The invention relates to the technical field of vibration speed sensors, and particularly relates to a WNN-based vibration velocity sensor amplitude-frequency characteristic compensation method. the WNN-based vibration velocity sensor amplitude-frequency characteristic compensation method adopts a steepest descent method introducing a momentum term for training a network weight value, a scale factor and a translation factor, and relates the initialization of wavelet network parameters with wavelet types, wavelet time-frequency parameters, learning samples and the like. After adopting the abovetechnical scheme, the WNN-based vibration velocity sensor amplitude-frequency characteristic compensation method has the beneficial effects of compensating for the amplitude-frequency characteristicsof a magneto-electric vibration speed sensor by using a wavelet neural network and a wavelet neural network parameter training and initialization method introduced in the invention, and the method is proved to be effective, has good robustness and can realize online compensation. Compared with the BP neural network of the same scale, the network training takes short time and has high precision,can better meet the real-time requirements, and has higher practical value in the testing field.