基于GM(1,1)与BP神经网络的卫星钟差预报

Clock bias prediction based on GM(1,1)and BP neural network

针对单一钟差预报模型的局限性,提出了一种基于GM(1,1)与BP神经网络组合的GPS卫星钟差预报方法。该方法首先用GM(1,1)对钟差进行建模预报,然后利用BP神经网络对GM(1,1)的预报残差建模,并进行外推预报,将GM(1,1)的钟差后续预报值与BP神经网络的残差预报值对应相加可得最终的钟差预报结果。用IGS提供的精密钟差数据进行试验,并与单一灰色模型进行了对比,结果显示,组合模型对未来6 h、12 h、18 h和24 h的钟差序列分别预报,平均预报精度分别为0.36 ns、0.41 ns、0.59 ns和0.84 ns,相比单一灰色模型分别提高了0.15 ns、0.17 ns、1.27 ns和2.54 ns,表明组合模型可有效地预报卫星钟差。

Aiming at the limitation of the single clock bias prediction model,a GPS satellite clock bias prediction method based on GM(1,1)and BP neural network is proposed.Firstly,the GM(1,1)is used to model and predict the clock difference,and then the BP neural network is used to model and predict the residual of the GM(1,1).The final clock bias prediction result is obtained by adding the predicted clock bias of GM(1,1)to the predicted residual of the BP neural network.The predictive tests are carried out by using the precision clock bias provided by IGS,and the results are compared with those of the single gray model.The results show that using the combination model to predict the clock bias sequences of the next 6 h,12 h,18 h and 24 h,the average prediction accuracy is 0.36 ns,0.41 ns,0.59 ns and 0.84 ns respectively,which is 0.15 ns、0.17 ns、1.27 ns和2.54 ns higher than the results of the single gray model respectively,indicating that the combination model can predict satellite clock bias effectively.