Abstract: In order to get clock products with high accuracy in real time, the rapid clock products were used to establish a short-term prediction model. First, the data batch was detrended by fitting and removing with polynomial. Then, the spectrum of residual after detrending was calculated with Fourier transformation. Thus, the modeling period and forecasting period can be confirmed with the periodic characteristics. After this, the RBF (Radial Basis Function) neural network was used to fit and forecast the clock errors. Since the RBF neural network was fit for nonlinear data modeling, this method can get better forecasting results after extracting linear trend and determining reasonable modeling data size. In fact, prediction results indicate that clock error products obtained by the proposed method are shown to have higher accuracy than the ultra-rapid products, and can satisfy decimeter accuracy positioning applications.

Key words: Clock error prediction, Rapid ephemeris, Radial basis function, Neural network, Spectrum analysis, Global position system