Accurate wind power forecasting is crucial for modern power systems, yet most deep learning models neglect spatial relationships between turbines. We propose GSTAformer, a graph-guided spatio-temporal model capturing both spatial and temporal dependencies through MIC- and PCC-built graphs; GraphSAGE for spatial feature extraction; multi-scale convolution for trend detection; and an improved Autoformer for temporal modeling. Experiments on SDWPF and GEFCom2012 datasets demonstrate GSTAformer’s superior performance, achieving a 24 h mean squared error (MSE) of 0.7480 and mean absolute error (MAE) of 0.6362 on SDWPF. This work integrates graph-based spatial modeling with enhanced temporal forecasting for medium-term wind power prediction, providing a coherent framework suited to complex wind energy scenarios.

High-precision wind power prediction improves grid stability and reduces curtailment losses. Existing methods face three limitations: static graphs cannot capture dynamic spatial correlations under weather changes, time series models miss multi-scale temporal features, and frequency-domain analyses lack physical constraints. We propose: (1) a dynamic distance correlation weighted graph that adaptively combines geographic and power correlations for weather–terrain coupling; (2) a spatio-temporal-frequency fusion framework integrating graph networks, bidirectional GRUs, and a patchwise sparse time–frequency module; (3) a turbine power curve-constrained frequency mixer for physical consistency. On the SDWPF dataset, our model achieves MAE reductions of 37.47–43.32% and RMSE reductions of 37.93–42.70% versus baselines, outperforming state-of-the-art methods. The approach demonstrates superior performance in complex spatio-temporal scenarios.