基于灰色系统和神经网络的钟差预报

Clock bias predicting based on grey system and neural network

为避免单一模型预报钟差的弱点,提出了一种基于灰色系统和神经网络(neural network,NN)的混合模型来实现钟差的预报,并给出了基于GM(1,1)模型和广义回归神经网络(generalized regression neural network,GRNN)进行钟差预报的基本思想、具体方法和实施步骤。针对神经网络算法易训练过度、泛化能力弱的问题,采用K重交叉验证法(K-fold cross-validation)提高网络的泛化能力。为验证该混合预报模型的可行性和有效性,利用实测GPS卫星钟差数据进行钟差预报精度分析,并将其与灰色系统模型和经典权线性组合灰色模型进行比较分析。结果表明,该模型具有较好的预报精度,优于另外两种模型。

For avoiding the weakness of single model in predicting clock bias, a hybrid method combining the grey model（GM） and neural network（NN） for predicting clock bias is proposed. The basic idea, prediction model and practical process of clock bias predicting based on GM（1,1） and GRNN（generalized regression neural network） are presented. In view of the defects of traditional NN, the K-fold cross-validation algorithm is employed for improving the generalization ability of GRNN. For verifying the feasibility and validity of the hybrid method, the clock bias predictions are carried out by using the real data of GPS satellites clock bias and the prediction precisions for different methods are compared. The results show that the prediction precision for the proposed method is better than those for the GM（1,1） and the weighted combination of GM（1,1）.