False Data Injection Detection in Nuclear Systems Using Dynamic Noise Analysis

The integration of nuclear reactors into the smart electric grid dictates the digitalization of Instrumentation and Control (I&C) systems, to support advanced networking capabilities (e.g., remote, semi-autonomous operation). However, continuous data transmission could potentially facilitate cyber events targeting nuclear installations. To secure nuclear communications against potential adversaries attempting to impersonate legitimate communicating parties, we propose a real-time processing scheme which does not rely on AI/ML, but uses noise analysis to verify the integrity of exchanged signals and commands. In this work, we evaluate the randomness of noise present in real-world nuclear I&C signals. After constructing a dataset consisting of steady state signals from PUR-1 all-digital reactor, three representative signals are used as case study. The disentropy of the autocorrelation function is used to quantify the amount of randomness present, while the signal observation time window is varied to determine the optimal monitoring length in a real-time cyber event detection scenario. We observe that several I&C signals contain additive components qualifying as white noise, which can therefore be used to evaluate their integrity. By performing several tests replicating cyber event scenarios, using original and adversary-manipulated data, we confirm the validity of our scheme for detecting False Data Injection (FDI) events.