At present, big data of power equipment condition shows explosive growth. Equipment failures and personnel errors may result in dirty data, affecting data quality and subsequent analysis results. Therefore, data cleaning is of great significance. Most researches focus on identifying and eliminating abnormal data directly, which destroys the integrity of data. In order to solve this problem, a data cleaning method based on improved stack noise reduction autoencoder is proposed in this paper. Firstly, particle swarm optimization is used to optimize the hyperparameters in the stack noise reduction autoencoder. Then the characteristics of the stack noise reduction autoencoder is used to extract and restore the data features to clean the data. Data quality is improved by repairing isolated data and filling in missing data. The proposed method is simple and efficient, improving both accuracy and integrity of data. Finally, the historical operation data of power equipment is taken as an example. The results show that the proposed method has better data cleaning effect than other existing classical methods, and it has good cleaning effect for data sets with different abnormal degree in different running states. The method proposed in this paper can be used to improve the quality of power equipment status data effectively.