Research on NWP wind speed error correction method based on EMD-Bi-LSTM modeling

In order to improve the prediction accuracy and computational efficiency of wind speed in large-scale wind farms, and to weaken the influence of numerical weather prediction (NWP) wind speed on the prediction accuracy of wind power. An empirical modal decomposition (EMD), and bi-directional long and short-term memory neural network (Bi-LSTM) combined with double-spanning residual network (DSRnet) improvement are proposed to correct the wind speed error of NWP. Firstly, the wind speed error data with time series characteristics are decomposed into the residuals (RES) and the intrinsic modal function (IMF) containing the same characteristics using EMD; Secondly, the components were predicted using an improved Bi-LSTM model; Finally, the results of the component predictions are used to reconstruct the corrected wind speed data. The results of simulation experiments in Pytorch framework show that the root mean square error and the average relative error of the former are 26.02% and 25.82% lower than those of the latter, respectively. It shows that this model has better prediction accuracy in correcting the error value of NWP wind speed.