Abdullahi Oboh Muhammed, “Artificial Intelligence (AI) Applications in Power Systems with Renewable Energy Integration”, PhD Dissertation, PhD in Engineering, Electrical Engineering Department, Khalifa University, United Arab Emirates, April 2025.As the power sector evolves with the proliferation of renewable energy sources (RES), the increasing demand for electric vehicles (EVs), and the integration of communication links in smart grids, modern power systems face escalating challenges to maintain stability and security. Unlike traditional power generators, inverter-based RES, such as solar photovoltaic (PV) and wind energy—lack inherent inertia, creating new vulnerabilities in power grid frequency stability and reducing resilience to disturbances. Furthermore, the increase in EVs charging complicates load forecasting and demand response, as charging patterns introduce variability at both localized and grid-wide levels. In addition, integrating sophisticated data links increases the risk of malicious attacks on power systems. Given these shifts and the interconnected nature of stability and security, real-time visibility into system behavior is essential. However, the influx of data from various grid components demands more than conventional monitoring techniques. Efficient distillation and interpretation of big data are necessary to capture complex high-speed dynamics and support timely decision-making. This research develops and applies advanced artificial intelligence (AI) techniques for real-time prediction of small-signal stability, time-varying inertia estimation, and detection and mitigation of cyber threats to address stability and cyber security challenges to ensure operational efficiency and service reliability. First, in real-time stability monitoring, this thesis proposes deep learning models, including long short-term memory (LSTM) networks, end-to-end convolutional neural networks with LSTM(CNN-LSTM), and convolutional-LSTM(Conv-LSTM) frameworks for forecasting small-signal stability by capturing spatial and temporal features inherent in the oscillatory parameters of power systems. Trained on IEEE benchmark system datasets, these models can allow operators to monitor oscillatory modes and damping rates and take prompt action during online operations. Another contribution of this research is a dual-stage, hybrid deep learning model designed to forecast dynamic inertia, a critical factor for maintaining grid stability amid time-varying renewable inputs. This approach employs an ensemble of LSTM with multilayer perceptron (LSTM-MLP) networks, Conv-LSTM models, and temporal convolutional networks with LSTM (TCN-LSTM) architectures, providing computationally efficient and robust inertia predictions in real-time, even under noisy and dynamic conditions, while outperforming existing models. Validated across IEEE test systems, these results achieve accurate predictions and ensure model resilience to variable input quality, supporting sustainable operations in volatile grid conditions. Beyond stability, this research addresses cyber-physical security challenges in power systems. In response to increasing cyber vulnerabilities introduced by digitalization, this thesis proposes novel AI-based cybersecurity frameworks for detecting, locating, and mitigating cyberattacks, such as false data injection and denial-of-service attacks, particularly in automatic generation control (AGC) systems. Utilizing advanced detection models with exemplary F1 and AUC scores, these frameworks effectively identify individual and coordinated cyber-attacks. Additionally, the mitigation models effectively restore compromised measurements, enabling AGC systems to operate reliably even under attack. Moreover, this study demonstrates the model’s ability to differentiate between typical PV power injections and cyberattack signatures, contributing to more resilient system defences.
| Date of Award | 8 May 2025 |
|---|
| Original language | American English |
|---|
| Supervisor | Mohamed El Moursi (Supervisor) |
|---|