Keywords
electricity theft detection, multi‑class, convolutional neural network, long short‑term memory network, terminals
Abstract
This paper addresses the difficulty of the accurately detecting electricity theft in complex grid environment and proposes a multi?category electricity theft detection method based on CNN?LSTM hybrid model. Firstly, the excellent feature abstraction ability of convolutional neural networks (CNN) is utilized to extract the non?periodic local features of one?dimensional electricity consumption data. Then, the long short?term memory (LSTM) is adopted to capture the correlation between daily power consumption data and extract periodic power consumption features to establish feature fusion layer network. After that, the feature vectors extracted by CNN and LSTM are horizontally splicing to obtain a new fusion vector. Based on this, the accurate detection of multiple types of electric theft behavior are realized. Experimental results show that the proposed method can accurately identify multiple types of electric theft behavior, and the detection results are more comprehensive and accurate than the existing detection methods.
DOI
10.19781/j.issn.1673-9140.2023.01.026
First Page
226
Last Page
234
Recommended Citation
LI, Jinjin; CHEN, Jueyu; and HUANG, Keying
(2023)
"Multi‑class electricity theft detection based on the CNN‑LSTM hybrid model,"
Journal of Electric Power Science and Technology: Vol. 38:
Iss.
1, Article 26.
DOI: 10.19781/j.issn.1673-9140.2023.01.026
Available at:
https://jepst.researchcommons.org/journal/vol38/iss1/26