Kernel Extreme Learning Machine Based on Particle Swarm Optimization for Prediction of Beidou Ultra-Rapid Clock Offset

Aiming at the problems of the complexity lying in the nonlinear characteristics of the satellite clock offset sequence and the lower accuracy of the ultra-rapid clock offset prediction, the kernel extreme learning machine algorithm is introduced into the Beidou ultra-rapid clock offset prediction. Firstly, the extreme learning machine is optimized, and the particle swarm optimization algorithm is introduced to select the kernel parameters and regularization parameters required by the kernel extreme learning machine. Then, the optimized method is applied to the ultra-rapid clock offset prediction, and the steps of using this method for the ultra-rapid clock offset prediction are given. Finally, based on the analysis of the measured Beidou ultra-rapid clock data provided by iGMAS, the data for a single day and multiple days are selected for short term prediction. The results show that in the short term prediction within 6h, the ultra-rapid clock offset prediction accuracy obtained from the optimization method adopted in this paper is obviously superior to the quadratic polynomial model and the periodic term model. Moreover, regarding the ultra-rapid clock difference prediction product obtained by this method, the prediction accuracies of GEO, IGSO and MEO satellites increase by 50.51%, 46.98% and 40.67% respectively, and their compliance with the final precision clock offset is significantly enhanced, compared with the ultra-rapid clock difference prediction product (ISU-P) provided by iGMAS. © 2019, Editorial Dept. of JA. All right reserved.