SGA-FL NIDS: A Similarity-Gated Asynchronous Federated Learning for Network Intrusion Detection

The rapid expansion of the Industrial Internet of Things (IIoT) has broadened the attack surface for cyber threats, making intrusion detection critical for securing the interconnection of heterogeneous endpoints and cross-domain business collaboration.However, existing approaches are prone to issues such as unstable global model convergence, high communication overhead, and significant update noise under non-independent and identically distributed (non-IID) data and link jitter conditions.To this end, this paper proposes a similarity-gated asynchronous federated learning-based network intrusion detection model (SGA-FL NIDS). It introduces similarity-gated asynchronous aggregation within the federated learning framework to filter stale or low-relevance updates. Concurrently, it employs a two-stage feature selection process combining Information Gain (IG) and Chimp Optimization Algorithm (COA) to reduce communication overhead without compromising detection accuracy.Evaluations on the Edge-IIoTset and Bot-IoT datasets show that the model converges rapidly and stably, with more balanced recognition across categories.The model attains an F1 score of 0.970 with an approximately 18\% reduction in communication overhead on Edge-IIoTset. It attains an F1 score of 0.969 with an approximately 16\% reduction in communication overhead on Bot-IoT.These results indicate that SGA-FL NIDS offers strong scalability and generalization in complex, dynamic, multi-domain IIoT environments.