Abstract: Vibration monitoring and analysis are an important means of fault diagnosis of hydropower units. How to filter noise from vibration signals to extract fault features effectively is a key problem. Therefore, a new method for the analysis of throw signals is proposed based on non-local means（NLM） and complementary ensemble empirical mode decomposition with adaptive noise（CEEMDAN） in this paper, which is applied in the swing monitoring and analysis of hydropower units. In this method, the signals are denoised by NLM-CEEMDAN.Several intrinsic mode function（IMF） are obtained and the sample entropy of each component is calculated for component classification. The denoising of the throw signal is completed by discarding the high-frequency noise components and noise component of the mixed information and noise components. Through an analysis of simulated signals and real-world signals, it can be found that this method is superior to the conventional decomposition component reconstruction and wavelet threshold denoising algorithms, which can remove noise more effectively, providing a new idea for the fault feature extraction of hydropower units.