a question about identifying low order model

Hello I have a high order plant to be identified. But for simplification, I want to use a low order model to identify the plant. Recursive least square(RLS) has been used to identify the plant, but the result is not satisfied. So I want to know why high order plant can't be approximated by low order model, and I want to improve identification method to complete this work.

I have found some papers about reduced order and robust identification, but I can't find I want.

Thank you for your help Wang

There could be many reasons why a lower order model would not be an adequate model for a system that is better described by a higher order model. They are *all* models so it bothers me that you refer to a "higher order plant" when it should be a "higher order model".

If there is an adequate ("high order") model of order "n" then you might ask yourself which poles or zeros can be removed from the model while keeping the adequacy of the model. Not knowing the model a priori makes that tough I should think.

In general, if a higher order model is known to be adequate then hoping that a lower order model will be adequate is just wishful thinking. You need to know.

I would start with the higher order model and start by removing the higher frequency poles and zeros - particularly those that are well damped - and testing the resulting model for adequacy. Under the assumption that this is a decent model, the critical frequencies should not move around.

If starting with an approximation of a lower order doesn't work, this suggests that the order of the model is inadequate. Well, under the assumption that the approximation method is decent and OK for the system at hand.

Fred