Modelos para Previsão de Potência Gerada e Identificação de Falha em Painéis Fotovoltaicos

Abstract

Solar energy is a promising alternative to meet the world’s growing energy demands. A photovoltaic system that has data storage technology allows the creation of a methodology for monitoring its operation, in order to reduce energy losses and maintenance costs. In this study, a technique is proposed using a hybrid model of graph convolutional neural network and recurrent neural networks with the objective to forecast the generated power of system. The model adjustment was performed and compared with models from previous studies. The developed one presented a result about 96% better than the other models, showing that considering the temporal and topological structure of the system is very relevant in the performance analysis.