I am relatively new to the concept of recoverable robustness. I am researching the robust version of an optimization problem. I currently have methods to address the problem with perfect knowledge. However, at the point of uncertainty realisation, there are a couple of scenarios that might affect the true solution:
- The objective value can become worse
- The objective value can become better
- The algorithm might not be able to find a feasible solution.
From my understanding, I should be able to re-use my algorithms with perfect knowledge and "fix" the produced solution so it produces the minimum amount of constraint violations and an objective value that is close to the optimal. How should one approach analysing the results produced from the uncertainty sets? Is that conceptually correct? Is someone aware of common recovery methods/strategies that can be used? (I presume one should not only rely on integer/linear programs to compute those.) I would appreciate if someone points me to some good literature on the topic.
A side note: Two other approaches I am considering:
- Modify the algorithms by relaxing constraints to produce a solution that can be used as a starting basis for a recovery.
- Use a more involved single algorithm. (I was unsure about this one as the literature suggests the algorithms with perfect knowledge should be able to produce a good enough solution that can be "fixed")