Utilização das redes neurais artificiais para previsão de radiação solar

Abstract

The use of solar energy obtained through solar cells installed in places with good incidence of solar radiation represents a great potential for electric generation. Given the great diffusion of this technology in Brazil, we investigated how the photovoltaic solar energy generated in a plant can be predicted through the correlation between the physical quantities: temperature and radiation. Solar energy requires a relatively high investment by the residential, commercial or industrial consumer, but it presents itself as viable, as it can produce savings in the monthly electricity consumption bill after its installation and full operation. This research took place within the scope of the Scientific Initiation Scholarship Program (PIBIC) developed at the IFPE campus Pesqueira, with the objective of studying how the ambient temperature influences the behavior of the incidence of solar radiation during a determined period. Considering that artificial intelligence techniques provide tools capable of estimating complex system data, we use artificial neural networks (ANN) to predict solar energy radiation, based on data provided by the IFPE campus Pesqueira solar plant. We developed several experiments to identify the most suitable configuration of the network. The artificial neural network used was of the Multilayer Perceptron type and the prediction was made using the neural network toolbox of the Matlab software. The result achieved was a network capable of predicting radiation based on temperature data with an average percentage of classification error of 1.21%, that is, a percentage of correct answers of 98.79%.