Two-stage Parameter Estimation for the Hammerstein Nonlinear ARX Systems

In this paper, a novel two-stage parameter estimation scheme for Hammerstein nonlinear ARX (Autoregressive Exogenous, ARX) system using multi-signal processing is studied. The designed combined signal consisting of binary signals and random signals is used to separate parameter estimation issue of the Hammerstein system, which facilitates parameter estimation procedure. In the first stage, based on input and output data of binary signals the unknown parameters for linear block are estimated by using recursive least square algorithm. In the second stage, to improve the convergence rate of estimation parameters, the weights of neural-fuzzy network are learned by using particle swarm optimization algorithm according to random signals. The simulation results show that for the Hammerstein nonlinear ARX systems the proposed method can accurately identify the system parameters.