Previsão de velocidade do vento com série temporal multivariada utilizando rede neurais recorrente
Previsão de velocidade do vento com série temporal multivariada utilizando rede neurais recorrente
-
DOI: https://doi.org/10.22533/at.ed.9282428046
-
Palavras-chave: MV-LSTM, Previsão, Energia Eólica
-
Keywords: MV-LSTM, Forecast, Wind Energy
-
Abstract: Investigating time series is crucial for monitoring and administering electrical systems, especially in the context of wind energy systems, where the ability to predict wind speed with high precision is critical. Such forecasting enables efficient energy dispatch and contributes to mitigating risks associated with the inherent volatility of wind conditions. This study proposes and implements a Multivariate Long Short-Term Memory (MV-LSTM) model to analyze wind speed data, including average, maximum, and minimum hourly measurements. Additionally, a univariate LSTM network was developed for comparison. The data used in this study were collected through a sonic anemometer, from January 1 to December 31, 2015, in a native cerrado ecosystem at Fazenda Água Limpa (FAL) in the Federal District, Brazil. The findings indicate that including maximum and minimum speed vectors significantly improves the accuracy of wind speed forecasts. Moreover, the MV-LSTM model showed a reduction in forecast lag in situations of abrupt wind speed transitions, addressing a challenge previously identified by Xie et al. (2021). It suggests that the model could be further enhanced by adding additional layers and integrating with other machine learning methods to optimize predictive performance.
- Reginaldo Nunes da Silva
- Dario Gerardo Fantini
- Mauro Sérgio Silva Pinto
- Rafael Castilho Farias Mendes
- Marlos José Ribeiro Guimarães
- Antônio César Pinho Brasil Júnior