AI-based measurement architectures in electrical power systems

This doctoral dissertation investigates the integration of Artificial Intelligence (AI) with measurement systems to improve the performance, reliability, and robustness of modern electrical power systems. The increasing penetration of Renewable Energy Sources (RES), distributed generation, and smart grid technologies poses significant challenges in monitoring, forecasting, and fault management. To address these challenges, the research develops and investigates AI-based measurement architectures based on Machine Learning (ML) techniques, providing scalable, accurate, and deployable solutions. The work is structured around three main technical areas. The first addresses AI-based forecasting of electrical quantities within Distributed Measurement Systems (DMSs) deployed in smart buildings and hybrid power systems, improving prediction accuracy for monitoring purposes and enabling real-time energy management architectures and Digital Twins (DTs) in DMS-based environments. The second area focuses on fault detection and predictive maintenance in Medium Voltage (MV) distribution networks. ML-based diagnostic methods are developed for Cable Joints (CJs) and MV switchgears using both supervised and unsupervised approaches. A novel Health Index (HI) is introduced to quantify CJ aging and is estimated through Deep Learning (DL) models, supporting condition-based maintenance strategies. The third area integrates metrological principles into AI-based architectures. The impact of measurement uncertainty on AI model performance is systematically analyzed, leading to uncertainty-aware evaluation frameworks and uncertainty-driven Data Augmentation (DA) strategies to improve robustness and generalization. In parallel, AI is applied to metrological estimation: Artificial Neural Networks (ANNs) are used to estimate temperature-induced accuracy deviations in Instrument Transformers (ITs), while physics-informed models are developed for Rogowski Coils (RCs) to compensate thermal effects. These solutions are also implemented on microcontrollers, enabling local and real-time modeling of sensor behavior. Overall, this dissertation provides an interdisciplinary contribution by integrating advanced ML with domain-specific knowledge in power systems and metrology, supporting the development of future smart grids that are intelligent, reliable, and metrologically sound.