3 Presentations

• 
  10:30 AM - 10:55 AM

  A L-Shaped Method for Mid-Term Hydro Scheduling Under Uncertainty

  • Pierre-Luc Carpentier, presenter, Polytechnique Montréal
  • Michel Gendreau, Polytechnique Montréal
  • Fabian Bastin, Université de Montréal

  We propose a new approach for solving the hydrothermal generation scheduling problem (HGSP) under uncertainty. The aim of this problem is to coordinate hydroelectric generation, thermal generation and market transactions while minimizing the expected net operating cost. We consider the mid-term planning horizon which typically cover 6-60 months with weekly of monthly time steps. We partition the set of time periods in two consecutive stages. Each stage typically corresponds to several months. We assume that random parameters are driven by a time- and space-correlated stochastic process which loses memory of previous realizations at the end of the first stage. We exploit the special structure of the resulting stochastic program using a Benders decomposition method. We apply this method on Hydro-Québec's power system over a two-year horizon.

• 
  10:55 AM - 11:20 AM

  The Role of Hydrological Variable in Stochastic Dynamic Programming Apply to Hydropower Reservoir Operation

  • Pascal Côté, Rio Tinto
  • Quentin Desreumaux, presenter, Université de Sherbrooke
  • Robert Leconte, Université de Sherbrooke

  We study the impact of the hydrological variable in Stochastic Dynamic Programming (SDP) to solve optimization problems of managing a Hydropower System in British Columbia. We will demonstrate that using a real-time snow water data as the variable in SDP management policies proves to be of best effective, safer management, compared to a Markov or order p autoregressive model.

• 
  11:20 AM - 11:45 AM

  Adaptive Monitoring of the Progressive Hedging Penalty Parameter in the Context of Reservoir Systems Management

  • Luckny Zéphyr, presenter, Université Laval
  • Pascal Lang, Université Laval
  • Bernard Lamond, Université Laval

  Reservoir systems operations problems are in essence stochastic. This leads to very large stochastic models that may not be easy to handle numerically. We revisit the decomposition method developed by Rockafellar and Wets (1991) by proposing new heuristics to initialize and dynamically adjust the penalty parameter of the augmented Lagrangian function. Numerical experiments are realized to compare our heuristics to the traditional strategy of setting the parameter to a fixed value.