Due to the complexity of the operation environment,partial discharge (PD) detection is accompanied by a large amount of noise interference,resulting in the phenomenon of missed and false alarms,which affects the subsequent operation and maintenance of power equipment. In this paper,an adaptive denoising algorithm for PD fluorescence signal based on spectral kurtosis and improved empirical wavelet transform (EWT) is proposed by fluorescence fiber PD detection method. Firstly,the fast kurtogram is used to determine the compact support boundary of the Fourier spectrum of the fluorescence signal,and then the EWT is used to decompose the noise-containing fluorescence signal to obtain the useful signal components where the fluorescence signal is located. Finally,the wavelet threshold method is used to remove the residual noise from the useful signal components to obtain the denoised PD fluorescence signal. This proposed method is used to carry out denoising analysis on simulated fluorescent signals. The de-noising results are compared with the empirical mode decomposition-wavelet transform (EMD-WT) and the EWT method,which show that the method in this paper improves the signal-to-noise ratio,root-mean-square error and normalized correlation coefficient,proving that the method has good denoising effect. In addition,the denoising results of the measured signals demonstrate that this method in this paper has a higher noise reduction rate than the EMD-WT method or the EWT method does,as well as superior noise suppression capacity.