Detection Of Cyber-Attacks To Water Systems Through Machine-Learning-Based Anomaly Detection In Scada Data

Vulnerability of water infrastructure to cyber-attacks is proliferated as the operation of these systems is automated using networked systems. Reports of several cyber-attacks to water systems such as the 2006 Pennsylvania Water Company hack and the 2012 Tijuana River sewage spill has proven the water community concerns for the security of these systems. This study proposes a machine-learning-based anomaly detection methodology for early cyber-attack detection and warning that is tailored to water distributions systems security. Attacks are assumed to be manifested as man-induced anomalies in SCADA time series data that should be detected in a short time with low false negative/positive rates. The methodology is trained by and examined on the BATtle of the Attack Detection ALgorithms (BATADAL) dataset.