Slack removed by singularity and infeasibility in model?

[davidliqhfx]

Background: I am working on a energy system model (RMIP) where it minimizes the total system cost, while accounting for energy demand balancing constraints on a certain spatial and temporal resolution. I also have constraints on the upper limit of the possible energy production by the generation technologies, at the time slides of the temporal resolution. My solver is CPLEX.

Issue: Sometimes while CPLEX shows finding an optimal solution, the inequality constraint of production upper limit would get relaxed (and thus not truly an optimal solution) and lines of lines of “slack removed by singularity” are produced before the “optimal” solution was being found. Any advice on how to avoid of this happening and making this constraints more binding?

Thank you for any advice!
Also: I cannot share my code.

[bussieck]

Numerical instabilities in an algorithm is sometimes caused by scaling issues. GAMS/Cplex has a "dataCheck=2" option (part of GAMS/Cplex option file) that can hint at model parts that can cause trouble. It is always best to "fix" things at the model level (hard to help if you can't share the source), but you can also try some GAMS/Cplex options, e.g. scaInd=1, numericalEmphasis=1, epRHS=1e-9, and epOpt=1e-9. The latter two only take effect for the two simplex algorithms.

-Michael