基于机器学习的卫星轨道预测混合模型研究

The orbital prediction of satellites plays a crucial role in satellite orbit design，precise orbit determination，autonomous orbit determination，and the implementation of space missions. Aiming at the issues of lacking spatial environment information and having low prediction accuracy in the satellite orbit prediction method based on dynamic modeling，a trajectory prediction approach using a hybrid model of CNN-BiLSTM and XGBoost based on the attention mechanism was proposed. This method is based on the prediction of the dynamic model and enhances the accuracy of short-term and medium-term orbit prediction by learning and correcting the errors of historical orbit predictions. The orbital data of the Saral satellite for 7 days，14 days，and 70 days，as well as the 7-day orbital data of the TerraSAR-X satellite，were employed along with the SGP4 orbit prediction model to design experiments. Comparisons were made between the proposed prediction model and LSTM，BiLSTM，and GRU to verify its validity.