4 présentations

• 
  10h30 - 10h55

  A multi-stage stochastic programming approach for an integrated maintenance and production planning with demand uncertainty

  • Hamid Reza Zarei, prés.,
  • Masoumeh Kazemi, Concordia University
  • Mustapha Nourelfath, Université Laval

  In this paper, we developed an integrated multi-stage stochastic programming model for production and maintenance planning problems for a multi-state system with random time to failure under uncertain demand for products in each stage. The proposed model is solved for small-sized problems and the results are compared with mean-value deterministic model.
  Keywords: Random demand, Maintenance optimization, Production planning
• 
  10h55 - 11h20

  Lagrangian dual decision rules for multi-stage stochastic integer programming

  • Maryam Daryalal, prés., University of Toronto
  • Merve Bodur, University of Toronto
  • Jim Luedtke, University of Wisconsin at Madison

  Multi-stage stochastic programs can be approximated by restricting policies to follow decision rules. This talk introduces Lagrangian dual decision rules (LDDRs) for multi-stage stochastic integer programs. We investigate techniques for using LDDRs to obtain bounds on the optimal value, and compare the strength of the relaxation from these different techniques.

  Keywords: Multi-stage Stochastic Integer Programming, Decision Rules, Lagrangian Relaxation
• 
  11h20 - 11h45

  Solving stochastic large-scale mixed integer linear problems for industrial production scheduling

  • Zayneb Brika, prés., École Polytechnique de Montréal
  • Michel Gamache, Polytechnique Montréal
  • Roussos Dimitrakopoulos, COSMO Stochastic Mine Planning Laboratory, Université McGill

  A new linear model is presented to address the topic of an open-pit mine production scheduling accounting for stockpiles and investment decisions in a stochastic context. The solution approach consists in first solving the linear relaxation using an extension of the Bienstock-Zuckerberg algorithm to the stochastic optimization. Then, a rounding heuristic based on the topological sorting is applied followed by a Tabu search with multiple neighbourhoods. A parallelization strategy is used to reduce the time spent creating the neighbourhoods. Real-sized instances are used to test the proposed method.
  Keywords: open-pit scheduling; stochastic mathematical optimization; Bienstock-Zuckerberg algorithm; parallelization.
• 
  11h45 - 12h10

  Investigating aggregate γ service level constraint in the stochastic lot sizing problem

  • Narges Sereshti, prés., HEC Montréal
  • Yossiri Adulyasak, HEC Montréal
  • Raf Jans, HEC Montréal

  In this research, we extend the stochastic programming model of the multi-item lot-sizing with γ service level for each individual product to a more practical setting where multiple service levels can be used in conjunction across multiple products to ensure that the business requirements are satisfied on an aggregate level.
  Stochastic Lot-Sizing, Aggregate Service Level