An Ensemble Deep Convolutional Neural Network Model for Electricity Theft Detection in Smart Grids

Abstract

Smart grids extremely rely on Information and Communications Technology (ICT) and smart meters to control and manage numerous parameters of the network. However, using these infrastructures make smart grids more vulnerable to cyber threats especially electricity theft. Electricity Theft Detection (EDT) algorithms are typically used for such purpose since this Non-Technical Loss (NTL) may lead to significant challenges in the power system. In this paper, an Ensemble Deep Convolutional Neural Network (EDCNN) algorithm for ETD in smart grids has been proposed. As the first layer of the model, a random under bagging technique is applied to deal with the imbalance data, and then Deep Convolutional Neural Networks (DCNN) are utilized on each subset. Finally, a voting system is embedded, in the last part. The evaluation results based on the Area Under Curve (AUC), precision, recall, f1-score, and accuracy verify the efficiency of the proposed method compared to the existing method in the literature.

Multi-layer defense algorithm against deep reinforcement learning-based intruders in smart grids

“…The algorithm calculates validation errors for the first training round to define a threshold. The threshold is set as shown in (17), where IQR stands for interquartile range. The model sends an attack signal once the test error exceeds the reconstruction error threshold.…”

“…Physical protection of all utilized assets in the network is expensive and impractical, especially in large-scale systems [17]. In fact, limited network information is always available that opens a gate for malicious activities.…”

A detection model of scaling attacks considering consumption pattern diversity in AMI

“…Owing to its disadvantages of one-way information flow, low user participation, etc., it has gradually been unable to adapt to the modern society. As a consequence, the smart grid emerges as the times require, which not only incorporates renewable energy resources such as solar energy and wind energy to support multiple energy supply, but also integrates the advanced metering infrastructure (AMI) to control the power layer, realizing the two-way flow of information and power (Zanetti et al, 2019;Rouzbahani et al, 2020;Choi et al, 2021;Sarenche et al, 2021;Chaudhry et al, 2022;Huang et al, 2022;Park et al, 2022). Specifically, smart meters which play a vital role in AMI are deployed in demand sides, i.e., users, to collect and upload information about power consumption and supply to the utility.…”

“…The utility then makes decisions on real-time pricing, and energy scheduling, among others, based on the uploaded information, and then feeds back the decisions to guide users supply and consume electricity smartly (Singh et al, 2017;Zheng et al, 2018;Choi et al, 2021;Chaudhry et al, 2022;Huang et al, 2022;Park et al, 2022;Verma et al, 2022). However, as smart meters are deployed in an open network environment, they are vulnerable to data integrity attacks, by launching which an adversary can seriously endanger the safe operation of the smart grid through tampering with the information in smart meters (Jokar et al, 2016;Hu et al, 2019;Jakaria et al, 2019;Yao et al, 2019;Zheng et al, 2019;Rouzbahani et al, 2020;Tehrani et al, 2020;Bhattacharjee and Das, 2021;Singh and Mahajan, 2021;Sun et al, 2021;Yan and Wen, 2021;Chaudhry et al, 2022;Mudgal et al, 2022;Verma et al, 2022). Therefore, the research on data integrity attacks detection is of significant importance and has become a research hotspot in the field of the smart grid (Jokar et al, 2016;Zheng et al, 2019;Tehrani et al, 2020;Ibrahem et al, 2021).…”

“…Recently, much work has been conducted on the detection for data integrity attacks in AMI, which is mainly divided into three categories (Jiang et al, 2014;Jokar et al, 2016;Yao et al, 2019), including statebased (Huang et al, 2013;Salinas et al, 2014;Leite and Mantovani, 2018;Lo and Ansari, 2013;McLaughlin et al, 2013;Aziz et al, 2020;Bhattacharjee et al, 2021b,a), game theory-based (Cardenas et al, 2012;Yang et al, 2016;Wei et al, 2018Wei et al, , 2017Paul et al, 2020) and classificationbased (Jokar et al, 2016;Singh et al, 2017;Ismail et al, 2018;Yeckle and Tang, 2018;Zheng et al, 2018;Fernandes et al, 2019;Jakaria et al, 2019;Punmiya and Choe, 2019;Zheng et al, 2019;Rouzbahani et al, 2020;Tehrani et al, 2020;Yan and Wen, 2021). As a result of the popularity of artificial intelligence technologies, the feasibility of machine learning to detect attacks in AMI has attracted much attention of a large number of researchers.…”