Análise de fator de sobre dimensionamento em sistemas fotovoltaicos com o uso de rede neural artificial

Abstract

Photovoltaic generation systems are today an alternative for those who want to invest in generating clean energy. Therefore, it is important that it is increasingly discussed, in academic and professional circles, how to correctly size these systems, generating an analysis that comes closer to the intended real generation. There is no consensus in academia on how the decrease in solar visibility during estimated periods reduces the energy generated in panels installed in rural and urban areas. A bibliographical review was carried out on the proposed topic, and studies from different places around the world were grouped in this work, elucidating the problem. Therefore, this study presents how to determine the oversizing factor to correct generation power in photovoltaic system projects, in regions with a long estimation period, in addition to determining the factors (parameters) that influence photovoltaic generation in these locations considered critical. For this, generation data from two photovoltaic systems were found, one installed in an urban area in Campo Grande/MS and one in a rural area in the city of Bela Vista/MS, which were later compiled together with meteorological data found from INMET. These data were implemented in an Artificial Neural Network (ANN) of the MultiLawer Perceptron (MLP) type, with the help of the WEKA simulator and, as a result, I hope that you obtain, with the smallest possible error, the correction factor in design for systems installed in regions with periods of drought, in addition to analyzing the difference between rural and urban systems in terms of dirt. The results were overwhelming, while error values of approximately 1% showed the efficiency in explaining and modeling the problem using MLP-type ANNs. Another important result was the scaling factor of around 4% for urban areas and 10% for rural areas, in systems without periodic maintenance. The study also shows the main meteorological factors that influence photovoltaic generation. Finally, we found the weights of Artificial Neural Networks that establish the knowledge of the network.