Multidimensional Prediction Regions of Wind Power for Robust Economic Dispatch

The dramatic increase in non-dispatchable wind power generation poses a significant challenge to power system operations. In this context, probabilistic forecasting becomes increasingly crucial for power systems, as it quantifies the uncertainty of volatile energy sources and supports various decisionmaking processes. This paper proposes a novel nonparametric probabilistic forecasting-based robust economic dispatch (NPF-RED) paradigm aimed at enhancing power system reliability and efficiency. The proposed NPF-RED methodology captures probabilistic information regarding uncertainties by evaluating the feature similarity between prediction targets and historical observations. A minimum volume uncertainty set in the form of multidimensional prediction regions (MPRs) based on similar samples is constructed using similar samples and optimized as a mixed integer programming (MIP). The proposed MPRs serve as inputs of subsequential operation decisions, which can maintain solution robustness while reducing conservativeness simultaneously. Comprehensive experiments conducted on IEEE 14-bus and 118-bus test systems using actual wind power data demonstrate that at a similar risk level, the NPF-RED approach effectively reduces the total cost by 2.56% and 0.50% in the 14-bus and 118-bus systems, respectively.