A Scoping Review of Machine Learning Applications in Power System Protection and Disturbance Management

TL;DR

This scoping review analyzes the state of machine learning applications in power system protection and disturbance management, focusing on fault detection, fault classification, and fault localization. It synthesizes 119 studies from 603 records, revealing widespread reliance on simulated data, methodological fragmentation, and limited real-world validation. To address these issues, the authors propose an ML-oriented taxonomy, terminologies, and dataset/reporting guidelines aimed at improving reproducibility and comparability, and they outline concrete directions for benchmark datasets, realistic validation (including hardware-in-the-loop and digital twins), and advanced architectures. The work highlights the practical significance of moving ML-based protection from theoretical promise toward field-ready solutions capable of handling increasingly dynamic grids with high DER penetration and decentralized operation.

Abstract

The integration of renewable and distributed energy resources reshapes modern power systems, challenging conventional protection schemes. This scoping review synthesizes recent literature on machine learning (ML) applications in power system protection and disturbance management, following the PRISMA for Scoping Reviews framework. Based on over 100 publications, three key objectives are addressed: (i) assessing the scope of ML research in protection tasks; (ii) evaluating ML performance across diverse operational scenarios; and (iii) identifying methods suitable for evolving grid conditions. ML models often demonstrate high accuracy on simulated datasets; however, their performance under real-world conditions remains insufficiently validated. The existing literature is fragmented, with inconsistencies in methodological rigor, dataset quality, and evaluation metrics. This lack of standardization hampers the comparability of results and limits the generalizability of findings. To address these challenges, this review introduces a ML-oriented taxonomy for protection tasks, resolves key terminological inconsistencies, and advocates for standardized reporting practices. It further provides guidelines for comprehensive dataset documentation, methodological transparency, and consistent evaluation protocols, aiming to improve reproducibility and enhance the practical relevance of research outcomes. Critical gaps remain, including the scarcity of real-world validation, insufficient robustness testing, and limited consideration of deployment feasibility. Future research should prioritize public benchmark datasets, realistic validation methods, and advanced ML architectures. These steps are essential to move ML-based protection from theoretical promise to practical deployment in increasingly dynamic and decentralized power systems.

Figures (6)

• Figure 1: This overview figure illustrates the analytical flow of the review. The study begins with the emerging challenges of modern power grids, which motivate the need for intelligent protection systems. These challenges are mapped onto core protection tasks, which are examined through the lens of machine learning - focusing on common approaches, architectures, and deployment considerations. The review concludes with structured contributions that synthesize insights, promote reproducibility, and guide future research.
• Figure 2: Conceptual taxonomy of machine learning paradigms in power system protection and disturbance management. Supervised learning learns from labeled data, unsupervised learning learns from unlabeled data, and reinforcement learning learns through interaction with an environment. Deep learning provides a cross-cutting family that learns hierarchical representations directly from raw data and enhances all three paradigms.
• Figure 3: Classical ml pipeline applied to power system protection and disturbance management. The first stage, Data Collection, entails acquiring voltage and current measurements from transmission infrastructure and system-level monitoring devices (e.g., protective relays, PMUs, and SCADA systems). The Feature Extraction stage focuses on transforming raw signal data into informative representations, including time-domain amplitude features, frequency spectra (e.g., via Fourier analysis), and multi-resolution decompositions (e.g., wavelet transforms), to emphasize fault-related characteristics. In the ml Models stage, the extracted features are employed to train and evaluate classifiers such as ann and svm, enabling accurate discrimination between normal and faulty system states. Finally, the Outputs stage illustrates the model's decisions, such as identifying specific fault types (e.g., Fault A) versus no-fault conditions, and initiating system-level actions including trip signals, alarms, or holding operations. This end-to-end pipeline is designed to enhance protection reliability, improve situational awareness, and support faster, more adaptive disturbance management in modern, decentralized power systems. Figure recreated and adapted after yang_machine_2021.
• Figure 4: Trend in annual publications related to machine learning-based power system protection, based on results from Google Scholar using the same keyword combinations defined in the search strategy (see Table \ref{['tab:database_queries']}). The plot illustrates a substantial increase in publication activity, particularly from 2018 onward, highlighting the growing research interest in this interdisciplinary field. The value for 2025 is extrapolated based on partial-year data.
• Figure 5: PRISMA-ScR flow diagram illustrating the structured study selection process used in this review. A total of 603 records were retrieved from Scopus and IEEE Xplore and imported into Rayyan for screening. After removing 55 duplicates, 548 unique studies were screened by title and abstract, resulting in the exclusion of 375 records. Full-text assessment led to the inclusion of 97 articles, with an additional 22 studies identified through manual search, totaling 119 included studies. The arrows in the diagram trace the sequential progression through identification, deduplication, screening, and eligibility assessment, clearly indicating how many records were retained or excluded at each step to ensure transparency and reproducibility.