Assessment of the Long-Term Hydrothermal Scheduling operation polices with alternative inflow modeling

Abstract

The Long-Term Hydrothermal Scheduling (LTHS) problem plays an important role in power systems that rely heavily on hydroelectricity. The purpose of the LTHS problem is to define an optimal operation policy that uses stored water as inexpensively as possible. A popular solution approach to this problem is called Stochastic Dual Dynamic Programming (SDDP). To incorporate the inflow uncertainties, the LTHS problem is modeled as a multi-stage linear stochastic problem. Although the Periodic Auto Regressive (PAR) model is considered the best model to forecast inflows, the PAR model may require nonlinear transformations that include nonconvexities in the problem. In the Brazilian LTHS problem, some modifications are applied in the PAR model to avoid the nonlinear transformations and negative energy inflow generation. However, these adjustments can still add nonconvexities to the problem. As a result, this paper describes three approaches to generate inflow scenarios that maintain the convexity in the LTHS problem and are compatible with traditional solution algorithms. We show the results considering a small hydrothermal configuration, as well as the Brazilian hydrothermal power system.