Abstract: Based on the measured vibration response data of the high arch dam discharge structure, aiming at the problems of various mixed noises under the action of discharge excitation, which will affect the accuracy of the structural working mode identification, a new filtering and de-noising method based on the improved wavelet threshold-empirical mode decomposition (EMD) joint algorithm is proposed in this paper. According to the measured vibration response data of the arch dam, the improved wavelet threshold is used to filter out most of the high frequency white noise, and reduce the boundary accumulation effect of EMD decomposition. After EMD decomposition, a detrended fluctuation analysis method (DFA) is utilized to further filter the white noise and the low-frequency current noise. The analysis results of engineering examples show that this method has better noise reduction effect compared with the wavelet analysis and the EMD method. It can accurately filter the low-frequency water flow noise and the white noise in the measured vibration response signal of the discharge structure. At the same time, the effective characteristic information in the signal is retained to the greatest extent. This method has provided convenience for extracting effective information of the arch dam discharge structure under noise interference and laid a foundation for the safety monitoring of the dam body structures.