Abstract: In view of the problem that traditional principal component analysis (PCA) ignores the correlation between coordinate components of stations, a method combining wavelet denoising and multi-directional principal component analysis (WD-MPCA) is proposed, which makes up for the shortcomings of traditional PCA. Compared with empirical mode decomposition and principal component analysis (EMD-PCA) and wavelet denoising and principal component analysis (WD-PCA). The WD-MPCA combination method has the highest accuracy. After denoising by WD-MPCA combination method, the mean median error is 0.83 mm, 0.85 mm and 8.30 mm respectively, which is 81.14%, 81.91% and 40.37% lower than that of the original coordinate residual time series. The WD-MPCA combination method fully considers the correlation between different components of each station. It can effectively remove high-frequency random white noise (WN) and low-frequency colored noise (CN), which is of great significance to the practical application and theoretical development of high-frequency Global Navigation Satellite System (GNSS) technology.