The prediction of inflows to the reservoirs of hydroelectric plants is of great importance in optimizing the operation planning, and aims to present a future scenario that may impact the energy generation process by increasing or decreasing the expected inflow. In this forecasting process, computational mathematical models based on neural networks are generally used. In this work we present a study of the application of Long Short-Term Memory (LSTM) Recurrent Neural Networks in the problem of forecasting the daily inflow of the Tucuruí Hydroelectric Power Plant (UHE) located in the Tocantins Araguaia Hydrographic Basin, in the horizon of 1 to 7 days ahead, considering the historical series of data measured by the National Water Agency (ANA) of UHE’s located upstream of its reservoir. The results obtained through the training of the model, showed the feasibility of its application to predict the daily inflow through the tests and analyzes carried out throughout the work, where the adjustment of each scenario presented was approximately 91% when the comparison was carried out. between the
computational values, with the original data portion of the set set aside for validation.
|
