During the processes of transformer data acquisition, transmission, and storage, various interferences introduce significant deviations and missing values in the original monitoring data. Thus, it is crucial to enhance data quality through effective data cleaning. To address the limitations of traditional transformer data cleaning methods in handling abnormal data imputation, this paper proposes a transformer state data cleaning method based on an improved Kernel Extreme Learning Machine (KELM). First, the Nutcracker Optimization Algorithm is employed to optimize the hyperparameters of KELM. Then, the regression characteristics of KELM are utilized to clean the data, enabling the repair of outliers and the imputation of missing values, thereby effectively improving the quality of transformer state data. Experimental validation using transformer state data from a region in China demonstrates that the proposed method is simple and efficient, simultaneously enhancing the accuracy and completeness of the dataset. Furthermore, the proposed method outperforms other existing classical methods in abnormal data cleaning and can be effectively applied to improve the quality of transformer state data.