Light Robustness

Abstract

We consider optimization problems where the exact value of the input data is not known in advance and can be affected by uncertainty. For these problems, one is typically required to determine a robust solution, i.e., a possibly suboptimal solution whose feasibility and cost is not affected heavily by the change of certain input coefficients. Two main classes of methods have been proposed in the literature to handle uncertainty: stochastic programming (offering great flexibility, but often leading to models too large in size to be handled efficiently), and robust optimization (whose models are easier to solve but sometimes lead to very conservative solutions of little practical use). In this paper we investigate a heuristic way to model uncertainty, leading to a modelling framework that we call Light Robustness. Light Robustness couples robust optimization with a simplified two-stage stochastic programming approach, and has a number of important advantages in terms of flexibility and ease to use. In particular, experiments on both random and real word problems show that Light Robustness is sometimes able to produce solutions whose quality is comparable with that obtained through stochastic programming or robust models, though it requires less effort in terms of model formulation and solution time.