MB10* - Stochastic and Robust Optimization 1

We study a competitive bi-level facility location problem with stochastic customer preferences and service levels. Customers choose facilities based on distance and service quality. We propose a single -level reformulation and a three-step deterministic policy (POLU) combining pessimistic, optimistic, and upgrade phases. Results show improved performance over deterministic and single-policy approaches.

We develop a Predictive-Prescriptive Analytics (PPA) framework for integrated production planning in sensor-driven smart factories (SFs). In the predictive layer, our approach uses sensor data to predict equipment reliability that affects the available production capacity. We explicitly avoid using point predicted deterministic parameters in our model to absorb prediction errors. Instead, we translate predictive uncertainty into covariate conditioned scenarios by exploiting the neighborhood structure of k Nearest Neighbors (kNN) and the leaf proximity structure of Random Forests (RFs). For each sensor state, we compute probability weights over historical outcomes and sample a finite scenario set that feeds a two-stage stochastic programming model. The prescriptive layer then jointly optimizes maintenance actions, production quantities, and machine activations over a short-term planning horizon. As the model grows fast in complexity, we develop a Progressive Hedging (PH) inspired fix-and-optimize approach. We evaluate the performance of our model through several experiment sets and finally assess the value of using our PPA framework relative to a conventional Sample Average Approximation (SAA) approach.

Keywords – Predictive Prescriptive Analytics, Smart Factory, Predictive Maintenance

We address hub location and flow allocation in capacitated agri-food networks distributing multiple perishable products through hubs. A key challenge is that distance-based deterioration is uncertain and varies across products and supplier–hub pairs. To account for this, we develop a budgeted robust optimization framework. Results show tractability and favor lower-deterioration suppliers.

This paper develops a stochastic facility location framework incorporating nonlinear congestion costs and uncertainty in waiting-time valuation. Using queueing theory and mixed-integer optimization, it compares linear, piecewise-linear, and robust models under minisum and minimax objectives, demonstrating how nonlinear and uncertain congestion significantly influence facility placement, system performance, and decision robustness.