Short-Term Wind Power Forecasting Based on Spatio-Temporal Adaptive Graph Convolutional Recurrent Network

The randomness and volatility of wind power pose significant challenges for short-term forecasting, requiring the model to capture both temporal dynamics and the spatial correlations among turbines. To address this issue, this paper proposes a Spatio-Temporal Adaptive Graph Convolutional Recurrent Network (STAGCRN). The proposed method dynamically constructs and updates the spatial relationship graph through node adaptive parameter learning (NAPL) and a data adaptive graph generation (DAGG) module, enabling more accurate modeling of spatio-temporal dependencies in wind power data. In addition, a spatio-temporal self-attention mechanism is introduced to enhance the model’s ability to capture both short-term fluctuations and long-term temporal patterns. By stacking multiple spatio-temporal adaptive graph convolutional recurrent layers, the model is capable of extracting complex nonlinear characteristics in wind power sequences. Experimental results based on real wind farm data demonstrate that the proposed method achieves significantly improved prediction accuracy and robustness compared with existing approaches in short-term wind power forecasting tasks.