Previsão de Vazões Diárias com Redes Neurais Artificiais e Impacto na Formação do Preço de Liquidação das Diferenças Horário

Abstract

As the forecast of inflows to hydroelectric plants is one of the input information in the process of SIN operation programming, it is important that these forecasts are increasingly assertive so that the outputs of this process are more consistent with the real conditions. The main goal of this work is to model a daily inflow forecast using neural network technique whose outputs are incorporated into the very short-term hydrothermal dispatch optimization program in order to analyze the impact on the spot price in the electric energy market. The results regarding the flow forecast showed that the predictor proposed was equivalent or more assertive than the models officially used in the energy sector for a considerable part of the hydroelectric plants evaluated, especially for the first forecasted day. On the other hand, the results referring to price indicated the close tracking of the curve created to the official reference, following even the movements of higher volatility. At the end, a price curve referring to the application of the realized flows is adopted, providing a conclusion about the affluence used as input data for the forecaster.