Re: model identification

can anybody help me please?




"mary " <maruetta@xxxxxxxxxx> wrote in message <h06eoq$nnr$1@xxxxxxxxxxxxxxxxxx>...
I think I was not clear enough in my last response.
I calculated the Box-Jenkins model concerning the input u1 and the output y looking for the best fit. I did the same computation again with the second input u2 and the output y. So now I have:

bestModel1 = ord1{ind}; %for u1
bestModel2 = ord2{ind2}; %for u2

which are my two Box-Jenkins models of my two inputs taken singularly (but which have the same output). Now,knowing that the Box-Jenkins has the form:

y(t)= B(q)/F(q)*u(t-nk)+C(q)/D(q)*e(t)

I'm interested in taking only the part which deals with the input of the two models I found (bestModel1 e bestModel2) and summing them,that is doing:

Ytot(t)=B1(q)/F1(q)*u1(t-nk)+B2(q)/F2(q)*u2(t-nk)

and then compare this Ytot(t) with my measured data (using compare and resid if it's possible).

thank you very much for your help.












"Rajiv Singh" <Rajiv.SinghNoSpam@xxxxxxxxxxxxx> wrote in message <h04608$f9g$1@xxxxxxxxxxxxxxxxxx>...
The code you show will compute the output of the model for zero initial
conditions. This is same as output of sim(bestModel1,[u1 u2],
'init','zero'). If you want compare this output to measured value, you can
simply plot them together using PLOT.

For comparison of measured vs simulated output, use COMPARE as in:
compare(bestModel1, data, 'init', 'z')

I am not sure what your end goal is, but to summarize the commands
available:

1. COMPARE: plots measured and simulated/predicted output together; can also
return fit values as output argument. See "help idmodel/compare"
2. PE: Compute prediction error for a given model.
3. RESID: Compute prediction error and perform correlation analysis on it
(whiteness test, correlation test); see "help resid".
4. SIM: Simulate the model - either view the simulated response in a plot or
fetch response data as output argument; see "help idmodel/sim"

Rajiv

"mary " <maruetta@xxxxxxxxxx> wrote in message
news:h03ss2$b05$1@xxxxxxxxxxxxxxxxxxxxx
ok your replay was fundamental for me,but I think I made the wrong
question.What I meant was this:
after I sum only the functions which deal with the data (not the error)
this way

y =
filter(bestModel1.b,bestModel1.f,u1)+filter(bestModel1.b,bestModel1.f,u2);

do I have to use compare or reside with the output data we have?


thank you very much and sorry for bothering you,
Mary






"Rajiv Singh" <Rajiv.SinghNoSpam@xxxxxxxxxxxxx> wrote in message
<h01cnf$ot1$1@xxxxxxxxxxxxxxxxxx>...
Hi Mary,
Yes this is what I suggested in essence. However, note that only one
COMPARE
call would be sufficient since you can call it with multiple models as
input
arguments. Use the model for which the fit value is maximum.

ord1 = cell(0,1);
for u = 1:1:nmax
for r = 1:1:nmax
for s = 1:1:nmax
for t = 1:1:nmax
ord1{end+1,1} = bj(datistima,[u,r,s,t,1]);
end
end
end
end

[yh,fit]= compare(datival,ord1{:});

fit = squeeze(fit);
[maxval, ind] = max(find);
bestModel = ord1{ind};

HTH,
Rajiv

"mary " <maruetta@xxxxxxxxxx> wrote in message
news:h01aj1$59d$1@xxxxxxxxxxxxxxxxxxxxx
well i should firstly thank you so much because you are very kind.There
is
one last doubt:
once we calculate all the possible fits using compare how can we choose
the orders which provide the best fit?
This is what I did:

for u = 1:1:nmax
for r = 1:1:nmax
for s = 1:1:nmax
for t = 1:1:nmax
ord1 = bj(datistima,[u,r,s,t,1]);
[yh,fit(u,r,s,t)]= compare(datival,ord1);
end
end
end
end

Is this what you mean?If it is how can we then choose the orders of the
model?using FIND?but how?


thank you so much again

best wishes
Mary











"Rajiv Singh" <Rajiv.SinghNoSpam@xxxxxxxxxxxxx> wrote in message
<h00n1k$s0h$1@xxxxxxxxxxxxxxxxxx>...
You can concatenate SISO models, as in M = [M1, M2]. See
idpoly/horzcat.

Rajiv


"mary " <maruetta@xxxxxxxxxx> wrote in message
news:gvuen1$t16$1@xxxxxxxxxxxxxxxxxxxxx
well,as i have to deal with two inputs and one output but i want to
calculate the box-jenkins taking an input singularly (to compare it
with
the one i would obtain if I did it MISO - multiple input
single-output)
how can I sum the two B-J models (one dealing with the first input
and
one
dealing with the second) and then compare them with the data?

thank u very much for your help







"Rajiv Singh" <Rajiv.SinghNoSpam@xxxxxxxxxxxxx> wrote in message
<gvospo$q8h$1@xxxxxxxxxxxxxxxxxx>...
Hi Mary,
arxstruc/selstruc are meant for ARX models only, however, they
would
indeed
give you some guidance on the corresponding orders for the BJ
model.
"na"
would indicate, roughly, the number of poles (orders of F and D),
while
"nb"
would tell you about order of B. As a starting point, I would
recommend
using na to define order of F, using nb to define order of B and nk
for
input delay. For the orders of the noise component (C/D), just
start
by
choosing something simple (lowest possible orders). Do a residual
test
(see
RESID) after estimation and based on those results slowly increase
the
order
of noise component related polynomials.

If you have ample time at your hand, you can also do an exhaustive
search
for best orders- choose a grid of (nb, nc, nd, nf, nk) values and
try
out
all combinations for estimation. For each estimated model, compute
percent
fit to an independent (validation) data set using COMPARE. Pick
orders
that
provide best fit to validation data.

HTH,
Rajiv


"mary " <maruetta@xxxxxxxxxx> wrote in message
news:gvn105$ko5$1@xxxxxxxxxxxxxxxxxxxxx
hello,i desperately need ur help!!
well,i have a series of input (there is more than one input
series)and
output data and i should identify a box-jenkins model,which,given
the
input data,identifies the output ones. I have to identify the
orders
and
the values of the parameters.
Are there functions like selstruc or arxstruc also for the
box-jenkins
model or should I implement a function for my purpose?

waiting for ur replies,
thank you so much




.

Relevant Pages

  • Re: model identification
    ... The output of this model may be compared to measured data using COMPARE. ... take into account the noise component. ... I calculated the Box-Jenkins model concerning the input u1 and the output ... y looking for the best fit. ...
    (comp.soft-sys.matlab)
  • Re: model identification
    ... once we calculate all the possible fits using compare how can we choose the orders which provide the best fit? ... RESID) after estimation and based on those results slowly increase the ... fit to an independent data set using COMPARE. ... output data and i should identify a box-jenkins model,which,given the ...
    (comp.soft-sys.matlab)
  • Re: model identification
    ... simply plot them together using PLOT. ... For comparison of measured vs simulated output, use COMPARE as in: ... Use the model for which the fit value is maximum. ... RESID) after estimation and based on those results slowly increase ...
    (comp.soft-sys.matlab)
  • Re: model identification
    ... I calculated the Box-Jenkins model concerning the input u1 and the output y looking for the best fit. ... and then compare this Ytotwith my measured data. ... simply plot them together using PLOT. ...
    (comp.soft-sys.matlab)
  • Re: smooth equation
    ... I have two sets of experimental data. ... I'd like to compare them, but the data are not obtain at the same ... My idea is to plot them with smooth. ... if you know that the data should be linear, I suggest to fit both sets simultanously and plot data - fit for both sets. ...
    (comp.graphics.apps.gnuplot)
Loading