Orbital prediction accuracy improvement method based on particle swarm optimization RBF neural network

Abstract

In response to the increasing demand for the improved accuracy in satellite orbit prediction, this study introduces a method to enhance the orbit prediction precision with RBF neural network utilizing particle swarm optimization. The research specifically targets low-Earth orbit satellites for the development and modeling process. The constructed PSO-RBFNN model is trained using the orbital error data from real satellites, enabling the prediction of future satellite error data. Then, the prediction error data is utilized to correct any errors that may arise during the calculation of orbit predictions in the dynamics model. In order to demonstrate the outstanding performance of the model proposed in this paper, we trained the PSO-RBFNN designed in this study and other common neural network models (such as BP, LSTM, and RBF) on data at different time intervals. The results confirm that the PSO-RBFNN has good prediction ability when applied to orbital error data. Furthermore, the study examines the impact of neural network architecture and training data on the performance of the model. Finally, this study evaluates the generalization capability of a model designed on one satellite to other satellites.