A Data-Centric Approach to Generate Invariants for a Smart Grid Using Machine Learning
Cyber-Physical Systems (CPS) have gained popularity due to the increased requirements on their uninterrupted connectivity and process automation. Due to their connectivity over the network including intranet and internet, dependence on sensitive data, heterogeneous nature, and large-scale deployment, they are highly vulnerable to cyber-attacks. Cyber-attacks are performed by creating anomalies in the normal operation of the systems with a goal either to disrupt the operation or destroy the system completely. The study proposed here focuses on detecting those anomalies which could be the cause of cyber-attacks. This is achieved by deriving the rules that govern the physical behavior of a process within a plant. These rules are called Invariants. We have proposed a Data-Centric approach (DaC) to generate such invariants. The entire study was conducted using the operational data of a functional smart power grid which is also a living lab.