ST-BiGRU-Transformer: A novel spatiotemporal fusion model for wind power ramp event prediction with dual-task learning

Rising global energy demand and worsening environmental issues have highlighted wind power’s importance as a clean energy source, yet its inherent variability—especially wind speed-driven ramp events—undermines grid stability. To enhance the accuracy of predicting wind power ramp events, this paper proposes an innovative spatiotemporal prediction model. The model integrates temporal sequence variations and spatial wind speed differences by combining Simpler Yet Better Video Prediction (SimVP), Bi-Directional Gated Recurrent Unit (Bi-GRU), and Transformer techniques, effectively capturing the dynamic changes in wind power. It exhibits superior performance in scenarios with drastic wind speed fluctuations during ramp events, and it predicts wind speed and wind power simultaneously. The accuracy of ramp event prediction is further enhanced by incorporating wind speed analysis into the model. A dual-loss function multi-task optimization strategy is employed to facilitate feature sharing and information fusion, significantly enhancing model performance and effectively preventing overfitting. Experimental results demonstrate the model’s high accuracy and robustness in wind power forecasting, which provides reliable decision-making support for the operation and management of wind farms.