Datadriven false data injection attacks against cyber-physical power systems

Abstract

Power systems are accelerating towards the transition to cyber-physical power systems (CPPS). Such CPPS include myriads of sensors that generate huge amounts of data. The information collected from all these sensing components enables, not only the enhancement of CPPS performance in terms of efficiency and reliability, but also the expansion of the threat landscape. Among the attack vectors, false data injection attacks (FDIAs) demonstrated that can severely impact energy management routines of CPPS. Existing data-driven approaches used to design FDIAs are often based on different assumptions and environmental conditions which could make them not realistic, and more importantly, detectable by bad data detection (BDD) algorithms. In this paper, we present existing data-driven FDIA methods evaluated under different conditions of measurement data. In addition, we propose a novel data-driven attack strategy based on robust linear regression (RLR). For all data-driven attacks, appropriate conditions are considered in terms of measurement data to develop evaluation case studies. The results show that our proposed RLR method performs better than other data-driven methods in most scenarios, even in the presence of outliers.