Aplicação de redes neurais LSTM em medição e verificação de resultados de projetos com fontes renováveis: estudo de caso de projeto com sistema fotovoltaico

Abstract

The measurement and verification (M&V) campaigns of results in energy efficiency projects play a fundamental role in evaluating the real benefits arising from the implemented actions. In projects that include the implementation of solar generation, one of the independent variables considered in the process is “solar irradiation”, on which the variation of energy provided by the system depends. With this in mind, a methodology with the application of LSTM neural networks (LongShort Term Memory) for the prediction of "solar irradiation" is proposed, with the objective of reducing the time for monitoring the project results from twelve to six months, taking in given the difficulties faced in long periods of monitoring. The evaluation of the method applied to the problem in M&V for projects with a photovoltaic system (SFV), showed satisfactory results, identifying an alternative solution with the application of LSTM neural networks.