基于小波变换和MC-ANNs并行模型的经验加速度预测

为对低轨卫星精密定轨过程产生的经验加速度进行拟合与预报,提出基于小波变换和MC-ANNs (Markov chainartificial neural networks)的并行模型。引入小波变换技术,结合神经网络的深度抽象以及马尔科夫链的动态随机的双重优势,缓解串联式马尔科夫链-神经网络模型随着时间推移导致误差变大的缺点。对某低轨航天器的实测星载的经验加速度预测分析结果表示,相较于BP (back propagation)神经网络模型、串联式马尔科夫链-神经网络模型,该模型的预测精度平均提高29.83%、16.04%,应用于轨道确定与轨道预报中,可提高卫星定位的精度。

Empirical acceleration prediction based on wavelet transform and MC-ANNs parallel model

To fit and forecast the empirical acceleration generated by the low earth orbit satellite determination,the parallel model based on wavelet transform and MC-ANNs was proposed.With the introduction of the wavelet transform,the depth abstractness of neural network and the dynamic randomness of Markov chain were combined to minimize inaccuracies accumulated over time from the series Markov chain-neural network model.When it came to the use of predicting and analyzing the empirical acce- leration of a low orbit spacecraft,the prediction accuracy of this model compared with that of the BP neural network model and the series Markov chain-neural network model can be increased by an average of 29.83% and 16.04% respectively.The accuracy of satellite positioning can be improved by applying this model in orbit determination and orbit prediction.