Research On The Short-term Stochastic Prediction Model Of Navigation Satellite Clock Offsets

In the global navigation satellite system,accurate positioning is essentially an accurate measurement of time.As the most precise instrument in time measurement,atomic clock is widely used in navigation satellite system.The clock offset of satellite-borne atomic clocks is one of the main errors in precise positioning of navigation satellites.This paper mainly studies the method of satellite clock difference prediction accuracy,the contents are as follows:First of all,error sources of precise positioning of navigation satellites and frequency characteristics of satellite-borne atomic clocks were analyzed.Secondly several traditional satellite clock prediction models were investigated,including polynomial model,gray model and kalman filter model.Based on the data of Beidou satellite,the performance and advantages of various models for short-term prediction were compared.The experimental results show that the quadratic polynomial model with additional period term has the best effect in the short-term prediction of beidou satellite clock difference.Third,the sparse auto regressive model was obtained by using the sparse theory to make the parameters of the AR model sparse.The stochastic part of the satellite clock difference was corrected by using the model,and a more perfect combined clock difference prediction model was obtained.The experimental results show that the sparse autoregressive model combined with quadratic polynomial model can significantly improve the prediction accuracy of satellite clock difference.And then,based on the traditional model,the back propagation neural network(BPNN)is used to correct the random part of the satellite clock difference.By adjusting the structure of BPNN and comparing the application modes of various neural networks,an optimal application mode is obtained.The experimental results show that the combination of BP neural network and the quadratic polynomial model with additional period term can significantly improve the prediction accuracy of satellite clock difference,among which the IGSO satellite has the largest accuracy improvement.Finally,a comparative experimental analysis was conducted on the two combined prediction models of satellite clock offsets proposed in this paper.The results show that the accuracy of the sparse autoregressive model is higher in the clock prediction within 500 min.When the prediction time is extended to one day,the prediction accuracy of BP neural network is better.BP neural network can effectively solve the problem that the prediction accuracy of traditional extrapolation model decreases rapidly with time.