Electricity Theft Detection Using Deep Bidirectional Recurrent Neural Network

Electricity theft causes significant harm to social and economic development. In the past few years, it has attracted much attention that electricity theft detection based on electricity consumption data can help to solve this problem. A major challenge is that there are no explicit features in electricity consumption records. However, the existing machine learning-based detection methods mainly suffer from the following two disadvantages. (1) Handcrafted features and shallow-architecture classifiers have poor detection accuracy. (2) Most methods consider electricity consumption as static and cannot capture both the internal time-series natures and external influence factors well. To overcome the above shortcomings, we propose a novel method called Electricity Theft Detection using Deep Bidirectional Recurrent Neural Network (ETD-DBRNN), which can capture the internal characteristics and the external correlation by learning the electricity consumption records and influence factors representation. Experiments on real-world datasets validate the effectiveness of our method.