Aiming at the problem of transmission and storage of massive electricity consumption information collection data in smart grid construction, a related data compression scheme is designed in this paper. First, the transmission protocol is parsed on the user side, and the majority of the application layer numerical data is decomposed. The numerical data is further classified. According to the different distribution characteristics of the sub-class, a correlation compression algorithm such as self-encoding network or differential encoding is used to perform experiments to remove redundancy. Finally, the analysis and reporting of abnormal values are carried out, which is of great significance for the future construction of smart grid.