Edge federated learning with adaptive optimization and lightweight design facilitates collaborative security situation awareness in the Industrial Internet of Things

Abstract

As the scale of the Industrial Internet of Things expands, its security issues become increasingly prominent. Traditional centralized security architecture faces challenges such as high data processing latency and easy privacy leaks. To solve the above problems, a collaborative security situation awareness model that integrates edge computing and federated learning is built to optimize the security protection capabilities. The federated learning is applied to implement distributed model training and complete cross-device knowledge sharing while protecting data privacy. Edge computing is used to optimize model lightweighting, and then pruning, quantification and other technologies are used to reduce computing overhead and improve real-time response efficiency. The model introduces three key improvements: an adaptive gradient update with momentum to accelerate convergence, a multi-dimensional threat quantification function for unknown attack detection, and an abnormal node isolation mechanism to enhance robustness against malicious participants. On IDS2025 and NSL-KDD datasets, Edge-Federated Learning based Security Situation Awareness (EFL-SSA) achieves a throughput of 370-420 items/second 18-22% higher than Federated Averaging (FedAvg), average latency of 50-72 ms (15-20% lower than FedAvg) and local detection accuracy (96.5%). In actual multi-factory tests, the false positive rate is less than 2.3%, and the data leakage rateis less than 3% under typical attacks. The research provides a new security situation awareness solution with real-time, privacy and scalability for the Industrial Internet of Things, which has practical application value for collaborative security protection in complex industrial environments.