Smart Meter Data Analysis for Electricity Theft Detection using Neural Networks

Abstract

The major problem in electric utility is Electrical Theft, which is harmful to electric power suppliers and causes economic loss. Detecting and controlling electrical theft is a challenging task that involves several aspects like economic, social, regional, managerial, political, infrastructural, literacy rate, etc. Numerous methods were proposed formerly for detecting electricity theft. However, the previous works considered only the one dimensional (1-D) energy consumption data which apprehended the periodicity and were found inefficient in storing and retrieving the memory that resulted in a lower accuracy rate of detection. Hence, this research study intends Convolutional Neural Network combined with Bidirectional Long Short Term Memory based Recurrent Neural Network (CNN-RNN-BiLSTM) for overcoming the aforementioned problems in the detection of electricity theft. The CNN captures the global variables of 1-D data and identifies the non-periodicity and periodicity of 2-D electricity consumption records. RNN-BiLSTM extends the memory storage capacity of the neural network with bidirectional flow of information, thereby allowing learning order dependencies. The proposed method results of the predicted values of the electricity theft show better accuracy rate with reduced time during the training phase and reduced number of epochs. The proposed model helps to discriminate the customers for preventing fraudulent activities in the usage of electric power. The validation of the proposed method is carried out by comparing the method with the existing Support Vector Machine (SVM) and multi-class SVM models. The comparative results prove that the proposed CNN-RNN-BiLSTM model of electricity theft detection works efficiently.