A fast automatic detection and classification of voltage magnitude anomalies in distribution network systems using PMU data

• Published in: Measurement : journal of the International Measurement Confederation Vol. 192; p. 110816
• Main Authors: Fuentes-Velazquez, Jose, Beltran, Ernesto, Barocio, Emilio, Angeles-Camacho, Cesar
• Format: Journal Article
• Language: English
• Published: London Elsevier Ltd 31.03.2022; Elsevier Science Ltd
• Subjects: Algorithms; Anomalies; Directed acyclic graph; Electric potential; Electricity distribution; Feature extraction; Monte Carlo simulation; Neural networks; PMU data; Recurrence quantification analysis; Signal classification; Supervised machine learning; Support vector machines; Voltage; Voltage magnitude anomaly; Directed acyclic graph; PMU data; Voltage magnitude anomaly; Supervised machine learning; Recurrence quantification analysis; Monte Carlo simulation
• ISSN: 0263-2241; 1873-412X
• DOI: 10.1016/j.measurement.2022.110816

Timely anomaly detection and classification in voltage signals for distribution systems allows the design of preventive and corrective actions to avoid damage or loss of equipment. In this paper, an approach that combines a robust recurrence quantification analysis (RRQA) for features’ extraction that allows anomaly detection and classification through a multiclass support vector machine (SVM) algorithm is proposed. This approach is robust to noise, is free a filtering stage, has a low computational burden and is easy to interpret, becoming viable for online monitoring of distribution systems. For its validation, case studies with the presence of voltage magnitude anomalies (VMA) events that occur during normal operation conditions are analyzed. Thus, synthetic records generated through a Monte Carlo model are compared with other algorithms based on similar strategies. Finally, the proposed approach is assessed using PMU records installed in a distribution system to show its performance in a real world environment. •Low dimensional feature space provides a fast and accurate training stage.•Local detection through a sliding window and a subsequent classification.•An algorithm based on RRQA and SVM to identify Power Quality events from PMU signals.