Machine learning-based state estimation in electrical power and multi-energy systems: A comprehensive review

This paper presents a comprehensive review and analysis of the application of machine learning (ML) methodologies to state estimation (SE) in power systems (PSs) and multi-energy systems (MESs). Traditional PS-SE, heavily reliant on weighted least squares and Kalman filter techniques, has experienced challenges, such as difficulties to track the dynamics of the inverter-based resources (IBR) in transmission networks (TNs), limited observability and topology variations in distribution networks (DNs), or susceptibility to various types of data anomaly in both TNs and DNs. These challenges are being amplified in MES-SE; for instance, energy systems exhibit different dynamics and sampling rates, some of them may be unobservable, and the impact of bad data can propagate across energy systems. The emergence of ML brings notable potential for overcoming these challenges; for example, “good” predictions are derived from large sets of historic data, data classification and anomaly detection are automatized and made more reliable, complex relationships within models can be established, and algorithms can learn from generated results. This paper presents the fundamentals and challenges of PS-SE and MES-SE, provides an overview of relevant ML methods, and highlights the advantages of applying these methods in both PS-SE and MES-SE. The study concludes with a real-life example and outlines promising research directions for the application of ML in enhancing PS-SE and MES-SE results.