Distributionally Robust Hub Location
Research output: Journal Publications and Reviews (RGC: 21, 22, 62) › 21_Publication in refereed journal › peer-review
|Journal / Publication||Transportation Science|
|Online published||16 Jun 2020|
|Publication status||Published - Sept 2020|
|Link to Scopus||https://www.scopus.com/record/display.uri?eid=2-s2.0-85092734632&origin=recordpage|
We study the adaptive distributionally robust hub location problem with multiple commodities under demand and cost uncertainty in both uncapacitated and capacitated cases. The hub location decision anticipates the worst-case expected cost over an ambiguity set of possible distributions of the uncertain demand and cost, and the routing policy, being adaptive to the uncertainty realization, ships commodities through selected hubs. We investigate the adaptivity and tractability of the distributionally robust model under different distributional information about uncertainty. In the uncapacitated case in which demand and cost are independent and costs of different commodities are also mutually independent, the adaptive distributionally robust model is equivalent to a nonadaptive classical robust model and the second-stage routing decision follows an optimal static policy. We then relax the independence assumption and show that the second-stage routing decision follows an optimal scenario-wise policy if either the demand or the cost is supported on a convex hull of given scenarios. We extend our analysis to the capacitated case and show that the second-stage routing decision still follows an optimal scenario-wise policy if the demand is supported on the convex hull of given scenarios. In terms of tractability, for all mentioned cases, we reformulate the distributionally robust model as a moderate-sized mixed-integer linear program, and we recover the associated worst-case distribution by solving a collection of linear programs. Through numerical studies using the Civil Aeronautics Board data set, we demonstrate the advantages of the distributionally robust model by examining its superior out-of-sample performance against the classical robust model and the stochastic model.
- Adaptive optimization, Distributionally robust optimization, Hub location