Originally posted by: ORA17

Hello Experts,

I am running a MIP problem which Cplex cannot close gap lesser than 99% Gap in 1 hour time (Same for more time).

Reduced MIP has 3913 rows, 15749 columns, and 55341 nonzeros.
Reduced MIP has 3552 binaries, 0 generals, 0 SOSs, and 0 indicators.

Any parameter tuning suggestion or model assistance techniques will be a great help.

I have attached Cplex log file.

Thanks in advance.

Did you check the various warnings printed in the log? In particular, the ones about potentially strange numbers and about multi-objective optimization?

It seems your model has a blended multi-objective. Solving them in a lexicographical way may be better.

Originally posted by: ORA17

Tried to reduce number of decimal points in constraints coefficient, but that did not help much.

Also as of now  modelling in Multi objective way is expensive for me.

Is there anything did you notice that I should take care rather than remodel?

What actions can be taken in order to get rid of CPLEX warning 1036 ?

Thanks.

Do you have constraints with fractions for coefficients, such as 3x + 1/2 y <= 11/13? If so, try multiplying both sides by the least common denominator (if not too large) to get integer coefficients, such as 78x + 13y <= 22.

Paul

Originally posted by: EdKlotz

What did you mean when you wrote:

>   Also as of now  modelling in Multi objective way is expensive for me.

Did you mean that creating multiple problems yourself would be time consuming for you in terms of the modeling API or language you are using?   If so, then use the multi objective feature that comes with CPLEX.    You can try it quickly by exporting an LP file of your model, then modifying the objectives.   See the CPLEX manual for details.

Or, did you mean that you already tried CPLEX's multiobjective feature, but that it too ran slowly?   If so, please upload the log file (with the multi objective display parameter set to 2) so we can see what's happening in that run.

Looking at the log file you provided, progress in both the best node and best integer stalls after the first 6 minutes of the run.    It's not clear whether the large gap is because much better solutions can be found, or because the dual bound (i.e. best node value) is weak.   Running CPLEX's multi objective feature on this model might help answer this question if you haven't done that yet.   

Regarding parameter settings to try, given that progress is lacking in both the primal and dual bound, I would suggest the following, i.e. aggressive cuts, symmetry and probing to try to tighten the formulation, and applying the RINS heuristic to find additional solutions.   Even if this run doesn't yield good solutions, it might shed more light on where the primary performance challenge lies.

CPXPARAM_MIP_Cuts_Cliques                        2
CPXPARAM_MIP_Cuts_Covers                         2
CPXPARAM_MIP_Cuts_FlowCovers                     2
CPXPARAM_MIP_Cuts_Implied                        2
CPXPARAM_MIP_Strategy_Probe                      3
CPXPARAM_MIP_Cuts_GUBCovers                      2
CPXPARAM_MIP_Cuts_Gomory                         2
CPXPARAM_MIP_Cuts_PathCut                        2
CPXPARAM_MIP_Cuts_MIRCut                         2
CPXPARAM_MIP_Cuts_Disjunctive                    2
CPXPARAM_Preprocessing_Symmetry                  5
CPXPARAM_MIP_Cuts_ZeroHalfCut                    2
CPXPARAM_MIP_Cuts_MCFCut                         2
CPXPARAM_MIP_Cuts_LiftProj                       2
CPXPARAM_MIP_Cuts_LocalImplied                   2
CPXPARAM_MIP_Cuts_BQP                            2
CPXPARAM_MIP_Cuts_RLT                            2
CPXPARAM_MIP_Strategy_RINSHeur                   200