Publicação

Previsão da potência média produzida por turbinas eólicas utilizando redes neurais

Ver documento

Detalhes bibliográficos
Resumo:This thesis presents the development and evaluation of artificial neural network models applied to short-term wind power generation forecasting in a wind farm. The main objective of the study was to explore the performance of Multi-Layer Perceptron (MLP) and Long Short-Term Memory (LSTM) architectures, using real data from wind turbines collected by SCADA systems between 2016 and 2017 in Portugal. Data preprocessing included normalization techniques, seasonal decomposition, and the construction of sliding windows for temporal modeling. Several configurations of MLP and LSTM were implemented, varying in the number of neurons, hidden layers, and training strategies, including early stopping and different data partitioning approaches. The evaluation employed statistical metrics to assess forecasting accuracy, including RMSE, NRMSE, and R2. Early stopping and randomized data splits were analyzed to enhance model performance and robustness. The models achieved results above 83% for the coefficient of determination (R2). The main objective of this work was to develop MLP and LSTM models capable of accurately predicting the average power output of a wind turbine in a short-term horizon. To achieve this, different layer configurations, neuron counts, and validation techniques were tested for both approaches. Additionally, both single-variable and multivariable inputs were considered for the LSTM models. Finally, the three best-performing models were selected based on the evaluation metrics and compared with existing studies in the literature.
Autores principais:Campos, Letícia Góes
Assunto:Artificial neural networks Wind power Multi-layer perceptron Python Forecasting Time series prediction LSTM
Ano:2025
País:Portugal
Tipo de documento:dissertação de mestrado
Tipo de acesso:acesso aberto
Instituição associada:Instituto Politécnico de Bragança
Idioma:português
Origem:Biblioteca Digital do IPB

Registos relacionados

Registos relacionados