Advanced robust scheduling technology is one of the most economical and effective means to deal with output uncertainty. Usually

uncertainty set is used for describing the prediction error of wind power generation. It does not require the probability distribution of prediction errors and can cover all realizations of wind generation prediction errors with given confidence level. In our work

existing models of uncertainty set are evaluated using real data in wind farm operations. The study shows that unreasonable uncertainty sets may occur if the correlation of prediction errors is neglected. Furthermore

a criterion of wind generation uncertainty is proposed

based on which a new ellipsoidal modeling method is derived to incorporate the correlation of prediction errors. Given a confidence level in the sense of probability

the proposed uncertainty set can greatly reduce the measure of prediction errors with convexity guarantee. This facilitates mitigating conservativeness of robust optimization decision making process in power systems.