Concordia University

https://www.concordia.ca/content/shared/en/news/encs/info-systems-eng/defences/2019/08/27/ensemble-feature-learning-based-event-classification-cyber-physical-security-smart-grid.html

notice

Master Thesis Defense - August 27, 2019: Ensemble Feature Learning-Based Event Classification for Cyber-Physical Security of the Smart Grid

August 21, 2019

 

Chengming Hu

Tuesday, August 27, 2019 at 1:30 p.m.
Room EV003.309

You are invited to attend the following M.A.Sc. (Quality Systems Engineering) thesis examination.

Examining Committee

Dr. F. Mafakheri, Chair
Dr. C. Wang, Supervisor
Dr. J. Yan, Supervisor
Dr. A. Youssef, CIISE Examiner
Dr. C. Lai, External Examiner (ECE)

 

Abstract

The power grids are transforming into the cyber-physical smart grid with increasing two-way communications and abundant data flows. Despite the efficiency and reliability promised by this transformation, the growing threats and incidences of cyber- attacks targeting the physical power systems have exposed severe vulnerabilities. To tackle such vulnerabilities, intrusion detection systems (IDS) are proposed to monitor threats for the cyber-physical security of electrical power and energy systems in the smart grid with increasing machine-to-machine communication. However, the multi-sourced, voluminous, correlated, and often noise-contained data, which record various concurring cyber and physical events, are posing significant challenges to the accurate distinction by IDS among events of inadvertent and malignant natures.

Hence, in this research, an ensemble learning-based feature learning and classification for cyber-physical smart grid are designed and implemented. The contribution of this research are (i) the implementation and evaluation of an ensemble learning-based attack classifier using extreme gradient boosting (XGBoost) to effectively detect and identify attack threats from the heterogeneous cyber-physical information in the smart grid; (ii) the implementation and evaluation of stacked denoising autoencoder (SDAE) to extract highly-representative feature space that allow reconstruction of a noise-free input from noise-corrupted perturbations; (iii) the design and implementation of a novel ensemble learning-based feature extractors that combine multiple autoencoder (AE) feature extractors and random forest base classifiers, so as to enable accurate reconstruction of each feature and reliable classification against malicious events. The simulation results validate the usefulness of ensemble learning approach in detecting malicious events in the cyber-physical smart grid.




Back to top

© Concordia University