Search All of the Math Forum:
Views expressed in these public forums are not endorsed by
Drexel University or The Math Forum.
|
|
pietro
Posts:
355
Registered:
1/3/09
|
|
Re: fmincon: constraint normalization
Posted:
Mar 9, 2013 5:46 AM
|
|
"pietro " <bracardi82@email.it> wrote in message <kg4sk7$7pd$1@newscl01ah.mathworks.com>...
> Hi all,
>
> I'm using fmincon to optimize an objective function. I need to normalize all the constraints because the design variables have different units (i.e angles, lengths and torque) and I need to check the langrange multiplier otherwise they can't be compared. I normalized the objective function in this way: norm_f(x)=f(x)/f(x0), where f(x) is the objective function, f(x0) is the objective at the initial value and norm_f(x) is the normalzed one. I also normalized the constraint is this way:
>
> Br=repmat(B,1,length(A(1,:)));
> An=A./Br;
> Bn=ones(length(A(:,1)),1);
>
> where: A and B are the constraint matrices and An and Bn are the normalized one.
>
> the not normalized problem works well, also the one with the normalized objective function and with A and B as constraint matrices. Instead the fully normalized one doesn't work well, despite the normalization seems correct to me, in other words:
> isequal(A*x0-B,(An*x0-Bn).*B)
>
> ans =
>
> 1
>
> May you be kind to help me in understanding what's wrong with my code?
>
> thank you
>
> Best regards
>
> Pietro
Can't anybody help me?
Thanks
regards
Pietro
|
|
|
|
