Modelo Inteligente Utilizando Grafos para Previsão da Potência Gerada por Módulos de Painéis Fotovoltaicos

Abstract

The growth in demand for electricity in the world has intensified the alternative use of renewable energy sources such as solar energy. Solar panels are structures composed of photovoltaic cells, responsible for absorbing solar radiation and transforming it into electrical current. Photovoltaic Systems (PV system) performance analysis mainly helps to reduce energy loss and maintenance costs. However, most existing solutions for performance analysis need technical knowledge and often neglect spatial and temporal dependencies. In this paper, it is proposed a spatial-temporal graph neural network with the objective of automating the analysis of the PV system functioning. The classic technique using statistical models of time series is used for comparison purposes. The model adjustment was performed to predict future values, and performance metrics considering the model error were used to define the best fitted model. The model using neural networks and graphs presented better performance than the classic one, showing that considering the structure of the system is very relevant in the performance analysis.