Parameter Identification of Nonlinear Hammerstein Model Using Stacked Auto Encoder Network

This paper discusses a modeling and identification scheme for the Hammerstein model using the stacked auto encoder (SAE) network. The Hammerstein model is composed of a static nonlinear block and a dynamic linear block, in which the nonlinear block is modeled by SAE network and the linear block is modeled by auto-regressive with extra inputs (ARX) model, and special test signals composed of binary signal and random signal are employed to realize parameter identification of each block in the Hammerstein model. Firstly, the static nonlinear block and the dynamic linear block of the Hammerstein model are separated using the correlation characteristics of separable signals, and the parameters of the dynamic linear block are identified according to the correlation analysis method. Then, the weights of two hidden layers of the SAE network are updated based on conjugate gradient training and back propagation techniques to identify the nonlinear block.. Simulation results show that the proposed method can effectively identify the Hammerstein model and has achieved good identification accuracy.