中文摘要：

为了更好地反映导航卫星钟差特性以及提高导航卫星钟差中长期预报精度，在对卫星原子钟差预报建模时，除考虑卫星原子钟频移、频漂和频漂率等物理性质外，还考虑了卫星钟差的周期性变化和随机性等特点．在传统多项式预报模型基础上，采用泛函网络对卫星钟差的周期项和随机项部分进行建模，利用GPS导航卫星钟差数据进行预报实验，并与传统的多项式模型、灰色系统模型、差分自回归滑动平均（ARIMA）模型以及Kalman滤波方法的预报结果进行比对，结果表明，基于泛函网络建立的混合预报模型能有效减小导航卫星钟差的中长期预报误差．

英文摘要：

Establishing an accurate clock error prediction model has a considerably prac- tical significance for the navigation constellation. In order to reflect the real satellite clock error and predict the medium- and long-term clock error precisely, it is necessary to take the periodic and stochastic features of the clock error into consideration besides the clock＇s phase bias, frequency bias, and the linear rate of the frequency bias. Based on the traditional polynomial prediction model and functional network, a hybrid method is proposed in this paper. By using this method, the clock error series are fitted with the polynomial model firstly, and then the functional network is used to model the periodic and stochastic terms of clock error. We take the GPS satellites for example and compare five methods in predicting the clock error, which are the hybrid method proposed in this paper, the traditional polyno- mial model, the grey model, the autoregressive integrated moving average （ARIMA） model, and the Kalman filtering model. The results show that the hybrid method can reduce the prediction error effectively.