How Prodrisk solves the problem

Algorithm (combined SDP/SDDP)

Details of the algorithm are presented in [1] and [2]. We briefly explain the parts of the algorithm necessary to define a vocabulary used in the remainder of the documentation on the portal.

Prodrisk is a medium-term hydropower scheduling model treating inflow and power price as stochastic variables.

In each main iteration Prodrisk solves one forward simulation and a backward recursion. The purpose of a forward simulation is to find a set of reservoir trajectories and compute the expected profit (lower bound). The purpose of a backward recursion is to compute cuts for each stage that describe the relationship between the future expected profit seen from that stage and the reservoir level at the end of that stage. The backward recursion also provides the future expected profit seen from the beginning of the planning horizon, which provides an upper bound.

After Prodrisk has reached convergence, or the maximum number of iterations has been reached, a final simulation is performed. The final simulation is a final forward simulation using the strategy (represented by cuts) created. However, we allow the final simulation to include some more details than the standard forward simulation.

Prodrisk is sometimes run twice in order to create head coefficients in a first round and then use these coefficients in the final run. We will refer to this as the initial run and the final run.

Prodrisk decomposes the overall scheduling problem into decision problems representing one week. Each decision problem is formulated as a linear programming (LP) problem.