System Identification Toolbox: How can we modify the starting parameters for the armax-algorithm?

One can calculate ARMA-coefficients using the armax-algorithm from the Systems Identification Toolbox:

estimatedPolymodel=armax(iddata(outputdata,inputdata,tsample),[na nb nc nk], opt);
ARcoeff=estimatedPolymodel.A
MAcoeff=estimatedPolymodel.B

How can we modify the starting parameters for the algorithm (to accelerate and improve the results)?

Answer 1

[Rajiv Singh]

You can set the A, B, C values explicitly, as in

estimatedPolymodel.A = ARCoeff

Or, call the IDPOLY constructor with A, B, C polynomials initialized to the desired values.

A = [1 .1]
B = [0 0 4 1]
C = [1 .3]
m = idpoly(A,B,C,'Ts',1)
% refine model's initial parameter values 
m2 = armax(data, m)

% but I wanted to update only A and B but not C!
m.Structure.C.Free = false;
m3 = armax(data, m)

[Urs Hackstein]

Dear Rajiv Singh,

thank you very much for your prompt and precise answer which explains how to set fixed values as start parameters for A, B and C. But how can you define intervals of constraint for A, B or C, e.g.

?

[Rajiv Singh]

Look under the "structure" property. In addition to "Free", it also offers "Minimum" and "Maximum" specifications that you can set for individual coefficients.