Electricity theft detection in integrated energy systems considering multi-energy loads

Abstract

The significant progress has been made in electricity theft detection, but most classic works focus on electricity theft detection in residential environments, neglecting other locations such as hotels, industrial plants, and street lights. Moreover, these works typically limit their scope to power systems alone, without considering heating and cooling systems. To this end, this paper aims to discuss the electricity theft detection in integrated energy systems where industrial plants are typically categorized. Firstly, we conduct a theoretical, qualitative, and quantitative analysis of the correlation between multi-energy loads (i.e., electrical, heating, and cooling loads), which provides insights into the motivation for considering these correlations in electricity theft detection. After that, multi-energy loads are projected into graphs where adjacency matrices represent their correlation and feature matrices denote their consumption readings. Furthermore, a Chebyshev graph convolutional network (ChebGCN) is proposed to detect malicious users by capturing latent features and correlations from the graphs. Simulation results demonstrate that the incorporation of heating and cooling loads can significantly enhance the performance of various machine learning models for electricity theft detection. Additionally, the detection performance of the proposed ChebGCN is consistently better than both classical and state-of-the-art machine learning models, no matter whether the fraud rate of the dataset is low or high.

Keywords

• False Data Injection
• Electricity Theft
• Advanced Metering Infrastructure
• Deep Learning
• Integrated Energy Systems