Identification of Malicious Data Attacks in a Smart Grid Network Using Spectral Clustering

The growing incidents of cyber intrusion into the smart grid (SG) network has been a major challenge for protecting the critical infrastructures. During cyberattack, an adversary can exploit the vulnerability of the complex cyber-physical (CP) infrastructure and hijack the decision-making process executed at the control center. The erroneous data can develop significant disruption in the operational and economical activities carried out at the control center. Thus, early identification of malicious data is necessitated to prevent the possible hazardous operations and avoid blackouts in the network. The SG network being one of the highly complex network, the task of recognizing corrupt sensor measurement with minimal computational cost is of utmost priority. In this regard, dividing the large geographical network into several regions can be an effective solution to accomplish the detection of compromised sensor measurements. Considering the efficacy of clustering algorithms to distinguish datasets of different operating scenarios, the present work is designed to detect the intrusion of corrupt data based on a spectral clustering approach. The spectral clustering approach for the intrusion detection task has been executed by incorporating the sensitivity of branch power flows to bus power injections for a network topology and parameter-dependent injection shift factor (ISF) matrix. The identification of data attacks has been validated for the benchmark IEEE 9 bus system in real-time simulation environment.