Elucidating the barriers in estimating false data injection attacks in process control systems

Abstract

Process control systems (PCSs) ensure efficient, safe, and high-quality chemical production. However, their growing reliance on communication networks increases vulnerability to cyberattacks, such as false data injection attacks (FDIAs). FDIAs modify data transferred over the PCS communication links. Addressing FDIAs requires detecting their presence, estimating their parameters, and mitigating their impact. This study examines the identifiability of FDIA parameters—precisely, the ability to estimate these parameters using observable process data. We present cases where attack parameters cannot be uniquely determined from observed data and assess their implications for attack estimation. First, we consider the case when the attack’s impact is indistinguishable from process disturbances or measurement noise. Second, we consider the case when the relationship between the observed data and the attack parameter values is not unique. Both cases result in a lack of identifiability in the attack parameter values from the process data. While the former is tied to inherent process and attack properties, the latter can be addressed using longer observation windows in estimation schemes. We explore these issues through different estimators and demonstrate their impact using a process example. Despite these challenges, we also present scenarios where accurate FDIA estimates can be achieved. Finally, we apply one of the estimators to a chemical process under simultaneous additive and multiplicative FDIAs, showcasing its effectiveness and validating its ability to estimate attack parameters accurately.