Abstract: With the development of smart grids, a large amount of monitoring data is generated during the operation of power transmission and transformation equipment. However, due to data acquisition, transmission and storage problems, missing data often occur in practical datasets, which affects the accuracy of subsequent data analysis and equipment condition evaluation. To address this issue, this paper analyzes the characteristics and causes of incomplete data and investigates three representative imputation methods, including the K-nearest neighbor (KNN) method, predictive mean matching (PMM), and principal component analysis (PCA). Experiments are conducted on WINE, SPECTF and GLASS datasets from the UCI repository under different missing rates, and the performance of the three methods is evaluated using the root mean square error (RMSE).The results show that the PCA-based imputation method achieves lower RMSE and higher accuracy in most cases compared with PMM and KNN methods, is suitable for high-dimensional power equipment data with strong variable correlations. Furthermore, potential improvements combining neighborhood information and machine learning methods are discussed.