Análise preditiva da geração fotovoltaica via algoritmos de inteligência computacional: modelagem e estudo de caso da Usina Solar Bom Jesus da Lapa - BA
Carregando...
Data
2022-01-27
Autores
Bergamo, Henrique Postingel
Ramos, Lucas Gomes dos
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Estadual Paulista (Unesp)
Resumo
O Brasil passa, atualmente, por um momento de diversificação de sua matriz energética. De um lado, ainda temos grande incerta com relação ao reestabelecimento anual dos reservatórios, e de outro vemos um crescimento e diversificação em ritmo acelerado das mais variadas fontes de energia alternativas. Neste cenário, estudos que possam diminuir estas incertezas, e ainda alavancar a confiabilidade dessas novas fontes de energia são de grande importância para o setor elétrico brasileiro. Neste cenário, em que se busca compreender e predizer dados futuros, modelos matemáticos computacionais têm sido aplicados com sucesso em diferentes áreas de conhecimento, com destaque especial para o setor elétrico. Desta forma, este trabalho objetiva estudar o comportamento da geração fotovoltaica de uma usina solar, definindo parâmetros que possam subsidiar estudos em outras usinas similares. Tal tarefa foi realizada a partir do uso de ferramentas de Análise Exploratória de Dados (AED) e de modelos de Aprendizado de Máquina (AM). A fim de viabilizar a metodologia proposta e conseguir validá-la em um cenário real, este trabalho combinou dados da usina solar de Bom Jesus da Lapa com dados meteorológicos da referida região, promovendo assim um estudo de previsibilidade da geração da energia elétrica. Para o desenvolvimento da pesquisa, foram exploradas quatro diferentes metodologias de AM: Florestas Aleatórias (Random Forest), Máquinas de Vetores de Suporte (Support Vector Machine), Aumento Extremo do Gradiente (XGBoost – Extreme Gradient Boosting), e Rede Perceptron Multicamadas (Multilayer Perceptron). Os modelos foram codificados e validados a partir de bases de dados reais de geração de energia fotovoltaica em conjunto com dados meteorológicos, cujo arcabouço computacional desenvolvido poderá servir para fomentar planos de eficiência energética tanto em âmbito governamental como no setor elétrico brasileiro em geral.
Brazil is currently going through a sensitive moment in the maturing of its energy one hand, we still have great uncertainty regarding the annual reestablishment of the reservoirs, and on the other we see a growth and diversification at an accelerated pace of the most varied alternative energy sources. In this scenario, studie matrix. On s that can reduce these uncertainties, and also leverage the reliability of these new energy sources are of great importance to the Brazilian electricity sector. In this scenario, in which one seeks to understand and predict future data, computational math ematical models have been successfully applied in different areas of knowledge, with special emphasis on the electricity sector. Thus, this work aims to study the behavior of photovoltaic generation in a solar plant, defining parameters that can subsidize studies in other similar plants. This task was accomplished through the use of Exploratory Data Analysis ( EDA ) tools and Machine Learning (ML) models. In order to make the proposed methodology viable and to validate it in a real scenario, this work combine d data from the Bom Jesus da Lapa solar plant with meteorological data from that region, thus promoting a study of the predictability of electricity generation. To develop the research, four different ML methodologies were explored: Random Forest, Support Vector Machine, Extreme Gradient Boosting, and Multilayer Perceptron Network. The models were coded and validated from real databases of photovoltaic power generation in conjunction with meteorological data, whose developed computational framework may serv e to foster energy efficiency plans both at the governmental level and in the Brazilian electricity sector in general.
Brazil is currently going through a sensitive moment in the maturing of its energy one hand, we still have great uncertainty regarding the annual reestablishment of the reservoirs, and on the other we see a growth and diversification at an accelerated pace of the most varied alternative energy sources. In this scenario, studie matrix. On s that can reduce these uncertainties, and also leverage the reliability of these new energy sources are of great importance to the Brazilian electricity sector. In this scenario, in which one seeks to understand and predict future data, computational math ematical models have been successfully applied in different areas of knowledge, with special emphasis on the electricity sector. Thus, this work aims to study the behavior of photovoltaic generation in a solar plant, defining parameters that can subsidize studies in other similar plants. This task was accomplished through the use of Exploratory Data Analysis ( EDA ) tools and Machine Learning (ML) models. In order to make the proposed methodology viable and to validate it in a real scenario, this work combine d data from the Bom Jesus da Lapa solar plant with meteorological data from that region, thus promoting a study of the predictability of electricity generation. To develop the research, four different ML methodologies were explored: Random Forest, Support Vector Machine, Extreme Gradient Boosting, and Multilayer Perceptron Network. The models were coded and validated from real databases of photovoltaic power generation in conjunction with meteorological data, whose developed computational framework may serv e to foster energy efficiency plans both at the governmental level and in the Brazilian electricity sector in general.
Descrição
Palavras-chave
Matriz energética, Fontes de energia alternativa, Modelos de Aprendizado de Máquina, Recursos energéticos, Aprendizado do computador, Energia solar, Geração de energia fotovoltaica