VI Trajectory Generalization Across Smart Meter Edge Clients using Federated Learning

Privacy preservation in smart home applications is increasingly vital with the rise of AI-generated content (AIGC), which poses the risk of misusing user profiles for nefarious purposes. This concern is especially relevant in Non-Intrusive Load Monitoring (NILM), where AI-powered algorithms detect and disaggregate the energy usage of electrical devices. Training deep neural networks for event detection in NILM typically requires diverse data sources, such as smart meter readings from household appliances. Sharing these data across networks can compromise the integrity of the data. Despite extensive research in this area, there is a lack of real-world evidence to support the effectiveness of proposed solutions. This paper introduces the use of Federated Learning (FL) to enhance privacy in NILM applications. We established an experimental lab environment with edge devices to test the viability of FL in realistic scenarios. Our laboratory experiments demonstrate that FL not only improves the generalization of machine learning models but also preserves data integrity, showcasing its potential for practical application.