摘要
低轨卫星在增强全球导航定位系统中具有重要意义,其超稳振荡器(USO)钟差的高精度实时预测是实现厘米级乃至亚厘米级定位精度的关键。然而,传统多项式外推与灰色模型等方法难以兼顾钟差序列的周期性与非线性特征,在短期或长期预报中均存在明显不足。文中以GRACE-FO卫星USO钟差数据为研究对象,首先采用改进的中位数绝对偏差方法进行粗差检测与预处理,再通过基于局部加权回归的季节-趋势分解将序列拆分为趋势项、周期项和残差项,最后利用长短期记忆网络对残差项开展多步预测,以评估不同预报时长对精度的影响。GRACE-C卫星在5 min短期预测中的RMSE可降至0.079 ns,而GRACE-D卫星在同时长的RMSE达到0.065 ns。
Low Earth Orbit(LEO)satellites play a crucial role in enhancing Global Navigation Satellite Systems(GNSS),where high-precision real-time prediction of Ultra-Stable Oscillator clock bias is essential for achieving centimeter-level to sub-centimeter-level positioning accuracy.However,it is difficult for traditional methods such as polynomial extrapolation and grey models to simultaneously capture the periodic and nonlinear characteristics of clock bias sequences,exhibiting significant limitations in both short-term and long-term predictions.To address this challenge,this paper focuses on GRACE-FO(Gravity Recovery and Climate Experiment Follow-On)satellite USO clock bias data.First,an improved Median Absolute Deviation(MAD)method is employed for outlier detection and preprocessing.Then,Seasonal-Trend decomposition using Loess(STL)is applied to decompose the sequence into trend,seasonal,and residual components.Finally,Long Short-Term Memory(LSTM)networks are utilized for multi-step prediction of the residual component to evaluate the impact of different prediction horizons on accuracy.The RMSE for GRACE-C can be reduced to 0.079 ns in 5-minute short-term predictions,while GRACE-D achieves an RMSE of 0.065 ns for the same duration.
作者
杨恒颀
郑秀伦
李萌萌
程彤
柯剑峰
YANG Hengqi;ZHENG Xiulun;LI Mengmeng;CHENG Tong;KE Jianfeng(School of Environment and Spatial Informatics,China University of Mining and Technology,Xuzhou 221116;State Key Laboratory of Spatial Datum,Xi’an 710054)
出处
《测绘工程》
2025年第6期8-17,共10页
Engineering of Surveying and Mapping
基金
国家重点研发计划项目(41874039)。
关键词
低轨卫星
超稳定振荡器
钟差预报
STL分解
LSTM模型
LEO satellite
ultra-stable oscillators
satellite clock bias prediction
STL decomposition
LSTM model
