Intelligent and Secure Detection of Cyber-attacks in Industrial Internet of Things: A Federated Learning Framework

Abstract

The increasing integration of traditional industrial systems with smart networking and communications technology (such as fifth-generation networks, software-defined networking, and digital twin), has drastically widened the security vulnerabilities of the industrial internet of things (IIoT). Nevertheless, owing to the lack of sufficient instances of high-quality attacks, it has been incredibly difficult to resist the cyberattacks that directed at such a substantial, complicated, and dynamic IIoT. This work introduces an intelligent federated deep learning framework, termed FED-SEC, for automatic and early identification of cyber-attacks against IIoT infrastructure. In particular, a new convolutional recurrent network designed to detect cyberattacks within IIoT data. Then, a secure federated learning scheme presented to promote making use of mobile edge computing to enable the distributed IIoT entities to cooperate together to train a unified model for cyberattack detection in a privacy-preserved manner. More, a safe communication channel constructed via an improved Homomorphic Encryption scheme aiming to keep the model parameters secure against any leakage of inferential attacks, especially throughout the training procedure. Massive experimentations on multiple public datasets of IIoT cyberattacks proved the high-level efficacy of the FED-SEC in discovering different categories of cyber-attacks against IIoT and the superiorities over cutting-edge approaches.