Volume 3: Innovative Solutions for Energy Transitions: Part II

Study on Energy Theft Detection Based on Customers’ Consumption Pattern Wen Xiong, Ying Cai, Li Wang, Yufan Zhang*, Qian Ai, Zhaoyu Li, Yue Wang, Shuangrui Yin



With the development of industrial Internet of Things (IoT) for smart grid, the amount of data in end user side increases sharply. However, the digitizing also brings great possibility for energy theft. Inspired by the good performance of deep learning models and the computation efficiency of the convolutional neural network (CNN), in this work, we present a deep CNN based energy theft detector. By learning the statistical pattern in the customers’ consumption pattern, the detector is supposed to make correct classification. In reality, the whole dataset tends to have a small portion of energy theft data. To overcome such data imbalance, data of malicious consumption behavior is synthesized according to the predicted energy theft patterns. The experiment is conducted on the open source dataset. The proposed method is compared with the support vector classifier (SVC)-based method. The results show that the proposed method is more robust against the changes of non-malicious consumption behavior and can achieve better classification performance. Moreover, accelerated by GPU, the proposed method is more suitable for real time detection.

Keywords energy theft detection, CNN, consumption pattern, SVC, IoT

Copyright ©
Energy Proceedings