• Speaker: Mike Hewitt, Rochester Institute of Technology
  • Title: Scenario Grouping in Methods for Stochastic Network Design
  • Location: Room V129, Mathematics Building (Callaghan Campus) The University of Newcastle
  • Time and Date: 10:00 am, Fri, 2nd Mar 2012
  • Abstract:

    We present a technique for enhancing a progressive hedging-based metaheuristic for a network design problem that models demand uncertainty with scenarios. The technique uses machine learning methods to cluster scenarios and, subsequently, the metaheuristic repeatedly solves multi-scenario subproblems (as opposed to single-scenario subproblems as is done in existing work). With a computational study we see that solving multi-scenario subproblems leads to a significant increase in solution quality and that how you construct these multi-scenario subproblems directly impacts solution quality. We also discuss how scenario grouping can be leveraged in a Benders' approach and show preliminary results of its effectiveness. This is joint work with Theo Crainic and Walter Rei at University of Quebec at Montreal.

  • [Permanent link]