基于小波分析和神经网络的卫星钟差预报性能分析

Performance Analysis of Satellite Clock Bias Based on Wavelet Analysis and Neural Network

为了有效地进行卫星钟差预报和更好地反映卫星钟差特性,提出了一种基于小波分析和神经网络的4阶段混合模型来实现卫星钟差的预报,并给出了基于小波分析和径向基函数(Radial Basis Function,RBF)神经网络进行卫星钟差预报的基本思想、预报模型和实施步骤.采用"滑动窗"划分数据,利用神经网络预测小波分解和去噪后的钟差序列各层系数,更精确地把握钟差序列复杂细致的变化规律,从而更好地逼近钟差序列.为验证该混合预报模型的可行性和有效性,利用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神经网络模型进行比较分析.仿真结果显示,该模型具有较好的预报精度,可为实时GPS动态精密单点定位提供较高精度的卫星钟差.

In the field of the real-time GPS precise point positioning （PPP）, the real-time and reliable prediction of satellite clock bias （SCB） is one key to realize the real-time GPS PPP with high accuracy. The satellite borne GPS atomic clock has high frequency, is very sensitive and extremely easy to be influenced by the outside world and its own factors. So it is very difficult to master its complicated and detailed law of change. With the above characters, a novel four-stage method for SCB prediction based on wavelet analysis and neural network is proposed. The basic ideas, prediction models and steps of clock bias prediction based on wavelet analysis and radial basis function （RBF） network are discussed, respectively. This model adopts ＂sliding window＂ to compartmentalize data and utilizes neural network to prognosticate coefficients of clock bias sequence at each layer after wavelet analysis and wiping off noise. As a result, the intricate and meticulous diversification rule of clock bias sequence is obtained more accurately and the clock bias sequence is better approached. Compared with the grey system model and neural network model, a careful precision analysis of SCB prediction is made to verify the feasibility and validity of this proposed method by using the performance parameters of GPS satellite clocks. The simulation results show that the prediction precision of this novel model is much better. This model can afford the SCB prediction with relatively high precision for real-time GPS PPP.