摘要
针对传统钟差预报模型初始偏差值较大的问题,本文采用长短时记忆网络(LSTM)优化传统钟差预报模型进行短期钟差预报。首先利用二次多项式(QP)模型、灰色模型(GM)和LSTM对前12 h的数据进行建模,分别预报0.5 h,1 h和3 h的钟差;进而用LSTM预测出的第一个数据作为真值,将传统模型预测出的钟差与其相减得到一个差值;最后将传统模型预测出的钟差减去差值得到最终的钟差预报值。本文选取BDS-3的C19,C30,C32和C374颗卫星进行实验,对比分析了QP,LSTM-QP,GM和LSTM-GM 4种模型的预报精度。实验表明,LSTM-QP和LSTM-GM模型在预报0.5 h时,其精度相较于QP和GM模型最高约提高了20.2倍和17.5倍,在预报3 h时,其最高约提高了1.4倍和1.9倍。因此,利用LSTM对初始偏差值修正有效提高了传统模型的预报精度。
In order to solve the problem of large initial deviation of the traditional clock bias prediction model,this paper uses LSTM to optimize the traditional clock bias prediction model for short-term prediction.Firstly,quadratic polynomial(QP)model,grey model(GM)and LSTM are used to model the data of the first 12 h,and the clock bias of 0.5 h,1h and 3 h are predicted respectively.Then the first data predicted by LSTM is used as the truth value,and a deviation value is obtained by subtracting the clock bias predicted by the traditional model.Finally,the clock bias predicted by the traditional model is subtracted from the error value to the final clock bias forecast value.In this paper,four satellites of C19,C30,C32 and C37 of BDS-3 are selected to carry out experiments,and the prediction accuracy of four models(QP,LSTM-QP,GM and LSTM-GM)are compared and analyzed.The experimental results show that the accuracy of LSTM-QP and LSTM-GM models is about 20.2 times and 17.5 times higher than that of QP model and GM model at 0.5 h,and 1.4 times and 1.9 times higher than that of QP model and GM model at 3 h.Therefore,using LSTM to correct the initial deviation can effectively improve the prediction accuracy of the traditional model.
作者
押少帅
齐娟
YA Shao-shuai;QI Juan(Anhui University of Science and Technology,Huainan 232001 China)
出处
《科技创新与生产力》
2022年第3期58-61,共4页
Sci-tech Innovation and Productivity
基金
大地测量与地球动力学国家重点实验室开放基金(SKLGED2020-3-9-E)。
