The participation of new energy reseurces in market trading needs accurate metering of new energy power generation. In order to establish a verification method for wind power generation metering data, a step-by-5tep k-means clustering algorithm based on discrete wavelet transform is proposed to explore the variation laws of wind power fluctuation, so that a basis for missing data fitting and abnormal data correction of wind power generation curves can be built. The frequency domain analysis of wind power generation data is carried out by using discrete wavelet transform, and the generation curve is decomposed into different frequency ranges. For the frequency domain results after multiple wavelets transforms, step-by-tep k-means clustering is carried out from the lowest frequency component to the highest frequency component. This processing method can etfetively ensure that the information of stroke fluctuation is not lost in the clustering process. Typical curves extracted based on clustering are used to fit the missing data. The proposed method is simulated by using the wind power generation data of some wind farms in East China, and the effectiveness of the proposed algorithm is verified.