This PhD project deals with the Modelling of Cascading Events in Power Systems and their Online Identification with Machine Learning, considering the integration of Renewable Energy Sources. Cascading events involve highly complex dynamic phenomena and in some cases can pose significant challenges to the stability and reliability of power grids, leading even to blackouts. The intermittent nature of renewable generation introduces additional complexities, as the system dynamic behavior following a contingency becomes more unpredictable. Consequently, there is an increasing need for cascading event identification methods that can effectively handle these emerging challenges and ensure secure network operation. Machine Learning methods can extract complex relationships from power system data, by capturing the underlying dynamics, offering a promising tool for the accurate and timely identification of the online system state. In addition, due to the extensive installation of Phasor Measurement Units in modern power systems, it is possible to acquire measurement data related to electrical system variables in close-to-real time. The thesis first delves into the understanding of cascading events appearance, as defined by the discrete action of protection devices, using detailed dynamic simulations and considering uncertainties associated with network operating conditions, contingencies and renewable generation. To address the online nature of the problem, supervised machine learning methods that utilize measurement data are developed. Different contemporary machine learning approaches are investigated, to identify the most suitable techniques for the detection of the appearance of cascading events, formulated as a binary classification problem, and the prediction of the reason of the upcoming cascading event, formulated as a multi-class classification problem. Furthermore, this thesis explores the challenges associated with the application of machine learning models on power system data, such as the online inference time, class imbalance, practical considerations related to measurement data and investigates techniques for model explainability to enhance the trustworthiness of the developed models. The contributions of this thesis lie in the development of machine learning-based techniques for online identification of cascading events in power systems, enabling more proactive and efficient situational awareness. These insights have the potential to significantly enhance the resilience and stability of power grids, minimizing the risk oflarge-scale blackouts and improving the overall reliability of the system.Georgios Nakas is sponsored through Engineering & Physical Sciences Research Council (EPSRC) Research Excellence Award (REA) and is supervised by Dr. Panagiotis Papadopoulos and Professor Graeme Burt.
Date of Award | 27 Mar 2024 |
---|
Original language | English |
---|
Awarding Institution | - University Of Strathclyde
|
---|
Sponsors | University of Strathclyde |
---|
Supervisor | Panagiotis Papadopoulos (Supervisor) & Graeme Burt (Supervisor) |
---|