关键词: 低轨卫星, 大气密度, 轨道预报, NARX神经网络

Abstract: Errors of orbit determination and prediction for low earth orbit (LEO) satellites mainly arise from the lack of accuracy in existing atmospheric density models. The lack of observation methods and insufficient understanding of physical mechanism of the upper atmosphere have brought difficulties to the modelling of atmospheric density. Two line element (TLE) was used to calibrate the MSIS atmospheric model, aiming at getting a localized density model along the orbit. Then a predictor was built based on the nonlinear autoregressive neural network with exogenous inputs (NARX). It uses calibrated MISIS model and a set of proxies of solar and geomagnetic activities to predict localized density values along the future orbit of a satellite. This model was applied for different types of satellite orbits and tested for different prediction windows. Comparison with the predictor based on the MSIS model shows a decrease in the mean error of the proposed model, which throws new light on improving the accuracy of LEO satellites’ short-time prediction.

Key words: low earth orbit satellite, atmospheric density, orbit prediction, nonlinear autoregressive neural network with exogenous inputs (NARX)