Modified Neural Network with Hysteresis Operators and Adaptive Learning for Tracking Control of Piezoelectric Stack Actuator

According to the proposed four-layer modified neural network with adaptive learning, an adaptive learning model is designed, and the Play operator weight function update algorithm, the dead-zone operator weight function update algorithm, and the hybrid model program are studied. The hysteresis nonlinearity of a piezoelectric stack actuator at multiple frequencies was tested separately, and the root mean square error (RMSE) of five control methods, including the without control, classic PI and DZ model, and the four-layer modified neural network with an adaptive learning model, were compared through experimental studies. The experimental results show that compared with the without control condition, the RMSE of the classic PI and DZ model is reduced by 67.98% at a frequency of 1 Hz, which can effectively reduce the hysteresis nonlinearity of the piezoelectric stack actuator and has a good hysteresis compensation effect. Compared with the classic PI and DZ model, under the four-layer modified neural network with an adaptive learning model, the RMSE of the piezoelectric stack actuator is reduced by 15.34% at 1 Hz, and the error can still be reduced by 67.75% even at 10 Hz. Indicating that the four-layer modified neural network with an adaptive learning model still has a good hysteresis compensation effect at a wider frequency band.