Long-Term Prediction of Clock Offsets Based on PSO-LSSVM*

Abstract

In the autonomous operation mode, the ground station cannot monitor the clock offset of navigation satellite. In order to restrain the long-term drift of on-board time relative to the ground navigation system time, it is necessary to make long-term prediction of the navigation satellite clock to ensure the accuracy of the timing service. A Particle Swarm Optimization - Least Squares Support Vector Machine ((PSO-LSSVM) model is established for the long-term prediction of the navigation satellite clock offset. Taking the precise clock offset of GPS satellite as an example, it is proved that the satellite clock offset has long term memory by means of the rescaled range (R/S) analysis method, and it is pointed out that the sliding prediction mode is suitable for the satellite clock offset. Then the LSSVM prediction mode is established, and PSO algorithm is used to optimize the length of the input historical clock offset series and regularization factor. Finally, the PSO-LSSVM prediction model is compared with the Quadratic Polynomial (QP) model and the grey model (GM). The results demonstrated that the prediction accuracy of PSO-LSSVM was better than that of QP and GM models in 30 days, and the prediction error did not have significant time accumulation effect.