Modelagem preditiva de geração de energia eólica utilizando redes neurais e dados climáticos

Abstract

In recent years, wind energy has gained significant importance in the Brazilian energy matrix, driven by global concerns about climate change and the substantial contribution of emissions from the burning of fossil fuels. In this scenario, wind energy emerges as a crucial alternative to mitigate and control the greenhouse effect, sparking a growing interest in dedicated studies in this field. This study investigates public data from wind power plants in Brazil with the purpose of applying them in a machine learning model capable of predicting the electrical generation of these facilities. To ensure the robustness of the results, public climatic data is integrated, exploring the relationship between meteorological data and wind energy production. An LSTM neural network model was developed, and a thorough analysis of the data was conducted, employing preprocessing techniques such as outlier removal and handling of missing data. Additionally, the feasibility of using climatic data was evaluated, considering the geographical distance between the wind farms and meteorological stations. This comprehensive approach seeks not only to anticipate wind energy generation but also to understand the direct influence of weather conditions on this process. The data analysis resulted in a set of 13 wind farms applied to the LSTM model. The results demonstrate that the use of climatic variables positively influences energy generation predictions.