Date of Award

January 2023

Document Type


Degree Name

Doctor of Philosophy (PhD)


Electrical Engineering

First Advisor

Naima N. Kaabouch


The traditional power grid is no longer a practical solution for power delivery due to several shortcomings, including chronic blackouts, energy storage issues, high cost of assets, and high carbon emissions. Therefore, there is a serious need for better, cheaper, and cleaner power grid technology that addresses the limitations of traditional power grids. A smart grid is a holistic solution to these issues that consists of a variety of operations and energy measures. This technology can deliver energy to end-users through a two-way flow of communication. It is expected to generate reliable, efficient, and clean power by integrating multiple technologies. It promises reliability, improved functionality, and economical means of power transmission and distribution. This technology also decreases greenhouse emissions by transferring clean, affordable, and efficient energy to users. Smart grid provides several benefits, such as increasing grid resilience, self-healing, and improving system performance. Despite these benefits, this network has been the target of a number of cyber-attacks that violate the availability, integrity, confidentiality, and accountability of the network. For instance, in 2021, a cyber-attack targeted a U.S. power system that shut down the power grid, leaving approximately 100,000 people without power. Another threat on U.S. Smart Grids happened in March 2018 which targeted multiple nuclear power plants and water equipment. These instances represent the obvious reasons why a high level of security approaches is needed in Smart Grids to detect and mitigate sophisticated cyber-attacks. For this purpose, the US National Electric Sector Cybersecurity Organization and the Department of Energy have joined their efforts with other federal agencies, including the Cybersecurity for Energy Delivery Systems and the Federal Energy Regulatory Commission, to investigate the security risks of smart grid networks. Their investigation shows that smart grid requires reliable solutions to defend and prevent cyber-attacks and vulnerability issues. This investigation also shows that with the emerging technologies, including 5G and 6G, smart grid may become more vulnerable to multistage cyber-attacks. A number of studies have been done to identify, detect, and investigate the vulnerabilities of smart grid networks. However, the existing techniques have fundamental limitations, such as low detection rates, high rates of false positives, high rates of misdetection, data poisoning, data quality and processing, lack of scalability, and issues regarding handling huge volumes of data. Therefore, these techniques cannot ensure safe, efficient, and dependable communication for smart grid networks. Therefore, the goal of this dissertation is to investigate the efficiency of machine learning in detecting cyber-attacks on smart grids. The proposed methods are based on supervised, unsupervised machine and deep learning, reinforcement learning, and online learning models. These models have to be trained, tested, and validated, using a reliable dataset. In this dissertation, CICDDoS 2019 was used to train, test, and validate the efficiency of the proposed models. The results show that, for supervised machine learning models, the ensemble models outperform other traditional models. Among the deep learning models, densely neural network family provides satisfactory results for detecting and classifying intrusions on smart grid. Among unsupervised models, variational auto-encoder, provides the highest performance compared to the other unsupervised models. In reinforcement learning, the proposed Capsule Q-learning provides higher detection and lower misdetection rates, compared to the other model in literature. In online learning, the Online Sequential Euclidean Distance Routing Capsule Network model provides significantly better results in detecting intrusion attacks on smart grid, compared to the other deep online models.