基于长短期记忆神经网络的导航卫星钟差预报

Study on navigation satellite clock error prediction based on LSTM

针对常用导航卫星钟差预报模型预报精度不高的问题,提出了1种基于长短期记忆神经网络的导航卫星钟差预报方法。建立了由2个长短期记忆神经网络层、2个舍弃层和1个全连接层组成导航卫星钟差预报模型。设计了递推式预报方式。递推式预报一步一预测,递推完成预报。执行过程分为3个阶段:首先利用历史卫星钟差准备训练数据,然后对预报模型进行训练,最后利用训练好的模型进行预报。试验结果表明,采用长短期记忆神经网络方法比二次多项式法的预报精度有较大提升,比线性法预报精度略有提升。基于长短期记忆神经网络的导航卫星钟差预报方法,为提高导航卫星钟差预报精度提供了1种新的思路。

Common models are not accurate enough to meet the needs of high-precision applications.To improve the accuracy of satellite clock error prediction,a method is proposed based on Long Short-Term Memory(LSTM)neural networks.A prediction model is designed,which is composed of two LSTM layers,two dropout layers and a fully connected layer.A step patten is designed.By’step pattern’,it means to make clock error prediction step by step until completion.The execution process includes three stages.At first,prepare the training data using history clock error.Then,train the prediction model.At last,make satellite clock error prediction by the model.By case analysis,the prediction method based on LSTM is much more accurate than quadratic polynomial model,and it is slightly more accurate than linear polynomial model.It is a new idea to improve the accuracy of satellite clock error prediction using deep learning method.