Adaptive Parameterisation for Efficient Detection of False Data Injections

Increasing interconnectivity in modern safety-critical cyber-physical systems (CPSs) renders them susceptible to attacks like false data injection (FDI). Due to computation and communication resource constraints, it is infeasible to encrypt all data exchanges in such systems. As the other alternative, the lightweight statistical detectors are system-agnostic in nature; attackers can launch stealthy FDI attacks with a high degree of sophistication. This research introduces an adaptive parameterisation method for stateful anomaly detectors to fill this security gap. The study includes a theoretical analysis for statistical evidence of stealthy data falsifications. The proposed adaptive detection framework has the capability to continuously observe the system’s behaviour in real time, with the goal of rapidly detecting FDI incidents using this statistical evidence. We propose a novel parameter tuning strategy to guarantee early detection of FDI attacks, keeping the false alarms to a minimum. Its efficacy is evaluated in CPS case studies from different domains and in an automotive Hardware-in-the-Loop (HIL) setup.