Identification of Cyber Attacks on Water Distribution Systems by Unveiling Low-Dimensionality in the Sensory Data

Modern water distribution networks (WDN) are evolving into cyber-physical systems comprising physical components, such as valves and pumps, and a cyber-layer of computing, sensing and communication devices. It is recognized that these cyber-components are vulnerable to malicious attacks that can range from sensor data manipulation to alteration of the control laws that govern the operation of physical components. This paper presents a data-driven, model-free optimization-based solution to the BATtle of the Attack Detection Algorithms (BATADAL) competition that aims to investigate various approaches for the detection of cyber-attacks in WDNs. Leveraging machine learning and system identification, this paper develops algorithms for detection of cyber-attacks on WDNs. The methodology relies on unveiling low dimensionality components in the sensory data as well as on the sparse nature of anomalies a natural property of cyber-attacks against large volumes of data. The chief computational instrument is convex optimization, for which computationally efficient solvers are available. The proposed methods are tested on simulated cyber-attacks from the C-Town network in EPANET and the provided data-sets.