Machine learning applications for real-time detection of cyber-physical attacks on water distribution systems

Τσιάμη, Λυδία-Μαρία; Tsiami, Lydia-Maria

dc.contributor.author	Τσιάμη, Λυδία-Μαρία	el
dc.contributor.author	Tsiami, Lydia-Maria	en
dc.date.accessioned	2020-12-14T16:54:01Z
dc.date.available	2020-12-14T16:54:01Z
dc.identifier.uri	https://dspace.lib.ntua.gr/xmlui/handle/123456789/52513
dc.identifier.uri	http://dx.doi.org/10.26240/heal.ntua.20211
dc.rights	Αναφορά Δημιουργού-Μη Εμπορική Χρήση-Όχι Παράγωγα Έργα 3.0 Ελλάδα	*
dc.rights.uri	http://creativecommons.org/licenses/by-nc-nd/3.0/gr/	*
dc.subject	Δίκτυα διανομής νερού	el
dc.subject	Μηχανική μάθηση	el
dc.subject	Στοχαστικά μοντέλα	el
dc.subject	Διάγνωση κυβερνοεπιθέσεων	el
dc.subject	Τεχνητά νευρωνικά δίκτυα	el
dc.subject	Water distribution networks	en
dc.subject	Machine learning	en
dc.subject	Stochastic models	en
dc.subject	Cyber-attack detection	en
dc.subject	Artificial Neural Networks	en
dc.title	Machine learning applications for real-time detection of cyber-physical attacks on water distribution systems	en
dc.title	Εφαρμογές Μηχανικής Μάθησης στη διάγνωση κυβερνοεπιθέσεων σε εταιρείες νερού σε πραγματικό χρόνο	el
heal.type	bachelorThesis
heal.classification	Διαχείριση Υδατικών Πόρων	el
heal.classification	Μηχανική Μάθηση	el
heal.language	el
heal.language	en
heal.access	free
heal.recordProvider	ntua	el
heal.publicationDate	2020-11-04
heal.abstract	Water distribution networks (WDN) deploy digital devices not only to monitor and control utility operations but also to increase automation and ultimately their efficiency. Although their digitalization is essential, it comes with a cost: it exposes the WDN to the risks of a Cyber-Physical System, i.e. cyber-attacks. The overall aim of this diploma thesis is to develop new and improve upon existing machine learning methods for cyber-physical attack detection on Water Distribution Networks. The innovation of this work resides in two main developments (a) the use of novel stochastic methods to generate the water demand timeseries needed to train existing machine learning models, in an effort to improve their overall performance in the presence of uncertainty and (b) the exploration and use of a novel family of machine learning methods that take both the spatial and temporal dimensions of a water network into account, in an effort to improve the ability of the model to represent the water network more accurately. To approach the first objective, we generate new, synthetic datasets for the study of cyber-physical attack detection on water distribution networks by performing simulations on a real medium size WDN under stochastically generated water demands. The second objective is approached by exploring the use of Spatio-Temporal Graph Neural Networks as cyber-physical attack detection tools. Finally, we test the detection performance of various ML algorithms (including SVDD, Autoencoder, Structural Convolutional Neural Networks) on our datasets and preexisting ones as well, and discuss.	en
heal.advisorName	Μακρόπουλος, Χρήστος	el
heal.committeeMemberName	Ευστρατιάδης, Ανδρέας	el
heal.committeeMemberName	Μαμάσης, Νίκος	el
heal.committeeMemberName	Μακρόπουλος, Χρήστος	el
heal.academicPublisher	Εθνικό Μετσόβιο Πολυτεχνείο. Σχολή Πολιτικών Μηχανικών. Τομέας Υδατικών Πόρων και Περιβάλλοντος	el
heal.academicPublisherID	ntua
heal.numberOfPages	161 σ.	el
heal.fullTextAvailability	false