An End-to-End Cost-Focused Wind Power Forecasting Framework Based on Convex Hulls

The traditional prediction methods primarily emphasize accuracy. This letter presents an end-to-end training framework that integrates a convex hull support function with the Mean Squared Error (MSE) for cost-focused learning. Specifically, by introducing a convex hull loss, decision cost information is incorporated into the model's learning process, while the MSE term ensures that the predicted results align with the actual distribution. Experimental results on the IEEE-33 bus system demonstrate that the proposed method achieves a 12.4% improvement in cost reduction compared with traditional forecasting methods. Furthermore, the proposed approach significantly improves training efficiency compared with other decision-focused baselines. The proposed framework provides a generalizable paradigm for cost-aware forecasting under renewable uncertainty.