Title of article :
Finite model order accuracy in Hammerstein model estimation
Author/Authors :
Hjalmarsson، نويسنده , , Hهkan and Mهrtensson، نويسنده , , Jonas، نويسنده ,
Issue Information :
روزنامه با شماره پیاپی سال 2012
Pages :
7
From page :
2640
To page :
2646
Abstract :
Hammerstein models is one of the most commonly used model classes used for identifying nonlinear systems. A static input nonlinearity followed by a linear dynamical part is an adequate way to model many real-life systems. This paper investigates the asymptotic (in terms of sample size) variance of Hammerstein model estimates. The work extends earlier results by Ninness and Gibson (2002) in the following ways. Not only frequency function estimation but estimation of general quantities is considered. The expressions are not restricted to be valid asymptotically in the model order. In addition, the results cover model structures having noise models and allow for data generated under feedback. The increase in variance due to the estimation of the input nonlinearity is characterized. In particular, under open loop operation, white additive noise and the assumption of a separable process, it is shown that the variance increase is exactly a term that was observed in Ninness and Gibson (2002) to result in good agreement with simulations. This term vanishes in the formal asymptotic in model order analysis in Ninness and Gibson (2002).
Keywords :
System identification , Hammerstein models , Prediction error identification , Asymptotic covariance
Journal title :
Automatica
Serial Year :
2012
Journal title :
Automatica
Record number :
1448885
Link To Document :
بازگشت