Redes neurais artificiais aplicadas à previsão de irradiância global horizontal no contexto de cidades energeticamente inteligentes

Abstract

Rapid population growth in the last decades and consequent accelerated urbanization have led to new urban problems that societies had not faced in the past centuries. Thus, the sustainable development of cities is compromised as a consequence of failure to meet the needs that arise. In this context, the smart cities emerged and, based on artificial intelligence, digital resources and communication technologies are proving themselves as a natural strategy to mitigate these problems. Among many areas served by a smart city, smart grids have gained focus, as global energy needs will grow by 30% until 2040. In addition, governments and society demand a solid insertion of renewable sources in order to guarantee the sustainability. The photovoltaic matrix is one of the renewable sources that fits this demand. The Brazilian government estimates that, until 2050, 13% of all residences in the national territory should be supplied by energy from photovoltaic production. However, its insertion is challenged by intermittent production, since the panels generate energy basically from Global Horizontal Irradiance, which is not uniform over time. Thus, an accurate forecast is beneficial because it reduces the costs and uncertainties besides avoiding annoyances due to the deviation between forecast and consumption. With the intention of predicting Global Horizontal Irradiance in the next hour (h + 1) on the campus of the Federal University of São Carlos, located in Araras- SP, it was used Artificial Neural Networks. A Multilayer Percetron architecture with Levenberg-Marquardt training algorithm was used, considering one and two hidden layers. The best results, in terms of the Root Mean Square Error (nRMSE) ranged from 5.9% to 6.8%. The data used as input signals to obtain these results were global horizontal irradiance, mean temperature and average wind speed. The prediction was accurate when compared to the literature.