When the navigation satellite GEO performs a station keeping, high-precision orbit elements are required. The analytical solution can hardly meet the high-precision requirement while the spectrum analysis method based on Fourier analysis theory has limitations. The Hilbert-Huang transform theory decomposes the signal adaptively into a finite number of intrinsic modes and residual signals that characterize the trend of the signal through empirical mode decomposition. It is strong adaptable and frequency sensitive, so that it can give spectral analysis to the oscular orbit ephemeris sequence effectively. By analyzing the data of Beidou 3G01 satellite positioned on November 9th, 2018, a clear time-varying frequency and a set of mean orbit elements that characterize the mean motion are obtained.