Identification of Hammerstein systems with dead-zone nonlinearities using modified CPLNN

A new one-stage identification method is proposed for Hammerstein systems in presence of non-symmetric dead-zone input nonlinearities. A modified continuous piecewise linear neural network whose activation functions are specified as the max - min linear functions is employed to describe the dead-zone element. Then a united parameterized model is derived to represent the entire system. Dead-zone parameters (thresholds and slopes) as well as the linear subsystem parameters can be calculated according to the proposed scheme. The main differences between the present method and the commonly used recursive methods lie in that the proposed model can be built without separating the nonlinear part from the linear part and no iteration procedure is needed in the parameter estimation. This method can be used without a priori knowledge of the dead-zone and is suitable for the modeling of Hammerstein systems with black-box nonlinear elements. Numerical experiments are presented to illustrate that it can be a promising tool for identifying Hammerstein systems with dead-zone nonlinearities.