A machine learning (ML) approach has been proposed to improve orbit prediction accuracy in previous studies. In this paper, the artificial neural network (ANN) model is investigated for the same purpose. The ANNs are trained by historical orbit determination and prediction data of a resident space object (RSO) in a simulated space catalog environment. Because of ANN’s universal approximation capability and flexible network structures, it has been found that the trained ANNs can achieve good performance in various situations. Specifically, this study demonstrates and validates the generalization capabilities to future epochs and to different RSOs, which are two situations important to practical applications. A systematic investigation of the effect of the random initialization during the training and the ANN’s network structure has also been studied in the paper. The results in the paper reveal that the ML approach using ANN can significantly improve the orbit prediction.
Artificial Neural Network–Based Machine Learning Approach to Improve Orbit Prediction Accuracy
Journal of Spacecraft and Rockets ; 55 , 5 ; 1248-1260
2018-05-31
13 pages
Aufsatz (Zeitschrift)
Elektronische Ressource
Englisch
British Library Conference Proceedings | 2018
|Estimating Atmosphere Density Variations to Improve LEO Orbit Prediction Accuracy
British Library Conference Proceedings | 1998
|