State Estimator and Machine Learning Analysis of Residual Differences to Detect and Identify FDI and Parameter Errors in Smart Grids

In the modern Smart Grid (SG), cyber-security is an increasingly important topic of research. An attacker can mislead the State Estimation (SE) process through a False Data Injection (FDI) on real-time measurement values or they can attack the parameters of the network that represent the system topology. While research has been done in SE bad data analysis, parameter attacks have proven to be difficult to distinguish from FDI attacks using physics-based techniques. Machine learning (ML) techniques have recently been used to enhance the detection of FDI attacks in an algorithm called Ensemble CorrDet with Adaptive Statistics (ECD-AS). ECD-AS, as developed, analyzes the real-time measurement values throughout the SG to detect FDI attacks. Parameter attacks, on the other hand, don't impact the measurements. Instead, the estimate of state variables of SE will be affected. This paper presents a collaborative framework of ML and SE for detection and identification of errors in system measurements and parameters. The residual difference space is introduced, created from the physics-based SE and ML-based measurement estimation processes, and analyzed by ECD-AS. A case study on the IEEE 118-bus system is presented. Numerical results show that the presented framework outperforms state-of-the-art method in detecting and identifying both FDI and parameter attacks.

Automated summary

Security in the modern Smart Grid is a significant topic, with attackers potentially misleading State Estimation through False Data Injection or attacking network parameters. Machine learning techniques are being used to enhance detection of FDI attacks, while distinguishing between parameter attacks remains a challenge.