A Robust and Energy-Efficient Federated IDS for IIoT Using Spiking Neural Networks and Differential Evolution with Adversarial Resilience

Abstract

The Industrial Internet of Things (IIoT) is faced with increasing cybersecurity threats that require lightweight, fast, and resilient intrusion detection systems (IDS). This study presents a novel federated IDS framework that integrates federated learning (FL), Spiking Neural Networks (SNNs), and differential evolution (DE). The use of SNNs within a federated context is a rare and innovative contribution that enables effective temporal feature extraction from IIoT traffic. DE is employed as a global optimization mechanism, enhancing robustness and generalization beyond conventional federated aggregation. To further strengthen resilience, synthetic adversarial noise is injected during training, allowing evaluation in realistic poisoning scenarios. The system is validated on the Edge-IIoTset dataset under both Independent and Identically Distributed (IID) and non-IID conditions. Results demonstrate 98.6% accuracy, 98.2% F1 score, and only a 0.6% performance drop under attack, with a compact 2.4 MB model and low-energy footprint suitable for edge deployment. These findings highlight the originality and practical impact of the proposed IDS, advancing secure and scalable IIoT applications.