Cybersecurity Defence Strategies of Distribution Synchrophasor in Smart Grids

Project summary

This project aims to develop innovative strategies for preventing critical infrastructures in bulk power systems from malicious False Data Injection Attacks (FDIA). Power grids, as typical cyber-physical systems (CPSs) are vulnerable to large scale FDIA that are hard to detect by existing approaches, often resulting in catastrophic power interruptions. This project will extract spatio-temporal signatures to detect potential FDIA on data from Phasor Measurement Units (PMUs) to ensure the data integrity of numerous PMU-based control strategies in the power industry. The outcomes will promote the stability and integrity of power systems as well as other tightly coupled, physically distributed CPSs.

Project description

Driven by the ever-increasing demand for accurate monitoring and advanced controls, modern power grids have become more and more dependent on a PMU-based monitoring system. Such a system consists of a large number of PMUs at multiple measurement locations which collect and transfer the measurement data in less than 30 milliseconds (synchronised by GPS) from all locations to data servers. Since most PMUs employ standard communication protocols - IEEE C37.118 without a cybersecurity mechanism, False Data Injection Attack (FDIA) on PMUs may be initiated which creates significant issues (e.g. power system instability and protection failure) and such an attack is difficult to detect as suggested by the National Institute of Standards and Technology.

Figure 1 Cyber-attack on PMUs and its impact on power grids

This research aims to address cybersecurity challenges and help protect the PMU from FDIA, allowing wide and safe adoption of numerous PMU-based monitoring and control applications.
Aim-1: Develop novel methods to completely and accurately extract spatio-temporal signatures from PMU measurements.
Aim-2: Build effective strategies for detecting sophisticated and large scale FDIA on PMUs by using advanced machine learning techniques.

Methodology
Task 1: Spatio-temporal signature extraction from PMUs. Proper signal processing techniques will be developed to validate raw PMU data quality. Then a novel feature extraction algorithm that combines multifractal characterisation and Time-Frequency (TF) mapping will be proposed to explore the most informative spatio-temporal signatures from PMU measurements.
Task 2: Light-weight deep learning-based FDIA detection. The extracted spatio-temporal signatures will be processed through advanced machine learning algorithms in order to recognise sophisticated and large scale FDIA on PMUs in real-time. Specifically, a light-weight multilayer perceptron will be proposed, which will perform as a deep neural network while being lightly parameterised like a shallow linear model. The new model’s sparse matrix operations also ensure high computational efficiency to support real-time FDIA detection.
Task 3: Validation and demonstration of the developed FDIA detection tool. The outcomes of Task 1 and Task 2 will be integrated into a ready-to-use tool for FDIA detection. Based on the PMU testing lab provided by NOJA Power, a thorough evaluation of the developed tool will be performed in terms of its real-world performance in detection accuracy, latency, and computational resources.

Innovation
(1) Accurate extraction of spatio-temporal signatures: The proposed feature extraction technique will fully reveal the multifractal characteristics of spatio-temporal signatures in PMU measurements and open up a new field for CPS cybersecurity research. It is the first time such multifractality has been discovered in PMU measurements. This groundbreaking approach is made possible because of the applicant’s many years of hands-on experience with PMUs and direct access to large amounts of PMU data.
(2) Identification of sophisticated and large-scale FDIA: Sophisticated FDIA (source ID swap) and large-scale FDIA (time tag mix) on PMUs are hard to detect by existing approaches. By completely and accurately extracting the spatio-temporal signatures, this project will explore the uniqueness of these signatures at each location and develop a novel light-weight deep learning method to identify FDIA in real-time.