一种CEEMD-TCN的BDS轨道误差预报算法

A BDS orbit error prediction algorithm based on CEEMD-TCN

针对长短期记忆网络(LSTM)对利用动力学模型预报的北斗卫星导航系统(BDS)轨道预报误差值拟合效果不优的问题,设计一种互补集合经验模态分解与时间卷积网络结合(CEEMD-TCN)的BDS轨道误差预报算法。该算法首先采用互补集合经验模态分解(CEEMD),对BDS轨道预报误差序列进行降噪处理,接着对降噪后的信号进行重构,最后将重构得到的轨道误差序列作为数据训练集进行时间卷积网络(TCN)训练。以地球静止轨道(GEO)、倾斜地球同步轨道(IGSO)、中圆地球轨道(MEO)上的三颗卫星的轨道误差序列作为研究对象,分别采用CEEMD-TCN、TCN、CEEMD-LSTM、LSTM四种补偿模型对卫星轨道预报在X、Y、Z方向上的误差进行短、中长期预报。实验结果表明:CEEMD-TCN相较于其他三种方法对BDS轨道误差预报效果最好,且CEEMD-TCN与CEEMD-LSTM中长期预报效果分别优于对应的TCN和LSTM,说明对中长期轨道误差序列进行降噪处理有利于提高预报精度。

In order to solve the problem that the fitting effect of Long-term and Short-term Memory Network(LSTM)on the orbit prediction error of BeiDou navigation satellite System(BDS) predicted by dynamic model is not good, a BDS orbit prediction error prediction algorithm based on the combination of Complementary Ensemble Empirical Mode Decomposition and Time Convolution Network(CEEMD-TCN) is designed. Firstly, the BDS orbit prediction error sequence is denoised by using the Complementary Set Empirical Mode Decomposition(CEEMD), then the denoised signal is reconstructed, and finally the reconstructed orbit error sequence is used as the data training set to train the Time Convolution Network(TCN). Taking the orbit error series of GEostationary Orbit(GEO), Inclined GeoSynchronous Orbit(IGSO) and Medium Earth Orbit(MEO) as the research object, the short-, medium-, and long-term prediction of X, Y, and Z directions of satellite orbit prediction error is carried out by using CEEMD-TCN, TCN, CEEMD-LSTM, and LSTM compensation models respectively. The experimental results show that CEEMD-TCN is better than the other three methods in BDS orbit error prediction, and CEEMD-TCN and CEEMD-LSTM are better than the corresponding TCN and LSTM in medium and long-term orbit error prediction, which indicates that the noise reduction of medium and long-term orbit error series is beneficial to improve the prediction accuracy.