Abstract: The parameter estimation problem of Hammerstein finite impulse response models and its application on the wind curtailment prediction field were considered. By adopting the maximum likelihood principle, the maximum likelihood estimate was obtained by minimizing the likelihood function. To reduce the impact of the unknown noise term, the maximum likelihood idea and the filtering theory were combined by changing the coupled nonlinear model into a parameter-independent model and to derive a filtering based maximum likelihood stochastic gradient algorithm for the Hammerstein system modeling on wind power curtailment prediction. The given simulation validates that the proposed algorithm can identify the wind power characteristic curve accurately and contributes to calculate the wind power curtailment prediction that shows its good practicability.

Key words: system identification, stochastic gradient, filtering theory, maximum likelihood, wind power curtailment prediction