Modeling hysteresis using hybrid method of continuous transformation and neural networks

A novel and simple approach to modeling hysteresis nonlinearities is proposed. The continuous transformation technique is used to construct an elementary hysteresis model (EHM), which forms a one-to-one relation between the input space and the output space of hysteresis nonlinearities. In theory, we can apply the output of the EHM as one of the input signals of a common neural network (NN) to approximate any kind of hysteresis nonlinearities, which meet any input signals satisfying an assumption. In order to validate the effectiveness of the proposed approach we use several sets of data from the proposed backlash-based hysteresis simulation models (BHSMs) for respective simulation testing. Then a set of real data measurements is used to evaluate the proposed approach. These results of simulation testing indicate that the proposed approach is simple and successful.