Forecast for wind power at ultra-short-term based on a composite model

It is valuable for the operations of power system by achieving high precision of prediction for the wind power at ultra-short term. To fully utilize the key information in the data of wind power and heighten the prediction precision, we construct a new combined method in this paper, which associates long short-term memory network (LSTM) and autoregressive moving average algorithm (ARMA), following the variational mode decomposition (VMD). In the beginning, the raw data is divided into three modes (mode one, mode two and decomposition error) by applying the VMD algorithm, since wind power data has the properties of strong intermittent and volatility, while it may lose key information in the process of decomposition. Then the LSTM algorithm is used to deal with the first mode, which presents the non-stationary tendency, while the ARMA method is adopted to handle the last two modes with different frequencies. Eventually, the results of each mode are summed to derive the total outcome. The application of Elia database on total wind power generation shows that the VMD-LSTM-ARMA method is appropriate for forecasting the wind power.