Abstract

The temperature of a photovoltaic panel has a significant influence on the generation of electricity. Several authors have developed models that allow to predict the panel temperature (Tp) from air temperature (Ta) and solar radiation. Schwingshackl et al., (2013) showed that including wind speed (v) in the models improves the accuracy of Tp prediction. In this work, Artificial Neural Networks (ANN) are used to determine the dependence of Tp of an operative photovoltaic power plant (PVP) on local meteorological variables. The study is based on data registered in the Luna PVP located in the semi-arid Coquimbo Region in Chile. Luna PVP is composed of polycrystalline panels and has a nominal power of 3.46 MW. It has sensors for Tp, Ta, v, global horizontal radiation (GSR), inclined radiation (TSR) and generated energy. Two cases are analyzed, in the first one, the ANN are trained with the Tp, TSR and Ta. Secondly, wind speed is added in the training. The two cases were compared with physical models found in the literature. It is found that considering v in the prediction of Tp improves the agreement between models and observations. Finally, we found that Tp values calculated with ANN models are more accurate than those calculated with physical models: the RMSE for physical models ranged between 3.2C and 7.2C, the RMSE for the ANN models was lower than 2.7C.