Federated Learning for Sustainable IoT Appliance Load Monitoring at the Edge Devices

Abstract Non-intrusive Appliance Load Monitoring (NALM) is an essential technique that enables tracking of household appliances' electricity usage, promoting eco-friendly practices and reducing energy monitoring costs. However, NALM implementation can be challenging due to privacy concerns, particularly in real-world applications. Federated learning is a promising solution that enables load monitoring model training and sharing while ensuring data security. Nevertheless, federated learning for NALM still faces various challenges, such as limited resources, edge model personalization, and scarcity of training data. In this paper, we propose a practical federated learning framework for NALM that addresses these challenges. Our approach utilizes collaborative data aggregation over federated learning, cloud-based model compression through filter pruning, and personalized edge and multi-task learning model building with unsupervised transfer learning techniques. Our experimental results, conducted using real energy data, demonstrate that our proposed load monitoring model achieves highly accurate personalized energy disaggregation, making it a state-of-the-art approach for non-intrusive appliance load monitoring at the edge client. By using our federated learning-based load monitoring model, we can minimize energy consumption while maintaining high learning performance and preserving user privacy. Future research could focus on further development and research to improve the efficiency of federated learning implementation and address remaining challenges in real-world NALM applications.