A new deep learning based electricity theft detection framework for smart grids in cloud computing

Abstract

Electricity theft is a widespread problem in smart grids with significant economic and security implications. Although users’ electricity consumption patterns usually show obvious periodicity, they also exhibit considerable stochasticity and uncertainty. Existing mainstream electricity theft detection methods are the deep learning-based ones, which struggle to capture reliable long-term dependencies from the complex consumption data, leading to suboptimal identification of abnormal patterns. Moreover, the massive data generated by smart grids demands a scalable and robust computational infrastructure that traditional systems cannot provide. To solve these limitations, we propose a new deep learning-based electricity theft detection framework in cloud computing. At the cloud server, we deploy an electricity theft detector based on the auto-correlation mechanism, called the ETD-SAC detector, which progressively decomposes intricate consumption patterns throughout the detection process and aggregates the dependencies at the subsequence level to effectively discover reliable long-term dependencies from users’ electricity consumption data. Experimental results show that the proposed ETD-SAC detector outperforms state-of-the-art detectors in terms of accuracy, false negative rate, and false positive rate.