To reduce the harm of wind power ramp events in advance, an effective wind power ramp prediction system is needed. A new ramp prediction approach utilizing suitable selected time windows as units is proposed in this paper. Two risk factors are defined based on the performance analysis of ramp prediction, and the risk minimization principle is applied to build the risk model. Combining the correlation analysis and the statistical analysis of ramp duration, model constraints are studied. The optimal time window of industrial data is computed for ramp prediction based on a support vector regression model. Four evaluation indicators are chosen to verify that the proposed approach improves the performance of ramp prediction, and that the risk model is effective to select an optimal time window for prediction. (C) 2016 Elsevier Ltd. All rights reserved.