Previsão do desempenho de um sistema solar fotovoltaico conforme dados meteorológicos da região

Abstract

Photovoltaic solar energy has been disseminated all over the world, and in Brazil this energy source has been getting considerable space in recent years, being stimulated mainly by the energy crises that the country faces. The use of photovoltaic systems offers several advantages, both to the final consumer and the country, contributing to the diversification of the energy matrix and reduction of dependence on fossil sources. However, associated with the benefits, as in any real system, when installed in regions with low incidence of solar irradiation this technology presents a loss in the efficiency of energy generation. Therefore, it is necessary to implement a preliminary study of the climatic conditions from the place where the photovoltaic system will be installed. As an alternative to this consideration, a study of the prediction of power production before its installation could be carried out, based on the local climatic information that directly influences the power generation, verifying the feasibility of the system implantation, and avoiding a investment. Therefore, this work aims to predict the viability of installing photovoltaic systems in other places, through an Artificial Neural Network (ANN). For this, the 3kWp photovoltaic system located in the Federal University of the Semi-Arid (UFERSA) Federal Campus was used as a model, in order to predict its behavior when installed in the states of Rio Grande do Sul and Pernambuco. The RNA was trained and validated with the support of Matlab®, and the training was implemented by inserting the input variables the data of temperature and solar irradiation of the city of Mossoró-RN, provided by the UFERSA weather station, and the generated electricity of the photovoltaic system was inserted as output variable. In order to enlarge the research and obtain the best result, two types of network with different operational proposals, feedforward and feedback were used in the training. After the implementation, some validations were carried out with different situations to verify the generalization capacity of each network in predicting the wanted output. Later, it was realized the power generation forecast in the states of Rio Grande do Sul and Pernambuco, through the methods of performance analysis, that the results are favorable for this application using feedforward network, since the feedback network did not perform well for a reduced amount of inserted samples