A great deal of basic data in the development of electricity smart grid have been constructed .These data are not only massive, high-frequency, dispersion and the nature of time and space, dynamic, complex data relationships,but also there is relevance and comparability between the data.Therefore, the traditional method of privacy protection will cause greater imformation loss and time consumption.Based on the idea of data classification this paper proposed a method which can protect the power rating information.Bottom-up generalization the quasi-identifier attributes, and randomized sensitive attributes,then generate a new data table after protection.The experiment results show that compared with the widely used algorithm MBF and GASCG,the method can greatly improve the efficiency of the privacy protection while reduce the less personal information and the effectiveness of data is greatly increased.