Medium- and Long-Term Orbit Prediction of Satellite Based on LSTNet

Abstract

Establishing high-accuracy satellite orbit prediction models is of great importance for completing space missions. Traditional satellite orbit prediction methods are mainly based on physical modeling. However, due to the complex perturbation forces of satellites in orbit, it is difficult to establish an accurate dynamic model and obtain high prediction accuracy. In this research, we propose a medium- and long-term satellite orbit prediction method based on long- and short-term time-series network (LSTNet). LSTNet is used to extract the long- and short-term dependencies and ultra-long-term repetitive patterns in satellite orbit sequences, and Huber Loss is introduced to enhance the robustness of the model to orbit outliers, so as to conduct high-precision orbit prediction. BEIDOU IGSO 1 satellite orbit data is selected for simulation validation. The experimental results show that the proposed method outperforms the traditional dynamic orbit prediction model and other deep learning models in medium-and long-term orbit prediction. The prediction accuracy of the LSTNet model is also improved by the introduction of the Huber loss function.