Hello!
I’m wondering if there’s any way for JuMP (specifically Clp) to continue even when an infeasible solution is found during an SDDP system optimization.
Frequently running into problems where a primal solution is infeasible (yes this is an inherent issue with the model, and will be fixed eventually) however would like the solver to continue to run even when such an infeasibility is found, as is possible with AMPL.
Any direction would be greatly appreciated!
to continue even when an infeasible solution is found during an SDDP system optimization.
No. A fundamental assumption is that your system has relatively complete recourse and that it can never end up in an infeasible state.
Frequently running into problems where a primal solution is infeasible
This might be due to your model, but I find that Clp is unreliable and also reports feasible problems as infeasible after a number of iterations. Use Gurobi instead.
Read: https://odow.github.io/SDDP.jl/stable/tutorial/basic/06_warnings/
Thank you odow. Will look into Gurobi. We know our model’s feasibility space is convex so it’s likely an implementation / Clp issue.