Abstract:

Normal filter algorithms cannot achieve high precision due to the modeling uncertainty caused by the earth gravity and atmospherical drag. A model-free unscented particle filter(MF-UPF) combined with Gaussian process regression is presented to overcome modelling uncertainty. Gaussian process is used to establish a relative motion model of formation flying satellites in near-circular orbits using training data, which efficiently avoids degradation of filtering performance. Simulations and comparisons validate the superiority of MF-UPF for the relative motion estimation of formation flying.