Power prediction methods for offshore wind farm clusters: interpretable ASTGCN based on wind speed delay perception and spatial feature fusion

In offshore wind farm clusters, although wind can propagate freely, a significant wind speed delay exists among different clusters. This study reveals a significant phenomenon of wind speed delay between various wind farm clusters. Conventional wind power forecasting methods are unable to capture the dynamic coupling of wind speed delay, resulting in poor prediction accuracy. To address this issue, we propose an interpretable forecasting method based on ASTGCN. It incorporates wind speed delay awareness and spatial feature fusion. Firstly, a wind propagation matching algorithm and a corresponding training set generation method are proposed to reveal the true dynamic relationships. Secondly, ASTGCN is utilized to extract spatiotemporal features. Next, an adaptive gradient normalization method is adopted to facilitate training convergence. Additionally, a new graph construction method considering wind speed delay, namely the Fusion Graph Method (FG), is put forward. Validation using field measurements from European offshore wind farms shows that our proposed method remarkably improves the power forecasting accuracy of ASTGCN, confirming the necessity of modeling wind speed delay in wind power forecasting.