3 Presentations

• 
  03:30 PM - 03:55 PM

  A hybrid Stochastic dynamic programming - Tabu Search approach for long-term energy planning

  • Yves Alain Mbeutcha, presenter, École Polytechnique de Montréal
  • Michel Gendreau, Polytechnique Montréal
  • Grégory Émiel, Hydro-Québec

  The long-term Energy planning can be modeled and solved using classical Stochastic Dynamic Programming (SDP). However, SDP fails to represent adequately the risk brought by some inflows hypothesis on energy reliability of the Hydro-Quebec’s power system. We propose a Tabu-Search approach to improve SDP policies performance.

• 
  03:55 PM - 04:20 PM

  A Least Square Monte Carlo method applied to the Kemano system

  • Nicolas Léveillé, presenter, HEC
  • Michel Denault, GERAD - HEC Montréal
  • Pascal Côté, Rio Tinto

  A hydropower management policy is built using a Least Square Monte Carlo method. The inflows are simulated by the corporate partner Rio Tinto, using a hydrological model. Numerical experiments are conducted on the Kemano system in British Colombia.

• 
  04:20 PM - 04:45 PM

  A Q-learning approach for short-term hydropower generation

  • Mahdi Zarghami, presenter, Ecole de Technologie supérieure
  • Fausto Errico, École de technologie supérieure

  Stochastic dynamic programming (SDP) has been widely applied to hydropower optimization. However, space-state discretization and the course of modeling might significantly deteriorate the SDP performances. In this study we explore the Q-learning algorithm for the short-term management of a multi-reservoir system. Computational results prove the efficiency of the proposed algorithm.