Economically-driven spatiotemporal collaborative correction of high-precision wind power forecasting curves: aiming to more practical scheduling

Accurate day-ahead wind power forecasting is crucial for power system stability and economic operation, yet existing models often fail to account for the asymmetric economic impacts of prediction errors in real-world grid operations. To address it, this study proposes a novel spatiotemporal collaborative correction framework that simultaneously optimizes forecast accuracy and minimizes grid integration costs. Based on multi-site information fusion, an advanced spatiotemporal convolutional neural network with attention model that effectively captures complex spatiotemporal patterns through a unique integration of convolutional neural networks and attention mechanisms is developed. The results demonstrate significant improvements over benchmark forecasts from professional forecasting company: the coefficient of determination increases by 1.67 % to reach an impressive 93.009 %, while mean absolute error and root mean square error are reduced by 8.85 % and 9.38 % respectively. More importantly, our economically-driven asymmetric loss function achieves a remarkable 37.24 % reduction in integration costs by strategically penalizing costly over-prediction errors. Comprehensive seasonal analysis reveals particularly strong performance during challenging winter conditions, with integration costs decreasing by 51 % and prediction reliability significantly improving. This work establishes a practical, cost-aware forecasting paradigm that bridges the gap between prediction accuracy and economic dispatch requirements in modern renewable energy systems.