312 / 2024-04-24 12:45:42

Reliable Supply Network Design under Correlated Uncertainties in Supply, Demand, and Link

Reliable facility location,Correlated uncertain disruptions,Distributionally robust optimization,Cutting plane algorithm

（备注：收到了何荣川教授和陆晔教授的邀请）

Disruptive events in supply chains usually lead to uncertainties to the supply side, the demand side, and the links between them. These uncertainties are inherently correlated particularly when the disruptions are caused by natural disasters or systemic threats. This paper studies the supply network design problem under uncertain disruptive events, which can affect the demand side, the supply side (the availability of prepositioned inventory), and the links (the shipment capacities) between supply and demand nodes at the same time. We characterize the disruptive events with an unknown joint distribution, which belongs to an ambiguity set based on the marginal and cross disruption probabilities. The uncertainties across the demand and supply sides and the links between them are characterized by linear functions of disruptive events. A two-stage distributionally robust model is formulated to simultaneously minimize the fixed location-allocation cost, the inventory pre-positioning cost, and the expected transportation cost under the worst-case disruption distribution. To solve this challenging model, we deploy a cutting plane algorithm based on the Benders decomposition, where the separation problem to calculate the worst-case disruption distribution is solved by a column generation approach. We explore two interesting special cases focusing on bottleneck links and bottleneck inventory, respectively. For the first one focusing on bottleneck links with an application in disaster-relief network design, the robust model admits a tractable mixed integer linear programming reformulation. For the second one focusing on bottleneck inventory with an application in sourcing and capacity planning, the robust model is equivalent to a two-stage stochastic model after proving the closed-form worst-case distribution for the second-stage problem. Extensive numerical experiments, including a case study on the Jiuzhaigou earthquake, are conducted to validate the effectiveness and efficiency of the proposed models, reformulations, and algorithms.