Hybrid multi-stage decomposition with parametric model applied to wind speed forecasting in Brazilian Northeast

Wind speed forecasting is a challenging task, especially medium-term forecasting on 10 min sampling interval, due to the intermittent nature of wind. Several numerical methods were presented to improving wind speed forecasting accuracy, being the decomposition strategy reported as the key step to improve forecasting results, where the raw wind speed data is first decomposed into several patterns and then recomposed in a new time series. This can effectively reduce the uncertainty in the wind speed fore casting, leading us to believe that the precision of the forecast has a strong correlation with the preprocessing strategy, rather than with the forecasting technique. Based on this understanding, this paper combines two decomposition techniques, amplitude and frequency modulation-demodulation signal theory. The proposed decomposition approach is further combined with the parametric forecasting model named AutoRegressive Integrated Moving Average (ARIMA), resulting in a new ensemble learning method called Variational Mode Decomposition-Singular Spectral Analysis-AutoRegressive Integrated Moving Average (VMD-SSA-ARIMA). All forecasting results obtained by this ensemble are compared with those from the ARIMA. The ensemble results demonstrated stabilization of the forecasting errors. This indicates the ability of the proposed hybrid approach to decompose wind speed time series into uncorrelated components, reducing the errors from one up to a 12-steps-ahead forecasting horizon. (c) 2020 Elsevier Ltd. All rights reserved.

Automated summary

Wind speed forecasting is challenging, with decomposition strategy key to improving accuracy. The proposed method combines amplitude and frequency modulation-demodulation signal theory with ARIMA, resulting in a new ensemble learning method VMD-SSA-ARIMA. Ensemble results demonstrate stabilization of forecasting errors and reduction in up to 12 steps-ahead forecasting errors.