Federated Learning meets Electronic Digital Twins for Privacy-Preserving Hardware Trojan Detection in Smart Meter Infrastructures

The growing deployment of smart meters in modern power grids has introduced new security challenges, particularly the risk of Hardware Trojans (HTs), that can compromise system integrity and data reliability. Traditional centralized detection approaches raise concerns about data privacy and scalability. In this work, we propose a federated learning-based framework to detect HTs in smart meter networks, enabling decentralized model training without sharing raw data. Our method leverages local anomaly patterns while preserving user privacy and reducing communication overhead. A circuit-level modeling approach has also been proposed to develop suitable Electronic Digital Twins (EDTs), which are virtual representations of physical systems capable of replicating their dynamics and operational characteristics in real time. In critical infrastructures, such as smart meter networks, EDTs provide a safe environment to emulate typical metering behavior and HT attacks by applying controlled modifications to specific circuit components. Experimental results on simulated HT-injected datasets demonstrate the effectiveness of the proposed approach in accurately identifying malicious behavior, highlighting its potential for secure smart grid applications.