Decision-Focused Surrogate Modeling with Feasibility Guarantee

Surrogate models are commonly used to reduce the computational complexity of solving difficult optimization problems. In this work, we consider decision-focused surrogate modeling, which focuses on minimizing decision error, which we define as the difference between the optimal solutions to the original model and those obtained from solving the surrogate optimization model. We extend our previously developed inverse optimization framework to include a mechanism that ensures feasibility (or minimizes potential infeasibility) over a given input space. The proposed method gives rise to a robust optimization problem that we solve using a tailored cutting-plane algorithm. In our computational case study, we demonstrate that the proposed approach can correctly identify sources of infeasibility and efficiently update the surrogate model to eliminate the found infeasibility.