Agrupamento de curvas de carga para redução de bases de dados utilizadas na previsão de carga de curto prazo

Abstract

This work presents the use of clustering techniques in load curves for the similar days method for load forecasting, in order to obtain a reduced data to achieve a faster computational algorithm, while achieving similar or superior performance compared to those obtained by the traditional method that makes use of the original data set. The method allows to perform similar day load forecasting using short-term historical data from the consumption of electricity at consumers level, and related data, which allow tracing analogies to a future day. Conventional implementations of the method are used for comparison and validation. The scenario that provides the data for the studies, as well as the equipment, and data preprocessing stage, are presented. The methodology is validated using the cluster silhoute analysis. With the MAPE values was possible to verify the forecast, indicating superiority of the method based on clustered load curves.