How solve parametric linear system identification for non zero initial condition and noise on input ?
Hi Andrzej,
Your question seems to be a rather "general" one, because by and large, parametric identification should be easier than the non-parametric ones.
Have you checked out the methods described in the classic text:
L. Ljung, "System identification: theory for the user", Prentice Hall.
You may find most basic answers in that book.
-Sanjay
Dear All,
The system is not causal, that is why the use of identification techniques dedicated to the causal systems is risky and can't give good results. I propose a document to address this issue, witch the theme is 'Identification of fractional order systems using modulating functions method'. This method, we do not need any initial values which are usually unknown and not equal to zero.
Attached file relating to the question.
Best regards
Classical identification methods eg. linear regression LSE (System Identification: T. S. Soderstrom, Petre G. Stoica, et cetera) have good result for Gaussian noise on output and process variable. For noise on input obtain an acceptable errors eg. of quality index for the task LQ trajectory stabilization.
Regarding to the modulating functions I wonder what that function does do ?
Is this function a filter to estimated the expected values in the measurement points eg. for the ARMAX model?
or
Is the function a reconstructor of the transition of system from zero initial conditions to the current state ?
Based on publication: "Fast and Robust Online Dynamic System Identification"
can put mathematical theorem that:
Solving parametric identification of linear system for non zero initial condition base on diferential equation is not optimal solution,
so easily can proof using LSE algorithm and MSE quality index ?
- Badges
- Science topic
- Similar topics
- Engineering
- Control Theory