A Combined Data-Driven Method for Electricity Theft Identification

Electricity theft is one of the important sources of non-technical losses in the power grid, which has been an important problem faced by electricity enterprises. The data collected by the advanced metering infrastructure (AMI) makes it possible to identify electricity theft based on a data-driven method. However, among the data-driven methods, the methods based on supervised learning, unsupervised learning and state estimation have their advantages and disadvantages, which can not play a full role. This paper presents a combined data-driven identification framework, which combines the advantages of the supervised method (day-week-month load convolutional neural network (DWML-CNN)), unsupervised method (clustering by fast search and find of density peaks (CFSFDP)) and state estimation method. Firstly, the features are extracted based on the state estimation, then the unsupervised and supervised methods are used to determine the suspicion ranks respectively. Finally, a comprehensive evaluation is made according to the two suspicion ranks. Experimental results show that the framework has good performance on the data set of the State Grid Corporation of China (SGCC).