Evaluation of scenario reduction methods for stochastic inflow in hydro scheduling models

The long-term hydropower scheduling problem is inherently stochastic due to uncertainty in future reservoir inflow. We use Stochastic Dual Dynamic Programming (SDDP) to solve this problem. This work evaluate and compare three scenario reduction methods used to construct a multistage scenario tree which represents the underlying stochastic inflow process in the SDDP model. A case study is carried out to numerically assess the performance of the different scenario reduction methods. The performance is measured using out-of-sample simulation, simulating the solution strategies obtain with the various scenario models on an exogenously given set of inflow scenarios. Our results show that the choice of scenario reduction method impacts the solution to a hydropower operation planning problem substantially.