Estudo de previsão de irradiância solar usando regressão por vetores de suporte

Abstract

The intermittence of solar resource is one of the barriers for the use of photovoltaics plants in a electric power system. The growth of photovoltaic generation demands solar forecasting studies to allow the balance between consumption and electricity production. This work has as objective to present a study of solar global horizontal irradiance forecasting made from historical series of data on solar global horizontal irradiance collected in Fortaleza-CE between 2003 and 2005. The data series were divided in 70% and 30% for training and testing set respectively. The forecasting study was based on Support Vector Regression (SVR) algorithm. The algorithm was implemented in Python programming language through Scikit-learn machine learning library. The forecast horizon is 10 minutes. Afterwards predicted and measured data during testing period were compared and evaluation of errors were made. In general, it was observed that errors tend to increase during sunrise and sunset periods. From the analysis of the obtained results in this study, it was verified that 56,4% of forecasts got errors up to 15%, which represents a satisfactory result