![[2602.16181] Towards Secure and Scalable Energy Theft Detection: A Federated Learning Approach for Resource-Constrained Smart Meters](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.16181] Towards Secure and Scalable Energy Theft Detection: A Federated Learning Approach for Resource-Constrained Smart Meters
Summary
This article presents a federated learning framework for detecting energy theft in resource-constrained smart meters, addressing privacy and computational challenges while ensuring effective performance.
Why It Matters
Energy theft in smart grids poses significant economic and operational risks. This research introduces a scalable and privacy-preserving solution that leverages federated learning, making it relevant for enhancing security in smart grid infrastructures, especially as reliance on smart meters increases.
Key Takeaways
- Proposes a federated learning approach for energy theft detection.
- Addresses privacy concerns by integrating differential privacy techniques.
- Utilizes a lightweight multilayer perceptron model suitable for smart meters.
- Demonstrates competitive accuracy and efficiency on real-world datasets.
- Offers a scalable solution for next-generation smart grid infrastructures.
Computer Science > Machine Learning arXiv:2602.16181 (cs) [Submitted on 18 Feb 2026] Title:Towards Secure and Scalable Energy Theft Detection: A Federated Learning Approach for Resource-Constrained Smart Meters Authors:Diego Labate, Dipanwita Thakur, Giancarlo Fortino View a PDF of the paper titled Towards Secure and Scalable Energy Theft Detection: A Federated Learning Approach for Resource-Constrained Smart Meters, by Diego Labate and Dipanwita Thakur and Giancarlo Fortino View PDF HTML (experimental) Abstract:Energy theft poses a significant threat to the stability and efficiency of smart grids, leading to substantial economic losses and operational challenges. Traditional centralized machine learning approaches for theft detection require aggregating user data, raising serious concerns about privacy and data security. These issues are further exacerbated in smart meter environments, where devices are often resource-constrained and lack the capacity to run heavy models. In this work, we propose a privacy-preserving federated learning framework for energy theft detection that addresses both privacy and computational constraints. Our approach leverages a lightweight multilayer perceptron (MLP) model, suitable for deployment on low-power smart meters, and integrates basic differential privacy (DP) by injecting Gaussian noise into local model updates before aggregation. This ensures formal privacy guarantees without compromising learning performance. We evaluate our framewo...
