Abstract:Aiming at the problem of noise accumulation in deep learning model errors of orbital error sequence data, we propose an EEMD-LSTM combined prediction model and perform orbit prediction analysis with three types of satellite data: GEO, IGSO and MEO. The results show that the EEMD-LSTM combined prediction model can suppress the error accumulation of noise in satellite orbit forecasting, improve the accuracy of GEO, IGSO and MEO satellite orbit prediction, and fit the orbit prediction error of dynamic model well, and the average improvement rate(imp) of EEMD-LSTM combined prediction model for GEO, IGSO and MEO satellites increases by 2.70 percent point, 2.46 percent point and 8.33 percent point, respectively.