Upload benchmark results

[ZedongPeng]

TODO

1. batch_processing
   • GLOA
2. biofuel
   • Enumeration, stoped for all logic-based methods
3. CSTR @AlbertLee125
   • Not working for all methods.
4. ex1_linan_2023 @AlbertLee125
   • Hull reformulation.
   • GLOA
   • LOA
   • LBB
   • RIC
5. gdp_col @AlbertLee125
   • Hull reformulation
   • LBB does not find a feasible solution within 1 hour. (Unknown termination condition of intermediateNonInteger. Treating as infeasible.)
6. hda @tristantc
   • Hull reformulation
   • LBB does not find a feasible solution within 1 hour.
7. kaibel @tristantc
   • Not working

I will keep benchmarking the rest of the problems.

[bernalde]

Great results so far, for the solvers that involve MINLP I realized that GAMS has as default DICOPT. Let's replace it with BARON instead (at least we'll get good bounds)

[bernalde]

For the results, if using GAMS, we should also report the subsolvers (NLP, MIP and MINLP) used

[ZedongPeng]

No problem. I will also use Baron to solve the BigM and Hull reformulations. We can specify the four subsolvers (nlp_solver, mip_solver, minlp_solver, local_minlp_solver) in GDPOpt. Besides the minlp_solver, is there any other solver we should consider specifying? @bernalde

[bernalde]

I think BARON is state-of-the-art in MINLP and NLP but might be too expensive, especially for local_nlp_solver.
KNITRO is the best as local goes in my opinion. And SCIP as open source.

[ZedongPeng]

1. Use SCIP through GAMS for all subsolvers.
2. Use BARON for all subsolvers.