POSTER: A Semi-asynchronous Federated Intrusion Detection Framework for Power Systems

Abstract

Federated Learning (FL)-based Intrusion Detection Systems (IDSs) have recently surfaced as viable privacy-preserving solution to decentralized grid zones. However, lack of consideration of communication delays and straggler nodes in conventional synchronous FL hinders their applications within the real-world. To level the playing field, we propose a novel semi-asynchronous FL solution on basis of a preset-cut-off time and a buffer system to mitigate the adverse effects of communication latency and stragglers. Furthermore, we leverage the use of a Deep Auto-encoder model for effective cyberattack detection. Experimental evaluations of our proposed framework on industrial control datasets validate superior attack detection while decreasing the adverse effects of communication latency and straggler nodes. Lastly, we notice a 30% improvement in the computation time in the presence of communication latency/straggler nodes, thus validating the robustness of our proposed method.