Corruption-Resilient Online Event Detection in Energy CPS- A Kernel PCA Approach

The overarching goal of the project is to detect spurious or maliciously injected sensor data originating from cyber-attacks that can have similar appearance as system's dynamic response and correct the same. Energy CPS is chosen as an example system. Thus far we have obtained positive results using principal component pursuit and robust PCA approach. In this poster, a Kernel PCA-based approach is used for event detection, which is resilient to data anomaly due to spurious outliers or false data injection attack on synchrophasor data stream. Promising results are obtained from simulation and field data. The broader impact and planned education and outreach activities are also outlined.