Enhancing Privacy in Smart Grids and IOTs Systems by Using Federated Learning: Case Study

Abstract

To enhance privacy in smart grids (SGs) and internet of thing (IoT) systems, a Federated Learning (FL) framework is proposed for practical application. By leveraging the idea of decentralizing model training and keeping raw data local, the framework addresses the privacy and security challenges associated with data collection on centralized servers. The framework achieves high accuracy (98.2% on MNIST, 85.6% on CIFAR-10) while resisting poisoning attacks and scaling efficiently by integrating differential privacy and secure aggregation. A case study on energy demand forecasting confirms its real-world applicability. The results demonstrate the potential of FL for scalable, privacy-preserving data analysis in IoT and SGs, with future work focused on integrating other privacy-enhancing technologies such as blockchain (BC).