Proposta de um modelo preditivo para eventos de rampa de vento utilizando Random Forests

• Ano de defesa: 2024
• Autor(a) principal: CRUZ, Josélio da Conceição
• Orientador(a): OLIVEIRA, Denisson Queiroz
• Banca de defesa: OLIVEIRA, Denisson Queiroz , ALBUQUERQUE NETO, Francisco , ALMEIDA, Vinícius Albuquerque de , LIMA, Shigeaki Leite de
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Maranhão
• Programa de Pós-Graduação: PROGRAMA DE PÓS-GRADUAÇÃO EM ENGENHARIA DE ELETRICIDADE/CCET
• Departamento: COORDENACAO DO CURSO DE ENGENHARIA DA COMPUTACAO/DCCET
• País: Brasil
• Palavras-chave em Português: energia eólica; rampas eólica; matriz energética; sistema de energia; energias limpas; arvores de decisão; Random Forests.
• Palavras-chave em Inglês: wind energy; wind ramps; energy matrix; energy system; clean energy; decision trees; Random Forests.
• Área do conhecimento CNPq: Geração da Energia Elétrica; Ciência da Computação
• Link de acesso: https://tedebc.ufma.br/jspui/handle/tede/5237

Global investment in wind energy is growing, driven by its potential as a clean and renewable source, while facing the challenge of inherent volatility in wind production, especially concerning wind ramp events. These abrupt changes in wind speed pose significant concerns for the stability and efficiency of wind energy systems, electrical grids, and electricity markets. The dissertation focuses on predicting wind ramp events, underscoring the importance of enhancing the stability and efficiency of wind energy systems amidst volatile production. The main objective is to develop and validate a predictive model, specifically Random Forests, to anticipate these events, utilizing data obtained from LIDAR wind profilers. This strategy aims to mitigate challenges brought by wind generation variability, affecting both electrical grids and the electricity market. The methodology incorporates wind data collection and validation, employing Random Forests for event analysis and categorization, complemented by sensitivity analysis to test the model’s effectiveness. The choice of Random Forests is justified by their unique combination of simplicity, robustness, and performance. By combining multiple decision trees, Random Forests mitigate the risk of overfitting. Each tree is trained on a random subset of the data, preventing the model from becoming overly specific to the training set. Sensitivity analysis confirms the model’s ability to manage the complexity and variability of wind data, highlighting its utility in improving wind energy prediction and management. It is concluded that predictive modeling through Random Forests is a valuable contribution to the wind energy sector, enabling safer and more efficient integration into energy matrices. The implementation of these predictive models is recognized as an essential advancement in dealing with wind production volatility, fostering sustainability and energy security.