附加周期项与随机项补偿的卫星钟差预报算法  

Satellite clock offset prediction algorithm with additional periodic and stochastic terms compensated

作　　者：雷雨 赵丹宁 LEI Yu;ZHAO Danning(School of Computer Science&Technology,Xi’an University of Posts&Telecommunications,Xi’an 710121,China;School of Electrical&Electronic Engineering,Baoji University of Arts and Sciences,Baoji 721016,China)

机构地区：[1]西安邮电大学计算机学院,西安710121 [2]宝鸡文理学院电子电气工程学院,宝鸡721016

出　　处：《中国惯性技术学报》2024年第10期994-1000,共7页Journal of Chinese Inertial Technology

基　　金：国家自然科学基金(11503031);陕西省自然科学基础研究计划(2023-JC-YB-057)。

摘　　要：考虑到多项式卫星钟差模型拟合残差中仍存在显著周期性信号和非线性随机噪声,提出了一种附加周期项和随机项补偿的多项式预报模型和实时预报算法。首先,基于奇异谱分析方法重构钟差趋势分量、周期分量和随机分量,并利用快速傅里叶变换对重构分量进行频谱分析,分离和提取钟差的趋势项、主周期项和剩余项;其次,结合二次多项式和周期项模型拟合外推趋势项和主周期项,并获得多项式拟合残差;最后,利用神经网络对钟差剩余项和多项式拟合残差之和进行建模预测,其中网络模型通过超限学习机确定。利用GPS超快速卫星钟差实测数据对所提算法进行测试验证,结果表明,利用所提算法获得的GPS超快速卫星钟差预报产品,相比基于快速傅里叶变换的附加周期项的二次多项式模型,3 h、6 h、12 h和24 h的预报精度分别提高了59.65%、42.86%、32.50%和23.44%;相比国际GNSS服务组织提供的超快速钟差产品,预报精度分别提高了56.60%、48.57%、49.06%和47.87%。Considering that there are still significant periodic signals and nonlinear stochastic noise in fitting residuals of a polynomial model for satellite clock offset prediction,a polynomial prediction model with periodic and stochastic variations compensated is constructed and a real-time algorithm for clock offset prediction is proposed.Firstly,the trend,period and random components of clock offset are reconstructed using singular spectrum analysis(SSA),and the frequency spectrum of the reconstructed component is analyzed by the fast Fourier transform(FFT)to separate and extract the trend,predominant periodic and residual terms of clock offset.Secondly,a combined quadratic polynomial and periodic model is fitted to the trend and periodic terms,and the fitting residuals are then obtained together with the extrapolated trend and periods.Finally,the sum of the SSA residuals and polynomial fitting residuals are modeled and predicted using a neural network,in which the network model is determined by an extreme learning machine.The algorithm is tested and validated using the observed GPS ultra-rapid satellite clock offset.The results show that the proposed algorithm improves the prediction accuracy by 59.65%,42.86%,32.50%and 23.44%for the predicted 3 h,6 h,12 h and 24 h clock offset,respectively,contrast to the quadratic polynomial model with periodic terms derived from FFT.Compared with the predicted ultra-rapid clock offset products provided by the International GNSS service(IGS),the accuracy of the predicted 3 h,6 h,12 h and 24 h clock offset is improved by 56.60%,48.57%,49.06%and 47.87%,respectively.

关 键 词：钟差预报 周期变化 随机变化 奇异谱分析 神经网络 

分 类 号：P228.1[天文地球—大地测量学与测量工程]