Search All of the Math Forum:
Views expressed in these public forums are not endorsed by
NCTM or The Math Forum.



system identification question
Posted:
Jun 28, 1996 9:10 AM


Hello,
I am new at using the system identification toolbox as well as system identification theory in general. I am working with the equation:
dp/dt=Wp+S(t)
where W is an nxn matrix, and p, dp/dt, and S(t) are nx1 column vectors. This defines a system of differential equations. I have entered S(t) as input and p(t) as output into the system, and I want to find the elements of W.
I have tried to define a statespace model structure to import into ident to define the structure of this system. It looks like:
F=[NaN NaN; NaN NaN] G=[1 0; 0 1] H=[1 0; 0 1] D=zeros(2,2) K=zeros(2,2) x0=[0.5 0.5]' ms2=modstruc(F, G, H, D, K, x0); thguess=[1, 1, 1, 1] th2=ms2th(ms2, 'c', thguess)
model=pem([p S], th2)
In this notation, F is equal to the W that I am looking for. I am encountering three problems:
1.) If I use a thguess, it gives back exactly what I guessed as the elements of F, even though I know they are wrong, because in a simple test case, I know what W is already
2.) If I do not provide a thguess, it gives back zeros for all the NaN's.
3.) If I give it all of "the answers", namely I replace the NaN's in F with the actual values in the system, the model output still does not match with the actual output, in fact, the rootmean square deviation between the two is given as NaN.
The third problem leads me to beleive that the statespace structure I am defining is possibly wrong for my system.
Can anyone give me any advice or help? Thanks for reading this far.
Chris Holcomb Universtiy of Rochester Department of Physics and Astronomy



