Deceptive attack detection of control system using multi-scale principal component analysis

As industrial control system is threatened by sinusoidal attack, the mathematical model was established using a typical loop of the control system; the Fourier transform and wavelet analysis was used to analyze its different characteristics in the aspect of attack ability and concealment. Then, an algorithm was developed against sinusoidal attack using multi-scale principal component analysis (MSPCA) with online implementation. Simulation of sinusoidal attack on Tenessee Eastman (TE) process show that sinusoidal attack not only causes physical damage, but also conceals the damage. When the frequency of sinusoidal attack is higher, the attack can not be detected by the traditional principal component analysis (PCA) method, meanwhile, our method can detect the attack quickly and accurately. © 2018, Zhejiang University Press. All right reserved.