变权组合模型在卫星钟差预报的研究

Research on Satellite Clock Error Prediction Based on Variable Weight Combination Model

针对RBF神经网络、GM(1,1)模型和ARMA模型3种单一模型在卫星钟差预报中存在局限性的问题,文章提出3种模型组合的最优非负变权组合模型,利用该模型对GPS卫星钟差进行3 h,6 h,12 h,24 h不同时长的预报,取得了均方根误差分别为0.026 603 ns、0.183 017 ns、0.471 233 ns、0.457 649 ns的预报精度,证明了基于此3种单一模型的最优非负变权组合模型在钟差预报中的可靠性。

According to the shortcomings of single satellite clock error prediction model,in this paper,based on three kinds of single model,RBF neural network,gray GM(1,1)model and ARMA model,Using this model,the GPS satellite clock error is predicted for 3h,6h,12h,24h,The prediction precision of the root-mean-square error is 0.026603ns,0.183017ns,0.471233 ns and 0.457649ns,Facts proved that it has the reliability of the model forecast.