Prediction of Satellite Clock Errors with a Combination of the Gray Model and the Least-Squares Support Vector Machines

We propose a new method for predicting satellite clock errors.The new method combines the Gray Model (GM) and the Least-Squares Support Vector Machines (LS-SVM) regression algorithm.The method first builds different GM (1,1) models based on observational data of satellite clock errors.Future satellite clock errors are predicted from these models.The LS-SVM regression algorithm is employed to combine the prediction results of the GM (1,1) models in a nonlinear manner.This method keeps the advantages of the GM (1,1) models such as relatively small amounts of required observational data and the simplicity in data modeling, and in the mean time, it has all the useful features of the LS-SVM, including relatively few samples to be used, nonlinearity, and easy generalization.Therefore, the method improves the reliability and accuracy of prediction results over previous methods.Our experimental results show that the new method is feasible, effective, and practical.