Redes neurais artificiais e modelos de combinação para previsão de velocidade do vento
Wind power generation has been highlighted in Brazil due to the need to diversify the electrical matrix, which mainly depends on hydroelectric plants. However, wind speed is a resource that presents constant fluctuations throughout the day, months, and even years. In this context, wind speed predict...
| Autor principal: | Barchi, Tathiana Mikamura |
|---|---|
| Formato: | Dissertação |
| Idioma: | Português |
| Publicado em: |
Universidade Tecnológica Federal do Paraná
2022
|
| Assuntos: | |
| Acesso em linha: |
http://repositorio.utfpr.edu.br/jspui/handle/1/30180 |
| Tags: |
Adicionar Tag
Sem tags, seja o primeiro a adicionar uma tag!
|
| Resumo: |
Wind power generation has been highlighted in Brazil due to the need to diversify the electrical matrix, which mainly depends on hydroelectric plants. However, wind speed is a resource that presents constant fluctuations throughout the day, months, and even years. In this context, wind speed prediction is essential as it helps in the electrical system’s management, dispatch, and operation. In this sense, this research carried out an extensive comparison of traditional forecasting models, being the Linear Models (Auto-Regressive and Auto-Regressive and Moving Average models) and Artificial Neural Networks (Multilayer Perceptron - MLP, Radial Basis Function Networks - RBF, Extreme Learning Machine - ELM and Echo State Neural Networks - ESN ), versus combination models: Ensembles and Hybrid Error Correction Systems. The application of the methodology considered databases provided by the SONDA project (National Environmental Data Organization System) with wind speeds for the cities: Brasília, Florianópolis, Natal, Petrolina, and São Luís. The error of each predictive model was verified through the metrics Mean Absolute Error, Mean Squared Error, Average Relative Variance, and Index of Agreement. The ESN network reached 1st place in the performance ranking considering its placement for each city, being the model that presented the smallest error for Natal and Petrolina. The ARMA proved to be more suitable for specific cases of Brasília, Natal, and Petrolina. However, its classification in the general ranking was degraded due to its inaccurate performance in the other cases. The Ensemble occupied second and third places with Median Ensemble (Single Models without RBF) and Median Ensemble (RNAs without RBF), respectively. In the final ranking, Hybrid Systems were a negative highlight. |
|---|