Is it possible to set a target value

[Bakhvar M.]

Hi all

is it possible to set a target value in AIMMS? what I mean is that, I
know the maximum or minimum that is possible to be reached so I want
to tell AIMMS to search for a maximum or minimum value within the
small interval of [-a,a] rather than looking into the big interval of
[-inf,inf], so result is reached faster

one way might be to make constraints that limit the objective
function, but is this really efficient?

[Mauro Rozenberg]

I'm not sure it's gonna be useful or even work.
If you set a constraint like that you might end up making the model
unfeasible. Just picture that Aimms sets the bound at A (any value)
and you set a constraint that the equation to minimize has to be under
B. If B < A then the model is infeasible and if A < B then the new
constraint is useless.
What I can suggest you to make it faster (with out improving the
model) is to configure a tolerance for the optimization. This way you
won’t get ‘the optimal solution’ but it might be as good.
Another option is to set intervals but not objective function but to
the variables. This will also get you a non-optimal solution if the
intervals are not correct.
In my experience guiding the model by using start with values is only
done when working with non-linear models since you can have local
minimum or maximum. After all we use these tools so that we don’t need
to solve it!

Best,
Mauro.

[Bakhvar M.]

Thanks Mauro

in fact, the problem that I am working on, is non linear and I am
using to CONOPT to solve it,

the thing is that I have around 50000 variables and 48000 constraints
and my objective function is to minimise the sum of a few variables, I
was thinking about solving this problem in 11 steps in the first 10
steps I will find the minimum for groups of 500 variables and 480
constraints then in the last step I will solve the whole problem at
once with a smaller interval that comes from the minimum of the
results found in the previous 10 steps.

could you please explain how can I configure tolerance, it sounds
interesting

Regards

Bakhtvar M.

[Marcel Hunting]

Hi,

I think Mauro was referring to the options ‘MIP Relative Optimality
Tolerance’ and ‘MIP Absolute Optimality Tolerance’ which do not apply
to NLP.

It is not possible to set a target value for the nonlinear solvers in
AIMMS. I doubt that it will be effective to add constraints to
restrict the objective value.

You might want to switch on the nonlinear presolver as controlled by
the option ‘Nonlinear Presolve’.

Your suggestion of using 11 steps seems similar to a rolling horizon
approach (see Language Reference) and could work.

Marcel Hunting
AIMMS Software Developer

> > > function, but is this really efficient?- Hide quoted text -
>
> - Show quoted text -

[Mauro Rozenberg]

Hi,
I was refering to those options so my advise is no good.
You should try the options Marcel mentioned.

Best,
Mauro.

> > - Show quoted text -- Hide quoted text -