Methodology for target forecasting of water level in hydroelectric plant reservoirs under conditions of low inflow

Abstract

The “El Niño” phenomenon brings periods of drought to northern South America that negatively impact the level of hydroelectric plant reservoirs, which could reduce their energy production. In order to avoid reaching the minimum operating level before the end of the drought period, this research proposes a methodology based on data science for the target forecast of the level of hydroelectric plant reservoirs in low flow conditions. The goal is that the minimum operating level of the reservoir be reached on the estimated end date of the drought period, that is, March 31, 2024. It is applied to the data of the reservoir of a hydroelectric plant located in the northwest of South America, for which three sequential forecast horizons are used, allowing the models to be evaluated as these periods pass, using the metrics: MAPE, RMSE, and MAE. To meet the goal, the predictive sampling method of the Prophet forecasting technique is used. The results indicate that the technique is a useful additional tool for the plant dispatcher, with values for the performance metrics during the third forecast horizon of 0.045%, 48 cm, and 62 cm, for the MAPE, the MAE, and the RMSE, respectively.