Wind farm's real historical operating data is the basis of accurate prediction for wind power，output characteristics and the influence on the power system.Due to the fact that some historical operating data are mixed with missing and bad data，especially curtailed wind power data，this paper aims to extract the timing characteristics of curtailed wind power data，analyze them with segmental identification method and reconstruct fault data by two methods based on available historical data.One uses delayed correlation between wind farms' output to reconstruct data and the other adopts bidirectional time series weight ratio of auto-regressive moving average（ARMA）model to reconstruct time series by the wind farm output data itself.Finally，the real wind farm historical operating data in a China wind power base is studied as an example，which shows that proposed methods can identify and reconstruct the missing data effectively.