A Secure Federated Learning Approach to Smart Microgrid Stability Prediction

Abstract

This paper addresses the challenges posed by the proliferation of Internet-of-Things (IoT) based smart grids in modern power systems, which can threaten the stability and security of next-generation smart microgrids. Sharing sensitive information about suppliers and consumers to maintain communication with the primary grid can lead to breaches in confidentiality or availability, resulting in significant economic damage, loss of life, or national security threats. To reduce the risk of sensitive information exposure, the paper presents a secure federated learning framework that only allows microgrids to exchange their encrypted learned models that predict the grid's stability. The communication channel between the server and clients (microgrids) is authenticated using Transport Layer Security (TLS) protocols, and a Tree-based Group Diffie-Hellman (TGDH) group encryption scheme is employed to encrypt the model updates between the server and clients, ensuring the security of the data-link layer. Finally, the paper presents a comparative analysis to determine the impact of data sharing on the accuracy of stability predictions for each microgrid.