I'm Kai, a graduate student.
I recently worked on Benders decomposition to solve MILP (master problem is MILP, subproblem is LP) and read some papers. They always declare that Benders decomposition can be faster than solvers (Gurobi or others) with their improved method. However, I tried to use callback, pareto-optimality cut，something like them to improve the classic Benders decomposition. Although the code works and can provide optimal solution, it takes much more time than Gurobi. (P.S., Gurobi is so good.)
So, my question is if you guys have some experience that your own coding Benders can be faster than Gurobi. If so, how you improve in the Benders algorithm.
Thank you in advance and so appreciated if you can share your experience.
Please sign in to leave a comment.