Abstract: Hydraulic turbine is the key equipment for converting hydropower energy while the cavitation phenomenon will affect the safe and stable operation of the unit. Therefore, in order to accurately identify the cavitation faults and reduce the long-term harm of cavitation to the turbine, this work proposes a step-by-step hydraulic turbine cavitation fault diagnosis method based on PVMD-AE-SVM. Firstly, the noise reduction of the hydraulic turbine operation sound signal is carried out by an improved wavelet thresholding method. Then the multi-dimensional features of the fault signal are extracted using an autoencoder, and finally, fault identification tasks are accomplished utilizing a support vector machine. The experimental fault recognition accuracy on real operating data is 99.8%, with an error of 0, which proves the effectiveness of this method. Compared with t-SNE, ProbPCA and AE, the approach herein has the higher diagnostic accuracy, shorter diagnostic time and lower error, which proves its superiority. This method can effectively eliminate the noise signal during the operation of the turbine and realize accurate identification of cavitation fault in the unit, providing an effective solution for the fault diagnosis of turbine cavitation, and demonstrating strong popularization in practical engineering.