Leveraging Optimized Machine Learning for Anomaly Detection and Quality of Service Enhancement of PMUs Observation in Smart Grids
Abstract
With the fast infrastructure expansion of the electricity network, phasor measurement units (PMUs) have become essential in smart grids (SGs), offering real-time electrical measurements, e.g., voltage, current, and frequency. Using GPS, PMUs can enhance location and time synchronization, enable quicker response to disturbances, accurately detect faults, and support renewable energy sources integration. In this regard, PMUs’ measurement classification significantly improves SGs’ controllability, sustainability, and QoS. This paper introduces a machine learning-based classifier using extreme gradient boosting not only to detect realistic PMU’s observation anomaly but also to categorize the data buses and observations within an IEEE Bus system for QoS enhancement in SGs. To improve different data type classifications (normal operation, faults, line outages, generation changes, and load fluctuations) and mitigate the imbalance in the utilized classes, the paper utilizes four feature scaling techniques (standard, min-max, max-abs, and robust scaling) along with hyperparameter optimization. For anomaly detection, the model is a binary classifier. The model is a multi-class classifier for data type and bus number classification. In both cases, the model is examined with and without the scaling methods. To ensure the model’s robustness, we evaluated its performance with multiple metrics. The results show that the proposed classifier using min-max and robust scaling, achieved the best accuracy, ranging from 99.9% to 99.99%, outperforming benchmarks.
Keywords
How to cite
Abdalzaher, M. S., Shaaban, M. F., & Aburukba, R. (2025). Leveraging Optimized Machine Learning for Anomaly Detection and Quality of Service Enhancement of PMUs Observation in Smart Grids. IEEE Access, 13, 196959–196972. https://doi.org/10.1109/access.2025.3633928