I read a paper called "on-and off-line identification of linear state space models".In the section six, when I run more than dozens of the cycle of the intersection algorithm, the identified order of the system is wrong,however the algorithm works well if it is used one time,not many times.I get the intersection algorithm from the web of the MathWorks writed by Bart DE Moor.
what do I deal with the off-line identification? I do two times SVD to get the state sequence,then I can get the four parameters of the system ,and the result is good.
what do I deal with the on-line identification? I also do two times SVD to get the state sequence,then I can get the four parameters of the system ,then I delete the oldest eight input data, and put eight new data into the end of the input data,then I do the procedure several times, however the result is bad, at the beginning I identify the rank of the system is four ,but with the time increasing, the rank of the system is two. I think that this is the simplest method of the on-line identification,but I connot realize it. If someone will do me a favor, please left your e-mail address.
Are you doing recursive subspace identification? Are you doing off-line identification? The algorithms would be different. The first one requires SVD updates and assumes constant a rank system. The latter does it in one shot.
my email is [email protected]. If you're looking to do online identification, you need to update after each observation. However, some portion of the SVD need not needed to be re-computed. I have Marc Moonen's original thesis. I can look up the algorithm. Also, if you look at Guillaume Mercere's Researchgate page, you will find some recursive subspace papers he published.
For a RG discussion on the meaning of the 'online' (or 'on-line') terminology at this forum:
https://www.researchgate.net/post/What_the_meaning_of_online_experimentation
- Badges
- Science topic
- Similar topics
- Engineering
- Control Theory
- Control Systems
- Control Systems Engineering