The invention discloses a wind farm output power data restoration method considering the segmental characteristics of wind speed data. The method includes the following steps that: S1, repeated, missing and unreasonable abnormal data are screened out from obtained data, and the abnormal data are divided into continuous abnormal type data and local abnormal type data according to the lengths of continuous time sequences corresponding to the abnormal data; S2, as for the local abnormal data, an interpolation method is adopted to obtain restored wind farm output power data; S3, as for the continuous abnormal type data, whether the continuous abnormal type data contain segmentation points is judged through using normal data before and after the abnormal data and based on a maximum posterior probability, and an ARMA model is obtained through the normal data or a mode recognition method and based on the wind speed characteristics of each segment, and restored wind speed is generated through using the ARMA model based on the wind speed characteristics, and therefore, restored wind farm output power data can be obtained; and S4, the validity of the restored data is verified, and a restoration report is outputted. With the method adopted, the decision-making accuracy of auxiliary services in a power grid can be improved, and unnecessary system backup can be reduced.