Reserve demand is an important constraint in the optimization for daily generation scheduling， which is also important for the safe grid operation. With large-scale wind power integration， the traditional method of the reserve setting based on conventional energy resources cannot meet the demand of the grid operation security. Considering the intermittence of the wind power， and economical and security factor， an associated model between prediction error of wind farm output and varying reserve capacity is established by using clustering analysis to sift the data of wind farm output. Analysis of actual system examples shows the effectiveness of the proposed model.