Privacy-Enhanced Intrusion Detection and Defense for Cyber-Physical Systems: A Deep Reinforcement Learning Approach

Abstract

Cyber-physical systems (CPSs) will play an important role in future real-world applications through the deep integration of computing, communication, and control technologies. CPSs are increasingly deployed in critical infrastructure, industry, and homes to achieve a smart grid, smart transportation, and smart healthcare and to bring many benefits to citizens, businesses, and governments. However, the openness and complexity brought by network and wireless communication technology, as well as the intelligence and dynamic of network intrusions make CPS more vulnerable to network intrusions and bring more serious threats to human life, enterprise productivity, and national security. Therefore, intrusion detection and defense in CPS have attracted considerable attention and have become a fundamental aspect of CPS security. However, a new challenging problem arises: how to improve the efficiency and accuracy of intrusion detection while protecting user privacy during the intrusion detection process. To address this challenge, we propose a deep reinforcement learning-based privacy-enhanced intrusion detection and defense mechanism (PIDD) for CPS. The PIDD is composed of three modules: privacy-enhanced topology graphs generation module, graph convolutional networks-based user evaluation module, and the deep reinforcement learning-based intruder identification and handling module. The experimental results show that the proposed PIDD achieves excellent performance in intrusion detection accuracy, intrusion defense percentage, and privacy protection.