Combined Wind Speed Prediction Model Considering the Spatio-Temporal Features of Wind Farm

Ultra-short and short-term wind speed prediction is important in wind power systems, especially for turbine control and primary frequency regulation. Wind farms have more drastic wind speed variations and spatial features to be explored because the turbines are affected by spatial wake effects. For the characteristics of wind speed and wake effect, this paper proposes a TCN-ARIMA prediction model based on Kalman filter and graph convolutional neural network (GCN). Because of the instability of wind speed, the wind speed signal is divided into the principal component part and the unstable part by Kalman filtering, where the principal component part contains most of the information, and the spatial features between turbine and turbine are extracted by using multi-layer graph convolutional neural networks (GCNs); the temporal convolutional neural network TCN is used to predict the principal component, and for the unstable part containing disturbances part, the adaptive auto-ARIMA model is used to extract the temporal residual information that may be contained, Finally, the output is superimposed. The analysis shows that the method improves the capability of conventional wind speed prediction, and the auto-ARIMA model can also capture the residual time-series information, and the achieved prediction results have significant advantages.