Semi-Automatic Data Labelling of Smart Meter Data for Electricity Theft Detection

Kwok Tai Chui, Lap Kei Lee, Ryan Wen Liu, Mingbo Zhao, Miltiadis D. Lytras

Research output: Contribution to journalConference articlepeer-review


The global penetration rate of smart meter installation is ever-growing. The smart meters provide 24/7 continuous recordings of electricity data which is comprised of useful information related to electricity usage and consumers' behaviors. Anomalies such as electricity theft can be detected using machine learning algorithms. To build a robust and accurate detection model, sufficient labelled data is important to ensure good generalization that adapts properly to unseen data. Nevertheless, manual electricity data labelling is costly and unrealistic in a large-scale population. In this paper, a semi-automatic data labelling algorithm is proposed based on deep convolutional neural network to label the electricity data, with limited amount of labelled data. Results reveal that the algorithm could serve as tradeoff between costly manual labelling and performance of the detection model.

Original languageEnglish
JournalCEUR Workshop Proceedings
Publication statusPublished - 2021
Event2021 International Conference on Smart Systems and Advanced Computing, SysCom 2021 - Virtual, New Delhi, India
Duration: 26 Dec 202127 Dec 2021


  • Advanced metering infrastructure
  • Data labelling
  • Data science
  • Electricity theft
  • Machine learning
  • Non-intrusive load monitoring
  • Smart grid
  • Smart meters


Dive into the research topics of 'Semi-Automatic Data Labelling of Smart Meter Data for Electricity Theft Detection'. Together they form a unique fingerprint.

Cite this