Previsão de vazões naturais afluentes médias semanais para usinas hidrelétricas

Abstract

Water resource engineering has been largely employing flow forecasting models to support decision making in several areas. The Previvaz and Smap Diário models have already been used by ONS to forecast average weekly flows for up to six operating weeks in order to provide the necessary inputs for the optimization of the hydroenergetic resource in the planning of the operation of the Brazilian electrical system. Aiming at adding efforts to improve this forecast, this work proposed and analyzed the results of a weekly version for the Smap model, which was generated based on the calculation mechanism of the monthly version of Smap. In the case study, the weekly Smap was calibrated for two sub-basins of the SIN: the Camargos HPP basin (Grande basin); and the Santa Clara HPP basin (Iguaçu basin). A back test was carried out for both basins for the period between September 2019 and September 2020, comparing the flow forecast of the Previvaz, daily Smap and weekly Smap models with the flows observed in the same period. As a result, the Previvaz model, in the first week foreseen, propagated the trend of flows observed in the previous week and in the other weeks it tended to overestimate the lowest flows and difficulties in capturing peak flows. Both versions of the Smap model showed an adjustment when the rainfall used to calculate the flows was the observed rain, however both showed a tendency to overestimate the peak flows when using the rain predicted by the CFS V2. It is concluded that the weekly version of Smap performed better than Previvaz in most scenarios, indicating that it can be a model used for planning the operation. In addition, the weekly version of Smap also showed better results in the fifth and sixth week of the forecast horizon, indicating that the weekly model can be used to improve the result of longer forecast horizons.