Data-Mining-based filtering to support Solar Forecasting Methodologies

TL;DR: The experimental findings show that the hybrid model, combining a SVM approach with a clustering mechanism, to filter the data used for training, achieved promising results, outperforming the approaches without clustering.

Abstract: This paper proposes an hybrid approach for short term solar intensity forecasting, which combines different forecasting methodologies with a clustering algorithm, which plays the role of data filter, in order to support the selection of the best data for training. A set of methodologies based on Artificial Neural Networks (ANN) and Support Vector Machines (SVM), used for short term solar irradiance forecast, is implemented and compared in order to facilitate the selection of the most appropriate methods and respective parameters according to the available information and needs. Data from the Brazilian city of Florianopolis, in the state of Santa Catarina, has been used to illustrate the methods applicability and conclusions. The dataset comprises the years of 1990 to 1999 and includes four solar irradiance components as well as other meteorological variables, such as temperature, wind speed and humidity. Conclusions about the irradiance components, parameters and the proposed clustering mechanism are presented. The results are studied and analysed considering both efficiency and effectiveness of the results. The experimental findings show that the hybrid model, combining a SVM approach with a clustering mechanism, to filter the data used for training, achieved promising results, outperforming the approaches without clustering.