Detecting and locating cyber and physical stresses in smart grids using the k‐nearest neighbour analysis of instantaneous correlation of states

Abstract

Monitoring the state of smart grids and detecting abnormalities are challenging tasks due to their large size, distributed nature, and complex and stochastic dynamics. Large deployment of sensors and monitoring devices, such as Phasor Measurement Units, provides a large volume of data about the state of the system, which presents both new opportunities and challenges in monitoring these systems. Different types of cyber-attacks have targeted the important function of monitoring smart grids and the integrity of the collected data. The multivariate time series of the state of components are analysed through their instantaneous correlation. The effects of multiple cyber-attacks and single line failures in the system on the instantaneous correlations are discussed. Moreover, a machine learning method is proposed to use the features selected from instantaneous correlations in a k-Nearest Neighbour classification for real-time detection and locating of stresses. The presented method shows a promising performance, which has been evaluated under a large number of simulated scenarios.