New algorithm for multiple satellite faults detection and exclusion based on time series prediction

A new algorithm for multiple satellite faults detection and exclusion is proposed based on the prediction theories of time series analysis. Firstly, the observations of each satellite are modeled by an autoregressive moving average (ARMA). The model above is equivalent to a filter, which can reduce the noise errors of the observations and finally can enhance the identification ability of the algorithm in small faults. Secondly, based on the model, the test statistic is established by using the statistical characteristic of the prediction residual. If the prediction error in some epoch deviates from its normal range, we conclude that the observation in this epoch contains fault. Thirdly, in order to evaluate the performance of the algorithm objectively, a method for computing the exact probabilities of missed detection is designed so that a quantitative analysis method is proposed for the evaluation of the new algorithm. Finally, we validate the effects of the new algorithm by the users in the global range under the four constellations of BDS, GPS, Galileo and GLONASS. It is shown that the algorithms proposed by this paper can handle the isolated fault or the multiple faults under the single or multiple constellations.