Recurrent neural network approach for trajectory prediction of uncontrolled space object reentry

Abstract

The trajectory modeling and re-entry prediction of uncontrolled space object is a challenging research area. Many previous studies have been conducted by using orbital dynamics, optimization technique and parameter estimation. In this paper, we have proposed new approach to predict re-entry trajectory of space object by using recurrent neural network. These deep learning models are based on LSTM and sequence-to-sequence method. Both simulated dataset and real flight dataset were validated by comparing the predicted trajectory with ground truth data. The main results from this study can be an alternative or a supplement for enhancement of prediction accuracy for re-entered space object in the future, instead of classical physics-based re-entry prediction. We verified our strategy for uncontrolled re-entry objects including different reentry objects, and got precise prediction results.