GAMS World Forum

Hello All,

I have a problem using BONMIN solver for MINLP problem optimization, i am trying different configurations, most of them went well and an optimal solution found but in one configuration a message appears that no feasible solution found, but it gives another message with best possible solution= certain value, only reliable for convex model ??

When i made a small change in the original wind pattern in the same problem, the solver reach an optimal solution. That's mean there is no problem in the modelling.

Is there any suggestion, why the solver didn't find an optimal solution using the original wind pattern ? and how to fix this problem ?

I assume wind pattern is some kind of parameter in your model. It is possible for a model to become infeasible with slight change in parameter values. This does not conclude if your model is right or wrong.
I am not sure with BONMIN, but in general, does the solver terminate by saving 'proven infeasible' or by exceeding 'resource limit' and failing to find a feasible solution? If you are facing the latter case, it might be a good idea to solve rminlp (relaxed minlp) just before solving minlp. The relaxed problem is nlp and if it is infeasible, then the model needs to be inspected.
Another trick might be to minimize a constant (minimize 0) just to force the solver into finding a feasible solution before actually solving your problem.

- Atharv

Thank you abhosekar for your reply.

I tried RMINLP but using other solver IPOPT and it found an optimal solution in the cases in which BONMIN-MINLP couldn't find. Do you believe that i can rely on the output results of RMINLP ?

If RMINLP is feasible (irrespective of which solver you use), then it indicates that the relaxed nlp is feasible. you cannot use this solution as the integer variables may be at non-integer values. You can, however, use these values as a starting point for your minlp solve (if you write solve statement right after the solve statement for rminlp, GAMS automatically uses the previous solve as a starting point for the current solve).
It is still not clear if your problem is proven infeasible or fails to find a feasible point within the given time limit. It should be noted that if rminlp is feasible and minlp is proven infeasible, then the problem is integer infeasible which is difficult to debug following any specific method. It will then require you to get deeper understanding of your model and some trials (by relaxing some constraints).

- Atharv

Thank you very much abhosekar

Hello,

I was wondering if there are some tricks to speed up the operation and solution of MINLP BONMIN solver, it takes very long time.