Parameter estimation of neuro-fuzzy based Hammerstein controlled autoregressive autoregressive moving average systems

This paper presents a parameter estimation scheme for neuro-fuzzy based Hammerstein controlled autoregressive autoregressive moving average (CARARMA) systems. The dynamic linear part and the static nonlinear part can be decoupled with a special input signal. By the correlation analysis method, we can get parameter estimates of the linear part. Moreover, the unmeasurable variables are replaced by their estimates, and the recursive generalized extended least squares algorithm can be applied to identify parameters for the noise model and nonlinear part. The validity of the developed method is proved by an example.