What is the criterion of best fit in system identification toolbox?

sir i am finding model for pressure control loop using experimental data .I got best fit till is 86.47 so is it sufficient or it is always required above 90 please tell us that i tries many times.

Answer 1

[Rajiv Singh]

Normalized Root Mean Square (NRMSE) criterion is used to generate the fit % number. It is a measure of how much better the model is in reproducing the observed data relative to the mean of the data. See:

http://www.mathworks.com/help/ident/ref/goodnessoffit.html

As said before, this by itself is not an absolute test of model quality; it depends upon SNR, nature of noise and desired quality levels. But at the very least one should also evaluate the quality of the residuals as given by the Whiteness test and Correlation test; see: http://www.mathworks.com/help/ident/ref/resid.html

Answer 2

[David Sanchez]

It all depends on your system and on how much accuracy is required. Obviously, when the best fit is around 95%, your model would yield much closer approximations to the real system. But 86% could seem a poor approximation, but it could be enough in some others. It's your system you are modelling, test it and tell yourself whether the results fulfil your requirements or not.