Abstract: In view of the serious output vibration harmonic distortion caused by the nonlinear stiffness of the low-frequency electromagnetic standard vibrator, the acceleration distortions are analyzed firstly based on the power series equivalent principle of nonlinear characteristics. Then, in order to overcome the low efficiency and poor accuracy of the traditional uniform sampling method for the vibrator inverse model identifying samples, a nearest neighbor uniform design input-output sample extraction method is proposed. The uniform design method selects a small sample data set from the traditional sampling points, and then adds other sample points that cannot be accurately identified to the data set according to the nearest neighbor method to form the optimal training sample set. Further, the neural network inverse model of the vibrator to be controlled is identified, and the harmonic distortion open-loop control system is constructed after it is connected in series with the original vibrator model. Finally, within the whole working frequency band, the simulation and experimental analysis show that the neural network control method can control the vibration acceleration distortion with different displacement amplitudes within 2% of the standard requirements, and the proposed sample optimization neural network control method with nearest neighbor uniform design has better harmonic distortion suppression effect.

Key words: electromagnetic standard vibrator, inverse model, harmonic control, neural network, nearest neighbor uniform design