Abstract: Robust optimization, as an important tool for addressing the uncertainty of new energy output such as wind power, is widely used in microgrid optimization and scheduling. Traditional uncertain sets are not compact enough to accurately characterize wind power uncertainty, and there may be some outliers in the data enclosed by uncertain sets, resulting in overly conservative scheduling results. In response to the above issues, this paper proposes a two-stage robust optimization scheduling method for microgrids based on data-driven uncertain sets. Firstly, a conditional normal copula (CNC) model is constructed based on historical wind power data, and then the predicted values of wind power from the previous day are input into the CNC model to generate the next day's wind power samples. Then, a data-driven uncertainty set considering the temporal correlation of wind power is constructed through support vector clustering (SVC) and dimension decomposition. This uncertain set can more accurately depict the uncertainty of wind power and exclude outliers in wind power data, thereby reducing the conservatism of robust optimization while possessing outlier resistance. Secondly, a two-stage robust optimization scheduling model based on the aforementioned uncertain set is proposed and solved using the column and constraint generation (C & CG) algorithm. Finally, simulations prove that the uncertainty set constructed in this paper has lower conservatism compared to traditional uncertainty sets, and has good resistance to outliers in wind power data.