Deterministic versus stochastic dynamic programming for long term hydropower scheduling

Abstract

This paper is concerned with the performance assessment of deterministic and stochastic dynamic programming approaches in long term hydropower scheduling. In order to focus the analysis on the stochastic nature of inflows, only single reservoir systems are considered, where the so-called “curse of dimensionality” is not a concern. The paper reviews the framework of dynamic programming for hydropower scheduling, highlighting the differences between deterministic and stochastic approaches. The performances of both methods are evaluated through simulation with the historical inflow records considering hydro plants from different river basins in Brazil. The conclusions indicate that the deterministic approach leads to operation rules which maintain the reservoirs fuller than does the stochastic one, providing greater hydropower efficiency but also greater water spillage. Despite these differences, both approaches yield a similar performance, which suggests that no significant benefit is obtained from considering stochastic inflows in long term hydropower scheduling. Furthermore, in critical hydrologic periods where no spillage occurs, the deterministic approach leads to a better performance than does the stochastic one.