Electricity Theft Detection in Incremental Scenario: A Novel Semi-supervised Approach based on Hybrid Replay Strategy

Deep learning (DL) has achieved great success in the field of electricity theft detection (ETD). Most existing studies have used supervised mode to complete the DL-based ETD, but they do not have the capability of incremental detection, especially in small sample size scenarios. To address this prob...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:IEEE transactions on instrumentation and measurement Jg. 72; S. 1
Hauptverfasser: Yao, Ruizhe, Wang, Ning, Ke, Weipeng, Liu, Zhili, Yan, Zhenhong, Sheng, Xianjun
Format: Journal Article
Sprache:Englisch
Veröffentlicht: New York IEEE 01.01.2023
The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
Schlagworte:
Behavioral sciences
Biological system modeling
Boosting
Convolutional neural networks
Data models
Deep learning
Electricity
Electricity theft detection
Feature extraction
Irish smart energy trial
Support vector machines
Temporal convolutional attention networks
Theft
Variational autoencoder
ISSN:0018-9456, 1557-9662
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Deep learning (DL) has achieved great success in the field of electricity theft detection (ETD). Most existing studies have used supervised mode to complete the DL-based ETD, but they do not have the capability of incremental detection, especially in small sample size scenarios. To address this problem, this paper proposes a semi-supervised ETD approach based on hybrid replay strategy. From the data perspective, this paper designs a hybrid replay strategy that includes a variational autoencoder (VAE) and sample scrambling ranking (SSR) methods, and uses a "rehearsal" method to obtain incremental ETD capability. From the detection method perspective, this paper designs a semi-supervised ETD architecture that uses a temporal convolutional attention network (TCAN) as a feature extractor and uses contrastive learning to improve the utilization of unlabeled sensing samples, thus reducing the labeled sample size required for the fine-tuning process. Experimental results on the Irish smart energy trial (ISET) dataset show that the proposed scheme effectively solves the problem of incremental ETD in small sample size, and achieves 92.72%, 92.70%, 92.57% on accuracy, precision, and f1-score, respectively.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0018-9456
1557-9662
DOI:10.1109/TIM.2023.3324674