北斗卫星轨道预报方法分析

Research on the method of high precision orbit prediction for BDS satellite

针对动力学模型预报轨道误差随弧长增加而发散的问题,用深度学习长短期记忆神经网络模型对预报误差进行补偿,且对LSTM模型逐点迭代产生的误差积累问题,提出了总体经验模态分解和LSTM模型组合的EEMD-LSTM预报模型。采用LSTM模型补偿GEO、IGSO和MEO轨道误差较BP神经网络更能完备地学习误差特性,在短、中和长期预报中,两者均方根误差差值随预报弧长增大而增大,同时误差平均改进率(Imp)也明显提高,30 d内预报中增大的(Imp)高达28.6%。且EEMD-LSTM模型较好地抑制LSTM模型误差累积,在中长期的预报中RMSE和(Imp)的差值再变化,前者高达到21.13 m,后者高达到4.24%。EEMD-LSTM组合模型补偿功能的实现对未来GNSS卫星轨道预报方法研究提供了一种参考。

Aiming at the problem that the orbit error of dynamic model was divergent with the increase of arc length,the prediction error was compensated by the deep learning Long Short Term Memory neural network model,and the error accumulation of LSTM model was caused by point by point iteration,the EEMD-LSTM prediction model combined with Ensemble Empirical Mode Decomposition and LSTM model was proposed.Using LSTM model to compensate GEO,IGSO and MEO orbit errors can learn error characteristics more completely than BP neural network,in the short,medium and long-term forecast,the RMSE difference increases with the increase of the arc length,at the same time,the average improvement rate of error(Imp)was also significantly improved,and the increase of(Imp)days’forecast was as high as 28.6%.And EEMD-LSTM model can suppress the error accumulation of LSTM model,The difference between RMSE and(Imp)changes again in the medium and long-term forecast.The difference between RMSE and(Imp)was as high as 21.13 m in the former and 4.24%in the latter.The realization of the compensation function of EEMD-LSTM combined model has reference significance for the study of GNSS satellite orbit prediction method in the future.