Short term photovoltaic power production using a hybrid of nearest neighbor and artificial neural networks

Abstract

In this paper, a hybrid forecasting method of the output power of a photovoltaic generation system based on statistical methods and artificial neural networks is proposed. This forecasting is made in two stages. The first stage consists on the prediction of two meteorological variables, ambient temperature and solar irradiance; these predictions are made using a series of artificial neural networks trained with conditioned data. This training data set is made from measurements obtained from a meteorological station, but preprocessed with the K nearest neighbor algorithm to avoid a set of unrelated training data. The second stage consists on the characterization of the photovoltaic panels that will be used to transform the solar energy into electricity; this task is accomplished by a characterization of the photovoltaic panels using artificial neural networks. Two study cases were performed and the forecasted output power was compared with the measured one.