A medium and long-term intelligent orbit prediction method for LEO satellites based on segmented pseudo-drag coefficients

Abstract

In traditional orbit prediction (OP) tasks, the precision achieved for low earth orbit (LEO) satellites rapidly decreases over time because of the uncertainty of the space perturbation model and the satellite parameters. A machine learning method is proposed to improve the precision of medium and long-term OP. First, the concept of pseudo-drag coefficient is proposed, and the long-sequence OP task is transformed into a short-sequence pseudo-drag coefficient prediction problem. Second, historical segmented pseudo-drag coefficient datasets are established by the golden section method, and a convolutional neural network with a squeeze-and-excitation block and long short-term memory (CNN-SE-LSTM) is proposed to train and predict this coefficient. Furthermore, the predicted pseudo-drag coefficient can be combined with the dynamic OP model to achieve high-precision medium and long-term intelligent OP. Finally, practical measurement datasets are used to verify the effectiveness of the proposed method, and the results show that the method can significantly improve the OP precision of LEO satellites.