Reasonable and accurate modeling of uncertainty set of wind farm outputs is prerequisite in the robust dispatch of power systems. The modeling and evaluation methods for uncertainty sets of wind farm outputs are studied. Considering the completeness of wind power data

several kinds of uncertainty set constructing methods incorporating wind power forecast error relevance are proposed. When wind power data is sufficient

the proposed methods gives an analytical relationship between the confidential probability and the detailed model of uncertainty sets

and the sets are built in the convex ellipsoid or polyhedron form with high engineering applicability. For power systems with insufficient wind power data

the proposed methods can help dispatch centers predict the power output uncertainty of wind farms more reasonably with limited information

and thus make a more economic and reliable dispatch strategy. Further

a method to evaluate general uncertainty sets for power output of wind farms is proposed. The appropriateness of the proposed uncertainty set constructing methods and the effectiveness of the evaluation methods are validated based on the historical data from Irish transmission system operator EirGrid. This work is supported by the Foundation for Innovative Research Groups of the National Natural Science Foundation of China (No. 51321005) and the Special Grand from China Electric Power Research Institute (No. XTB51201303968).