A new method for the identification of Hammerstein model (Q1376277)

An identification algorithm using a hybrid model that consists of a multilayer feedforward neural network (used to identify the static nonlinearity) in series with a parametric ARMA model of fixed order (implemented for modeling the linear dynamics) is proposed for identification of discrete-time SISO and MIMO Hammerstein systems. To update the weights of the neural network and adjust the parameter estimates of the ARMA model simultaneously, a recursive updating procedure is developed, based on the backpropagation algorithm and the recursive least squares. Performance of the method is examined by numerical examples.