Hybrid Orbit Propagator Based on Neural Networks. Multivariate Time Series Forecasting Approach

The orbital trajectory of artificial satellites around the Earth requires frequent corrections in response to different perturbation forces. The necessary maneuvers can be designed in simulated space environments by propagating Two Line Elements with orbit propagators such as SGP4, which provides the orbital position information at a given epoch. In this work, a hybrid orbit propagator based on a neural network model is developed. Compared with previous models, the proposed neural network shows generalization capabilities for different space objects, which implies a potential benefit for the accuracy of any classical orbit propagator.