It is difficult to find suitable Preisach function for the classical Preisach model identified by the parametric strategy for the piezoceramic actuator which has complex construction. To improve the modeling accuracy

this paper presents an average hysteresis model as the compensation of the classical Preisach model

based on which a mixed Preisach model is defined as the weight sum of the classical Preisach model and the average hysteresis model. The weight parameter called hysteresis degree is used to describe the badness of the hysteresis linearity. As the classical Preisach model

the mixed Preisach model has the wiping-out and congruency properties which compose the representation theorem. The neural network is applied to accomplish the identification of the two models. It is experimentally proven that the predicted error of the mixed Preisach model reduces 1.77 compared with the classical Preisach model when driven by the triangle voltage and 1.26 when driven by the decay sinusoid voltage.