I have a problem with CANON functions of different Matlab versions (e.g. 2007b, 2010b, 2011a) returning inconsistent results for the same call parameters (sysd = canon(sysd0,'modal'):
- the input/output relation stays the same - OK.
- the diagonal A-matrix is permutated slightly - still OK,
- the B and C matrices differ significantly.
I undersdand the modal decomposition has in multiple solutions, however, as far as I know, there is no useful control over the CANON function realization. Furthermore in newer Matlab versions the canon function calls p-type subfunctions which makes them inaccessible.
The obtained model is being used for tuning of predictive controller. Different B-matrix subsequently yielding significantly different results for certain set of tuning weights.
I have the original data generated using Matlab 2007. As I want to replicate it, and further provide public the code for performing a benchmark test, I need to have the CANON function to return consistent results regardless of the Matlab version used... Or find an alternative way defining/norming the B & C matrices afterwards
Is there any way to get consistent modal decomposition?
I would say the problem is more with your controller than with the canon function. Indeed, a well-designed controller should be independent of the state representation you used: the controller manipulates the inputs using measurements given by outputs, sothe actual state representation you use should not matter.
Dear Boštjan Pregelj,
If you just intend to use CANON for state space representation in MATLAB, you may not view the results as you expect from the manual calculation. The MATLAB "Canon" function has its own way of representation form which are related to some texts published by US authors.
canon(sys,T) , performs the similarity transformation z = Tx on the state vector x of the state-space model sys with A,B, C, D matrices in a particular form which may be different from your case.
So better will be to use the canonical representation steps as per your method.
Regards,
Shwetha M
Thank you for suggestions!
I have already taken few measires.
- I have askd withinin the Matlab forums first, but got no answer within a month.
- Afterwards I have asked Matlab support directly, but only got answer, that there are differences between M2007b and later versions. However, i find diferences also between later versions (M2010b, M2011a).
@Philippe Martin. We have been tuning the MPC by defining some weights via b*b' (LTR) approach. I know this is not the best for certain cases, but the differences in b matrices were of factor of 2 to 10 or more.
- Badges
- Science topic