Abstract
Electric energy metering plays a crucial role in ensuring fair and equitable transactions between grid companies and power users. With the implementation of the State Grid Corporation’s energy Internet strategy, higher requirements have been put forward for power grid companies to reduce costs and increase efficiency and user service capabilities. Meanwhile, the accuracy and real-time requirements for electric energy measurements have also increased. Electricity information collection systems are mainly used to collect the user-side energy metering data for the power users. Attributed to communication errors, communication delays, equipment failures and other reasons, some of the collected data is missed or confused, which seriously affects the refined management and service quality of power grid companies. How to deal with such data has been one of the important issues in the fields of machine learning and data mining. This paper proposes a collaborative fitting algorithm to solve the problem of missing collected data based on latent semantics. Firstly, a tree structure of user history data is established, and then the user groups adjacent to the user with missing data are obtained from this. Finally, the missing data are recovered using the alternating least-squares matrix factorization algorithm. Through numerical verification, this method has high reliability and accuracy in recoverying the missing data.