Outliers processing method of navigation satellite telemetry data based on time-varying neural network

Abstract

In view of the characteristics of dense and non-stationary outliers in remote sensing data of navigation satellite in complex space environment, a method of eliminating outliers in residual test based on time-varying radial basis neural network was proposed. In the method of outliers elimination, the time-varying radial basis neural network (RBF) is firstly modeled according to the telemetry data. After the training network is stable, the residuals of the original sequence and the fitting sequence based on RBF neural network are calculated. Then the residual is tested by the adaptive threshold value to determine the outliers in the telemetry data. Finally, the method is proved to be effective in detecting isolated outliers and speckled outliers by practical application.