Clark, JohnGope, ProsantaAlromih, Arwa2023-10-092023-10-092023-09-25https://hdl.handle.net/20.500.14154/69349Energy plays an essential role in our lives. Merging the existing electricity networks with distributed energy resources and information and communications technology (ICT) changes how companies and customers generate, distribute, and consume energy. This integration transforms the legacy electricity networks into smart systems, or what is currently known as the Smart Grid (SG). Smart grid infrastructure has been one of the major industrial revolutions that has attracted widespread adoption across the globe. Therefore, they can be the target of major security risks as they are not inherently secure. In this sector, sensors’ and meters’ data are the main factors in any decision-making process. This introduces the need to develop appropriate security mechanisms that ensure data integrity. One of the main attacks against data integrity in energy networks is energy theft. This attack can be made by injecting false consumption data into the network. The consequences of a successful energy theft attack on smart grid systems can be severe and far-reaching as it can result in power outages and physical damage to equipment which can be a safety hazard to individuals. Therefore, secure techniques are needed to detect any anomalies or injection attempts and smart meter data integrity should be considered and ensured. In this thesis, we propose three machine learning (ML) based energy theft detectors that address the existing challenges facing current research in this domain. In particular, we consider the impact proposed by prosumers in launching new types of energy thefts and how to detect them. We also show how to fully utilise data from multiple sources for better detection performance. To decrease the probability of any privacy breaches caused by the use of customers’ data, privacy-preserving approaches are proposed. Lastly, we tackle the significant impact on demand management caused by energy thefts by proposing a combined energy theft detector with demand management. The findings presented in this thesis show that our approaches can accurately detect energy thefts, with minimal information leakage. Moreover, the results are also promising in providing a clear link between reliably managing demand when energy theft is considered.136enEnergy theftssplit learningsmart gridprivacy preservingprivacyThesis