Abstract: Aiming at the problem that noise exists in the acoustic emission signal of hydraulic turbine cavitation, which affects the effective extraction of signal features, a feature extraction method of acoustic emission signal of hydraulic turbine cavitation based on the combination of noise reduction and local mean decomposition（LMD） of optimized variational modal decomposition（VMD） and Birge-Massart strategy was proposed. In view of the significant influence of penalty factor and decomposition mode number on the decomposition results in VMD algorithm, the minimum value of dispersion entropy difference correlation coefficient was proposed as the objective function, and Harris Hawk Optimization（HHO） was used to optimize the parameters of VMD. The signal was decomposed with the VMD of the optimal parameters, and a series of Intrinsic Mode Functions（IMF） were obtained. The correlation coefficient of each IMF is calculated, the IMF with correlation coefficient less than 0.1 were eliminated, the IMF greater than 0.5 were retained, while the IMF between 0.1 and 0.5 were denoised using the wavelet BM criterion, and reconstructed with the retained components. The reconstructed signal was processed by LMD, and the energy of the decomposed Product Function（PF） component was extracted as the signal feature. The experimental results show that there is a negative correlation between PF component energy and cavitation coefficient after optimized VMD combined noise reduction treatment and LMD treatment, which verifies the feasibility of the proposed method for cavitation state identification of hydraulic turbines.