Attack Detection of CPS System with Multi-sensors

Cyber-physical systems (CPS) are next-generation intelligent systems based on environment-aware computing, communication, and physical elements. They are widely used in security-critical systems and industrial control. The interaction of information technology and the physical world makes CPS vulnerable to various malicious attacks, thereby undermining its security. This work mainly studies the attack detection problem of sensors in CPS systems with transient faults. This study considers a system with multiple sensors measuring the same physical variables, and some sensors may be malicious attacked and provide erroneous measurements. In addition, this study uses an abstract sensor model where each sensor provides the controller with an interval of possible values for the true value. Existing methods for detecting sensor malicious attacks are conservative. When a professional attacker manipulates the sensor's output slightly or infrequently over a period of time, existing methods are difficult to capture attacks, such as stealth attacks. In order to solve this problem, this study designs a sensor attack detection algorithm based on fusion intervals and historical measurements. First, the algorithm constructs different fault models for different sensors, integrates historical measurements into the attack detection method using system dynamics equations, and analyzes sensor measurements from different aspects. In addition, combined with historical measurement and fusion interval, the problem of whether there are faults when the two sensors intersect is solved. The core idea of this method is to detect and identify attack by using pairwise inconsistency between sensors. This study obtains real measurement data from EV3 ground vehicles to verify the performance of the algorithm. The experimental results show that the proposed method is superior to the state-of-the-art algorithm, and has better detection and recognition performance for various attack types. Especially for stealth attacks, the detection rate and recognition rate are increased by more than 90%. © Copyright 2019, Institute of Software, the Chinese Academy of Sciences. All rights reserved.