Intrusion Detection in Smart Grid Measurement Infrastructures Based on Principal Component Analysis

Abstract

The extensive measurement infrastructure of smart grids is a vulnerable target for cyber attacks aiming at compromising reliable power supply. Thus, the detection of intrusion into the system and the identification of manipulated and false data is a key security capability required for future power systems. In this paper, we apply principal component analysis (PCA), together with a subspace analysis, to detect the presence of such false data injection (FDI) attacks. A key requirement for this method is a database of historical grid states that is used to compute the PCA transformation matrix. Each new grid state is then transformed based on this matrix to calculate its uncorrelated principal components. The presence of FDI attacks leads to a significant increase in the contribution of principal components that span the residual subspace. By comparing this projection against a threshold, the presence of compromised measurements can be detected. This is demonstrated by several case study simulations.