earth orbit (VLEO). However, the atmospheric density in VLEO is unclear, making it difficult to predict the satellite behavior and lifetime. During a stay-in-orbit and reentry mission of a nanosatellite, global positioning system (GPS)-based positioning was performed at 400 to 100 km using the Iridium satellite network. This can provide useful insights for estimating the atmospheric density in VLEO. However, the intermittency of the GPS data made these estimations difficult. We performed Gaussian process regression (GPR) by using GPS data as training data and reconstructed continuous data from the sparse positioning data. The proposed reconstruction method was flexible in selection of the kernel function. The velocity profile of the satellite was reconstructed using the obtained GPR results, equation of motion simulation of a three-degree-of-freedom mass-point system in a non-inertial coordinate system, and Bayesian optimization. The prediction performance and characteristics of trajectory reconstruction by GPR were verified.