ELECTRICITY THEFT DETECTION WITH CONVOLUTIONAL NEURAL NETWORKS BASED ON SMART METERS DATA

Abstract

Electricity losses is one of the most important factor of the power quality. Electricty losses include technical and non-technical lossess. Technical losses are caused by iron and copper losses in electric equipments such as generator, transmission line, transformer, motor. Non-tecnical losses, on the other hand, are normally caused by management problem, inaccurate metering data and electricity thief. In non-technical losses, electricity theft is the most popular type and account for high percent. However, this type of loss is very difficult to detect traditional methods. In this study, we propose to use deep learning method, based on a convolutional neural network model combined with a attention mechanism to detect electricity theft. The model was trained on the dataset collected from 42372 customers for 147 weeks. The dataset was obtained from China electricity utility company. The results show that the proposed model can detect the user's electricity theft with the AUC accuracy of 92.2%.