Enhanced collaborative intrusion detection for industrial cyber-physical systems using permissioned blockchain and decentralized federated learning networks

In the digital economy, the security of Cyber-Physical Systems (CPSs) is paramount. Despite the deployment of various Intrusion Detection Systems (IDSs) in industrial CPSs, the primary obstacles to the advanced research include class imbalance, privacy gap, model homogenization, and arbitrary aggregation. Thus, in this paper, a Permissioned Blockchain-enabled decentralized federated distillation Generative Adversarial Network (GAN), namely PB-fdGAN, is proposed for Collaborative IDSs (CIDSs) in industrial CPSs. The novel Local Differential Privacy (LDP)-external classifier GAN (ecGAN) addresses class imbalance and privacy concerns by integrating label conditions into the latent space and using Wasserstein distance for stability in semi-supervised learning. Additionally, a Decentralized Federated Distillation (DFD) scheme allows for collaborative model building without data exchange, enhancing privacy and diversity. Moreover, a Quality of Service (QoS)-consortium blockchain framework with a new QoS evaluation strategy ensures reliable and effective model aggregation. The experimental evaluation demonstrates the high effectiveness of PB-fdGAN in detecting various types of cyber threats and the superiorities over state-of-the-art IDS solutions.