Attack detection problems in the smart grid are posed as statistical learningproblems for different attack scenarios in which the measurements are observedin batch or online settings. In this approach, machine learning algorithms areused to classify measurements as being either secure or attacked. An attackdetection framework is provided to exploit any available prior knowledge aboutthe system and surmount constraints arising from the sparse structure of theproblem in the proposed approach. Well-known batch and online learningalgorithms (supervised and semi-supervised) are employed with decision andfeature level fusion to model the attack detection problem. The relationshipsbetween statistical and geometric properties of attack vectors employed in theattack scenarios and learning algorithms are analyzed to detect unobservableattacks using statistical learning methods. The proposed algorithms areexamined on various IEEE test systems. Experimental analyses show that machinelearning algorithms can detect attacks with performances higher than the attackdetection algorithms which employ state vector estimation methods in theproposed attack detection framework.
展开▼
