Model Selection Method Based on ARIMA Model in Outliers Detection of Satellite Clock Offset

Due to the interference of various uncertain factors, the abnormal disturbances often occur in the satellite clock offset data, which reduces the reliability of the performance analysis of the satellite clock, destroys the validity of the modeling and prediction of clock offset, and affects the accuracy of the navigation positioning results. As to this problem, on the basis of the autoregressive integrated move average (ARIMA) model, this paper establishes an outlier detection model of clock offset time series. Based on the principle of Bayes statistics, the problems of outliers detection and the outliers magnitudes estimation are transformed into a model selection problem. Through the approximate calculation of the posterior probability of the model, the measurement standard of the model selection is derived so the complex iterative computation is avoided. Simulation Test examples of GPS and BeiDou illustrate that the proposed method can detect the outliers effectively and estimate the magnitudes of outliers accurately in the clock offset sequence; furthermore, it can obtain higher prediction precision when the method is applied in the medium and long term prediction of the satellite clock offset.