Isolation of Integrity Attacks in Cyber-Physical Systems

This work introduces a robust methodology for isolating integrity attacks affecting a smart grid network, and in general cyber systems, i.e. systems controlled by an information and communication layer. The method approaches the problem holistically by classifying datastreams coming from the entire network into classes which represent its nodes. Features extracted from the wavelet and frequency domains are employed along with classifiers of diverse characteristics (Support Vector Machine, Gaussian mixture model, and reservoir network). A

wide range of integrity attacks is applied onto the IEEE-9 bus model which comprises the experimental platform. The isolation capabilities

of the proposed framework are shown by means of confusion matrices. We infer that the performance is more than satisfactory and proves the efficacy of the isolation-by-classification logic.