Quantifying the impact of scenario tree generation and reduction methods on the solution of the short-term hydroscheduling problem

This paper studies the properties of a stochastic optimization model for the short-term hydropower generation and reduction problem with uncertain inflows. The price of energy is not considered. The uncertainty of the inflows is represented using scenario trees. Backward reduction and neural gas methods are used to generate and reduce a full scenario tree. The objective of this work is to evaluate the impact of scenario tree generation and reduction methods on the solution of the optimization. First, statistical tests are done where the expected volume, the variance and the standard deviation of each scenario tree are calculated and compared. Second, operational tests are realized, where the scenario trees are used as input to the stochastic programming model and the value of the objective function and solution are evaluated and compared. The model are tested on a 14 forecasted days and for a 10 days rolling-horizon for two powerhouses with five turbines each located in the Saguenay-Lac-St-Jean region of the province of Québec in Canada.