Operations Research Seminar: Jorge Vera Andreo

In many applications, decisions are made in various stages or horizons. For instance, aggregate production planning decisions are done in tactical horizons and then the detail is managed in short term planning. Optimization models have been used for long in this area and one typical problem is how to deal with the inconsistencies that arise many times from uncertainties present in the diferent decisions stages. Stochastic approaches like 2-stage models have been used in this context as well as robust approaches. On the other hand, some optimization problems could be more sensible to data perturbations than others and the more sensible ones will have a higher risk of being affected by uncertainties. Theoretical developments in recent decades have studied condition and geometric measures associated to optimization problems, which can assess the sensitivity of a model to data perturbation and, hence, to the risk of uncertainty exposure. In this work we show how to connect those measures to a context of 2-stage decision models. We illustrate and evaluate the computational solution of these problems in some simulated experiments as well as a real industrial application.