LSTM-BP Model: Enhancing Prediction Accuracy for Medium Earth Orbit Satellites

To improve the low accuracy of the SGP4 model in short-term orbit prediction for medium Earth orbit satellites and the instability in LSTM model training, this paper proposes and develops an LSTM-BP prediction model. Compared to the SGP4 model, the LSTM model enhances the learning of historical data, better handling long-term dependencies and improving the model's representation capability and accuracy. The LSTM-BP model not only increases prediction accuracy but also mitigates the gradient instability issue in LSTM models to some extent. The model has been validated through several sets of generalization experiments, with a significant reduction in error following corrections. For example, along the X-axis, the maximum absolute error decreased from 54.9795 km to 2.3821 km.