Electronic Theses and Dissertations

Date

2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy

Department

Electrical & Computer Engineering

Committee Chair

Mohd. Hasan Ali

Committee Member

Alexander Headley

Committee Member

Mohammadreza Davoodi

Committee Member

Sanjay Mishra

Abstract

One major challenge in implementing the solar photovoltaic (PV) systems is integrating the generated DC power into AC systems. Efficient operation of these systems requires smart inverters that can communicate with Supervisory Control and Data Acquisition (SCADA) systems. Internet of Things (IoT) devices communicate various parameters and control signals with the SCADA system by using advanced communication technologies like 5G. However, communication lines are susceptible to cyber intrusions. Also, various set points of controllers of smart inverters can be compromised. Given these vulnerabilities, it is crucial to detect cyber intrusions effectively to ensure the reliable operation of grid connected PV systems. Conventional detection techniques use neural network-based approaches that cannot classify attacks effectively. Also, while numerous researchers have suggested Artificial Intelligence (AI)-based detection strategies, the identification of the attack's location has not been addressed in these approaches. Moreover, effective mitigation strategies for cyber-intrusions have been relatively underexplored in existing literature. To overcome these drawbacks, this dissertation develops a cyber-secured smart intelligent inverter for 5G-enabled grid-connected PV system by providing robust solutions for the detection, identification and mitigation of cyber-attacks. A novel Long Short-Term Memory (LSTM)-Extreme Gradient Boosting (Xgboost) method is proposed for the identification of Denial of Service (DoS) attack on the smart solar PV inverter system. Also, a novel Convolution extreme gradient boosting (ConvXgboost) method is proposed for not only detecting DoS and False Data Injection (FDI) attacks but also identifying the location and component of the system that was compromised. Moreover, three novel strategies such as an equation-based method, Reinforcement learning (RL) agent-based method, and a digital twin approach-based method are proposed. Simulation results demonstrated the effectiveness of the proposed LSTM-Xgboost method when compared to the conventional LSTM model. Also, the ConvXgboost method is effective for both the smart PV and PV fuel cell (PV-FC) systems when compared with the existing Convolution Neural Network (CNN) and decision tree (DT) strategies. Moreover, the proposed mitigation strategies are found to be effective, and performance comparisons indicate that the equation-based mitigation is the most efficient. Additionally, experimental testbed is developed to validate the equation-based mitigation strategy for the smart PV inverter.

Comments

Data is provided by the student.

Library Comment

PDF

Notes

Open access.

Download

Share

COinS