附加周期和神经网络补偿的实时钟差预报模型

Real-Time Clock Offset Prediction Model with Periodic and Neural Network Corrections

考虑到多项式模型拟合残差中仍存在显著周期信号及其他系统误差影响,提出构建一种多项式结合周期项与BP神经网络的北斗(BDS)超快速钟差预报模型,并利用实测超快速钟差数据进行算法测试验证。数值算例结果显示:利用本文模型得到的北斗超快速钟差产品,相比国内i GMAS超快速钟差产品(ISU)与德国地学中心超快速钟差产品(GBU),预报精度在3 h,6 h,12 h和24 h四个方面分别提升了26.14%,16.46%,12.68%和10.58%及10.34%,13.85%,8.17%和14.41%。

Considering the influence of the significant periodic signals and other system errors in the polynomial model fitting residuals, a real-time prediction model of BDS clock offset is proposed which is attached with the polynomial period and the back propagation （BP） neural network. The ultra-fast forecast clock offset data is used to verify the proposed algorithm. The numerical results show that compared to the China international GNSS monitoring ＆ assessment system products （ISU） and the uhra-fast of the German research centre for geosciences products （GBU） , the prediction precision of the BDS ultra-rapid clock products using the proposed model, is increased by 26. 14% , 16.46% , 12. 68% and 10.58% as well as 10.34% , 13.85%, 8.17% and 14.41% respectively in 3 h, 6h, 12 h and 24 h.