Nonlinear identification with input output data

Hello, I have a database about a valve, and I want to identify the dynamic of system ( mathematical model). The data has lots of random numbers for ECC and RSA coding, I tried to somehow do a little preprocessing by removing irrelevent numbers.

I don't know how to identify a system model and tried systemidentification Toolbox, but it didn't work and it only fit 1.08% at maximom with NARX. Is there any other toolbox or a method for non-linear identification? I just need a model of system for other porpuses, so I don't look for 90% fit!!

[Khaled Aljanaideh]

Hi Erfan,

Can you give information about the system you are trying to identify? Do you have a model structure that describes your system? If so, then you can use idnlgrey to estimate the parameters of the model

https://www.mathworks.com/help/ident/ref/idnlgrey.html

The accuracy of the Nonlinear ARX (NARX) model you obtain depends on the choice of regressors. Please try to add more regressors of different types and delays.

Another option you can try is Nonlinear Hammerstein-Weiner models, which you can find under the same dialog as the NARX model.

[erfan hassani]

Hi Khaled,

Unfortunately, I do not have any information about the system, I have input/output data of a fire fighting valve that I want to control with NN. I searched a lot for any model structure, but there is nothing about it.

I used both NARX and Nonlinear Hammerstein-Weiner models, but still, I can not determine the model.