Nonlinear Autoregressive Model With Exogenous Input Recurrent Neural Network to Predict Satellites’ Clock Bias

Yifeng Liang,Fangneng Li,Jiangning Xu,Pengfei Jiang

doi:10.1109/access.2021.3053265

Yifeng Liang, Fangneng Li + Show 2 more

Open Access

https://doi.org/10.1109/access.2021.3053265

Copy DOI

Abstract

The prediction of Satellites’ Clock Bias (SCB) plays an important role in optimizing the clock bias parameters in navigation messages, meeting the requirements of real-time dynamic precise point positioning and providing the prior information required for satellite autonomous navigation. Satellite-borne atomic clocks are often affected by many factors in space, which makes it difficult to describe the clocks’ bias and behavior with fixed model to achieve reliable high-precision prediction. The composition and characteristics of clock bias for satellite-borne atomic clock are described and analyzed, a clock bias prediction algorithm based on Nonlinear autoregressive model with exogenous input (NARX) recurrent neural network is proposed, the advantages of this model in SCB and other time series prediction are introduced in detail. The SCB data from four different clock types are selected for calculation and analysis. The comparative results show that, for both 6h and 24h forecasts, the accuracy and stability of NARX model are significantly better than three commonly used models, especially in the prediction of satellite cesium atomic clock.

Highlights

Global Navigation Satellite System (GNSS) itself is a high-precision time synchronization system
Nonlinear autoregressive model with exogenous input (NARX) recurrent neural network model is introduced into satellite clock bias (SCB) prediction
The SCB prediction model and algorithm flow based on NARX model are established

Summary

INTRODUCTION

Global Navigation Satellite System (GNSS) itself is a high-precision time synchronization system. Y. Liang et al.: Nonlinear Autoregressive Model With Exogenous Input Recurrent Neural Network to Predict SCB clock characteristics and prior information, which is usually difficult to obtain [5]; for ARIMA model, the determination of order p and q parameters is complicated, which will affect the final prediction accuracy [6]. In reference [8], the radial basis function (RBF) neural network is used to predict the clock bias of GPS satellite, and some better indexes are obtained compared with the traditional models. In reference [9], a wavelet neural network (WNN) clock bias prediction model is proposed, its prediction accuracy is significantly improved compared with the commonly used statistical models, including QP model and GM(1,1) model. The high-precision prediction of a series of GPS satellite clocks for 6h and 24h is carried out, proving the effectiveness and feasibility of the proposed method

NARX MODEL OF CLOCK BIAS PREDICTION

CHARACTERISTIC ANALYSIS OF SCB DATA

CONCLUSION