A novel transformer ordinal regression network with label diversity for wind power ramp events forecasting

Large-scale integration of wind energy and large-amplitude wind power fluctuation in minutes to hours, imposes unprecedented challenges of retaining reliable and secure power systems. Proper detection and precise prediction of wind power ramp events could help power systems better optimize scheduling and mitigate the impact of extreme events. Thus, this paper proposes a novel ramp event forecasting approach named Informer Ordinal Regression Network with Label Diversity (IFORNLD), without distributional assumptions and increasing the computational complexity. The proposed framework retains the superiority of the Informer architecture which can easily handle very long sequences with efficient computation and effective gradient in training, and the strengths of the ordinal regression with label diversity (ORLD) method, which creates a multi-output model that creates diversity with similar computational complexity. The result of case studies on three wind power datasets demonstrate that the proposed framework is superior than some benchmark models, validating its effectiveness and advantages.

Automated summary

This paper proposes a novel ramp event forecasting approach called IFORNLD, which effectively detects and accurately predicts wind power ramp events, outperforming some benchmark models.