PREVISÃO DE IRRADIAÇÃO SOLAR PARA OTIMIZAÇÃO DE SISTEMAS FOTOVOLTAICOS: UMA ABORDAGEM UTILIZANDO REDES NEURAIS

Abstract

Photovoltaic solar energy is a clean and renewable source of energy that uses solar radiation to produce electricity. Its development has been growing very fast in recent years due to technological improvement and government support for this electricity production source. One of the aspects that directly affects the efficiency of photovoltaic generation is the climatic factor, so studying the impact of climatological variables, such as solar irradiance, and proposing ways to mitigate such impacts is an important way to further provide for the growth of this source of energy generation. This study presents the development and evaluation of Long Short-Term Memory (LSTM) recurrent neural network models for medium-term solar irradiation prediction in the Redenção-CE region. Using historical solar irradiation data collected by a local weather station, three LSTM models with different architectures and hyperparameters were trained. The results demonstrated the models’ ability to capture complex temporal patterns and produce accurate predictions, with Model 003 standing out for achieving the lowest loss, MAE (Mean Absolut Error), and MSE (Mean Square Error) values. Analysis of the prediction plots revealed the models’ capability to reproduce daily seasonality and patterns present in the real data. These promising results pave the way for the application of LSTM models in optimizing photovoltaic systems in the region.