Ultra-short-term wind vector prediction based on a two-stage decomposition

Abstract Accurate wind speed and direction forecasting play a crucial role in promoting wind energy utilization. Based on the correlation of wind speed and direction and the non-stationarity of wind speed and direction sequences, this paper proposes a wind vector prediction model based on improved SSA and parameter optimization VMD two-stage decomposition. First, mutual information analysis of wind speed and direction is conducted, and the results show that wind speed information is beneficial for reducing the uncertainty of wind direction. After that, a two-stage decomposition wind vector prediction model based on improved singular spectrum analysis and parameter-optimized variational mode decomposition was con-structed. Wind speed and wind direction were first converted into east–west and north–south vector components; the component sequences, rather than raw circular direction angles, were then used as direct model inputs. The predicted component values were finally synthesized back into wind speed and wind direction to obtain the prediction data. Finally, the model and method pro-posed in this paper were applied to a dataset of a wind farm in Yunnan, China. Experiments showed that the wind vector prediction framework proposed in this paper could better capture wind speed and direction information in wind vector prediction. When the prediction errors of wind speed and direction were less than 1.0 m/s and 5°, respectively, the prediction accuracy reached more than 90 percent.