Federated Learning-Based Network Threat Detection System with Digital Twins for IoT Security

Abstract

With the widespread use of Internet of Things technology, network threats are increasing, posing a serious challenge to the security of Internet of Things systems. To address this challenge, an efficient network threat detection system is designed by combining digital twin technology and federated learning algorithms. The research first uses the decentralized and immutable characteristics of blockchain technology to securely store and verify the data in the Internet of Things network, while combining the digital twin technology to carry out virtual mapping of the Internet of Things entity, real-time monitoring of its status, and timely detection of potential threats. Subsequently, a comprehensive simulation of the Internet of Things system is conducted via the digital twin network to generate data samples. Following this, the federated learning algorithm is employed, enabling multiple participants to collaboratively train the model. This approach enhances the model's detection capabilities while safeguarding the privacy of local data. Additionally, a distributed architecture is adopted to facilitate the efficient processing and analysis of large-scale Internet of Things data. Finally, the proposed system is tested. The test results show that in terms of registration time, when the number of attributes is 10, the registration time of the research system is about 0.57 seconds, the registration time of the micro-step online threat intelligence platform is about 0.59 seconds, and the registration time of the Check Point Infinity platform is about 0.62 seconds. When the number of access policies is 10, the shortest encryption time of the research system is 0.52 seconds, the second is 0.54 seconds of the micro-step online threat intelligence platform, and the longest is 0.58 seconds of the Check Point Infinity platform. In comparison to the benchmark system, the proposed research system demonstrates superior efficiency during the registration and encryption phases. This is primarily attributed to the precise modeling of Internet of Things devices using digital twin technology and the efficient data processing capabilities inherent in the federated learning algorithm. Consequently, the research system offers a swifter and more effective solution for detecting network threats within the Internet of Things environment.