Data-Driven Attack Detection and Identification for Cyber-Physical Systems Under Sparse Sensor Attacks

Abstract

This paper studies the issues of data-driven attack detection and identification for cyber-physical systems under sparse sensor attacks. First, based on the available input and output data sets, a data-driven monitor is formulated with two objectives: attack detection and attack identification. Then, with the subspace approach, a data-driven attack detection policy is presented, wherein the attack detector is designed directly by the process data. A subspace projection-based attack identification scheme is proposed via designing a bank of projection filters to determine the locations of attacked sensors. Moreover, the sparse recovery technique is adopted to decrease the combinatorial complexity of the subspace projection based identification method. The attack identification is recast into a block-sparse recovery problem. Finally, the proposed methods are verified by the simulations on a flight vehicle system.