With the continuous improvement of GNSS observation technology, and the accumulation of decades of observation data of GNSS observation stations such as IGS station, more in-depth analysis of GNSS coordinate time series can be carried out. GNSS coordinate time series can be described as the sum of actual signal and noise. Modeling and correcting the noise in GNSS coordinate time series can play a good role in weakening the error in GNSS data. In this paper, the optimal noise model of the coordinate time series of 18 CORS stations in Zhejiang region is established. The experimental results show that the noise models of the coordinate time series of CORS stations in Zhejiang are diverse, and the N, E, U components show different noise characteristics, which is different from the existing models. The existing Flicker Noise + White Noise (WN+FN) combination model can only represent the optimal noise characteristics of 29.63% station components; When analyzing the anniversary amplitude, ignoring the selection of the noise model will lead to the underestimation of the anniversary amplitude. Compared with WN+FN, the impact of different complex noise models on the velocity is negligible, but the impact on the velocity uncertainty of each component of the station is large. Using only a single noise model will cause the underestimation of the station velocity uncertainty.