Fellner, D. (2024). Data Driven Detection of Misconfigurations in Power Distribution Systems [Dissertation, Technische Universität Wien]. reposiTUm. https://doi.org/10.34726/hss.2024.120150
Power distribution system; artificial intelligence; deep learning; operational data; malfunction detection; data-driven monitoring; Machine Learning; transformer profile disaggregation; misconfigurations; smart grids
en
Abstract:
The modern power system is undergoing fundamental changes to adapt to new requirements, caused by the need of sustainable provision and consumption of energy. These changes especially affect power distributions systems, as they have been designed to host very predictable and similar household loads. Nowadays, they are increasingly required to connect decentral renewable generation which shows volatile power infeed characteristics as well as novel electrified loads such as electric vehicles or heating systems. The integration of these novel devices can create problems of voltage band violations or overloading to the grid and therefore pose problems to the safe and reliable operation of the power grid. To counter such difficulties, the already mentioned, new devices often implement grid-supporting control functionalities as reactive power control curves or charging current controls. However, due to the former sole purpose of passing on energy in a very foreseeable and one-directional manner, the distribution grid lacks sensory capacities. Therefore, grid operators have no way of ensuring that the necessary grid-supporting functionalities work as intended. New monitoring capabilities are needed.The thesis provides a layout of a data-driven monitoring approach which works under the mentioned circumstances found in distribution grids. Assessments of the available data and their properties such as sampling rates or restrictions due to privacy issues were conducted. Following, methods were developed to detect misconfigurations of grid-connected devices' control functionalities. These detected abnormalities are also categorized by classifications methods. These methods are adjusted to the data available: on the substation level these are traditional Machine Learning methods, at the device level Deep Learning methods are employed. Also, data mining methods are developed and assessed to gain information about the Low Voltage grid despite the lack of sensors. The methods are tested and validated using data collected in laboratory setup as well as through simulations, which are in turn also validated using the laboratory data. Also, the development framework used for developing and assessing the methods and creating the simulation data is described. The results present the best suited approaches for monitoring grid-supporting functionalities in electrical distribution grids. They show that these can be integrated with the current sensing and metering infrastructure and show a good detection and classification performance which enables the implementation of a meaningful decision-support tool for Distribution Grid Operators.
en
Additional information:
Arbeit an der Bibliothek noch nicht eingelangt - Daten nicht geprüft Abweichender Titel nach Übersetzung der Verfasserin/des Verfassers Kumulative Dissertation aus vier Artikel