MIMO, Model order reduction
in [Matlab]
|
Prev: adaptive filter frequency domain sytem identification of fast block LMS
Next: replacing elements of matrix by its indices
From: Haavard Pehrson on 7 Mar 2010 07:09 Hello. I need some help with model order reduction. I have a MIMO nonlinear simulink model, that i have linearized. The linearization give a system with 11 input, about 80 states and 4 output, where some input dosn't affect some output at all (Gain from u1->y1 is zero). When I try to do a model reduction, I get some annoying dynamics from input to output that has zero gain in the full order model. In the reduced model, the gain from u1->y1 is not zero, but very small and it also has some oscilation. How can I avoid that? Kind regards Håvard
From: Star Strider on 7 Mar 2010 11:18 "Haavard Pehrson" <haavardpehrson(a)gmail.com> wrote in message <hn0511$ml4$1(a)fred.mathworks.com>... > Hello. > > I need some help with model order reduction. I have a MIMO nonlinear simulink model, that i have linearized. The linearization give a system with 11 input, about 80 states and 4 output, where some input dosn't affect some output at all (Gain from u1->y1 is zero). When I try to do a model reduction, I get some annoying dynamics from input to output that has zero gain in the full order model. In the reduced model, the gain from u1->y1 is not zero, but very small and it also has some oscilation. How can I avoid that? > > Kind regards > Håvard I'm not sure I completely understand your problem, but from your description it seems your system is neither controllable or observable. I suggest you determine the observability and controllability of your linearized model (original and reduced). If you model isn't too large, it might be worthwhile to post it, along with the results of your controllability and observability analyses. (If you have the Control System Toolbox, this is easy. If you don't, it's still not difficult but you'll need to consult a reference on linear control system theory.) Star Strider
From: Phil Goddard on 7 Mar 2010 21:09 When doing linearization of a Simulink model the first thing you want to do is use either the function minreal or sminreal on the resulting linear model. These eliminate any pole-zero pairs that may get introduced due to the structure of your model (rather than dynamics that you are really interested in). After using minreal/sminreal then do your model reduction. It may be that you are having problems due to the way that linmod (or linmod2) works. Phil.
From: Haavard Pehrson on 8 Mar 2010 11:13 Hi Thanks for your feedback. regarding the observability and controllabilty. I tried to use the function Wc= gram(sys,'c'), but I recieved an error message, and MATLAB help didn't recommend obsv due to numerically singularity for system with more than a handfull states. But the I am pretty sure that the process is both controllable and obsvervable from a logic point of view. It is possible to upload files here? I couldn't find out how. I tried to use minreal and sminreal, but I get the same results when I try to do a model reduction. One possible way to do this, is to extract each subsystem, do a model reduction on each subsystem and then build a new MIMO system from the reduced subsystem Kind regards Håvard Pehrson
From: Star Strider on 16 Mar 2010 12:54
"Haavard Pehrson" <haavardpehrson(a)gmail.com> wrote in message <hn37mj$hue$1(a)fred.mathworks.com>... > Hi > > Thanks for your feedback. regarding the observability and controllabilty. I tried to use the function Wc= gram(sys,'c'), but I recieved an error message, and MATLAB help didn't recommend obsv due to numerically singularity for system with more than a handfull states. But the I am pretty sure that the process is both controllable and obsvervable from a logic point of view. It is possible to upload files here? I couldn't find out how. > > I tried to use minreal and sminreal, but I get the same results when I try to do a model reduction. > > One possible way to do this, is to extract each subsystem, do a model reduction on each subsystem and then build a new MIMO system from the reduced subsystem > > Kind regards > Håvard Pehrson If your code isn't too long, copy-pasting it to your news post will work. I'm interested in your problem (although I cannot guarantee I can help you solve it). If it's too long to post (use your own judgment), or if you want to you can also send it as an e-mail attachment to me directly at: s t a r s t r i d e r ( a t ) s p e a k e a s y ( d o t ) n e t. Frank ========== |