10h30 - 10h55
Blackbox optimization of inflow scenario trees
Inflow scenario tree parameters are tuned using blackbox optimization methods. The scenario tree structure maximizing total energy production in a rolling-horizon scheme is compared to the energy production using median scenario of the inflows. Numerical experiments are conducted on a Rio Tinto hydropower system in Saguenay, Québec.
10h55 - 11h20
Comparison of model based and model free methods for water reservoir management
Stochastic dynamic programming and direct policy search using universal approximators for reservoir management are compared. The comparison focuses on the representation of the uncertainty and the consequence on the policy obtained. Numerical experiments are performed on the real case of Kemano located in British-Columbia, Canada.
11h20 - 11h45
Fast near-optimal heuristic for the short-term hydro-generation planning problem
Short-term hydro-generation planning can be modeled as a mixed integer linear program. However, for Hydro-Quebec’s system, the resulting MILP is too big to be solved in a reasonable time with commercial solvers. We developed a three phase-heuristic based on spatial decomposition that yields to fast near-optimal solutions. We will present this approach and give numerical illustrations on real problem instances.
11h45 - 12h10
The LSMC-S&D method applied to the Kemano system
A hydropower management policy is built using the Least Square Monte Carlo method with regression on the state and the decision (LSMC-S&D). Numerical experiments are conducted on RioTinto’s Kemano system in British Colombia.