Key Technologies Of Clock Offset Prediction In Navigation Satellite System

At this stage,the Beidou-3 navigation system has been fully networked.The high-precision positioning,timing and navigation services with Beidou as the core have gradually penetrated into all aspects of people’s lives and national construction.Beidou is a navigation timing positioning system based on the time and distance measurement system.The services that its satellites can provide are all based on the time-frequency reference of the spaceborne atomic clock.As the performance of the spaceborne atomic clock is affected by the changes in its own physical properties,periodic motion in orbit and solar radiation and other complex space environments,it is difficult to establish an accurate mathematical model for it.On this basis,this paper studies how to perform high-precision modeling,prediction and feature extraction on the clock error sequence of Beidou satellite-borne atomic clock observations.The main research contents are as follows:(1)This paper constructs an improved long and short memory neural network on the basis of the original cyclic neural network.The neural network model of long and short memory that is originally intelligent single-step prediction is designed with consideration of the characteristics of the on-board atomic clock.It is suitable for the ultra-short-term clock error prediction method of BDS-3 satellite.Compared with the traditional quadratic model,the gray system model and the differential integrated sliding autoregressive model under the ultra-short-term clock error forecasting,this method improves the single-hour forecast accuracy by 52.7%,66.5% and 69.8%.And the six-hour forecast accuracy is higher than the ultra-fast clock error forecast product of the International Global Positioning System Service Center.With the increase of a priori data,there is still room for further improvement in the prediction performance of the model.(2)This paper proposes a new nuclear extreme learning machine method,which aims to realize a 24-hour mid-and long-term forecasting method suitable for the BDS-3satellite.Based on the analysis,a multi-weight wavelet kernel function is designed according to Mercer’s law.The completed model was optimized with the whale optimization algorithm.Experiments show that the model is in the 24-hour forecast.The new model increases with the forecast time,the error cumulant is compared with the quadratic model,differential integrated sliding autoregressive and wavelet neural network In terms of lower,the average forecast accuracy is higher,and the ultra-fast clock error product is also included in the experiment.The experiment shows that the new model is better than the forecast part of the ultra-fast product in the 24-hour forecast accuracy.(3)This paper proposes a feature extraction method of fully adaptive noise ensemble empirical mode decomposition fusion spectrum analysis.In the residual sequence of the spaceborne atomic clock to remove the trend term,the proposed method is used to separate the characteristics of the periodic term,and then all Types of space-borne atomic clocks have been corrected for the period term,and the corrected space-borne atomic clocks are used to generate the on-board time reference.The results show that for different types of spaceborne atomic clocks,the method has improved the stability performance in ten thousand seconds by more than 10%,and effectively eliminates the deterioration of atomic clock performance due to periodical fluctuations caused by the orbital movement of the clock error data in the clock group.Compared with the traditional ALGOS algorithm,the performance of the new method is greatly improved compared with the traditional ALGOS algorithm under the condition of selecting a specific satellite with good long-term stability.