Abstract: Wind power is highly uncertain and volatile, and the day-ahead prediction accuracy of wind power needs to be improved.In order to improve the effect of day-ahead interval prediction for the wind power, an interval prediction method of wind farm dayahead output based on robust multi-label generation adversarial is proposed. First, the Pearson correlation coefficient is used to analyze the correlation between wind power output and various meteorological factors and historical wind power output, and construct an original data set containing meteorological factors of numerical weather prediction（NWP） and wind power output.Then, the predicted daily wind power is removed from the original data set to obtain a clustered data set. The k-means clustering is carried out to get the original data set with cluster labels. Then, many labeled scenarios are generated based on the robust auxiliary classification generation adversarial network. Finally, according to the known historical wind power output and the factors obtained by the NWP, the cluster label of the day to be predicted is determined. In the generated scenarios, multiple scenarios with high similarity to the wind power factors of the day to be predicted are selected according to the corresponding cluster labels to form a similar scenario set. Based on the average value and upper and lower limits of the wind power of the day to be predicted in the similar scenario set, the point prediction and interval prediction results of the wind power in 24 hours on the day to be predicted（the next day） are obtained, respectively. The superiority of the proposed method is verified by taking the actual wind farm data in Northeast China as an example.