Automatic Power System Event Classification Using Quadratic Discriminant Analysis on PMU Data

Rapid detection and diagnosis of events in power system wide area monitoring is of great interest to system operators, with event classification being a major aspect of diagnosing an event. Other event diagnostic aspects include the time the event occurred, location of the event, root cause of the event and magnitude of the event. Automatic event classification enhances the operators' ability to identify the types of events occurring in a system quickly, which helps to assist fast decision making when restoring power to the system. This paper proposes an approach for classifying power system events, namely Generation Dip, Loss of Load and Line Trip Events, by employing Quadratic Discriminant Analysis (QDA) on Phasor Measurement Unit (PMU) data in combination with a forward selection technique. QDA is a commonly used supervised, statistical technique for data classification, and works by finding a combination of features that separates the data into different classes by modelling the difference between them. Historical power system event data is used to construct an event database, and as new events are detected the methodology automatically classifies the event based on the effect on power system variables. The reliability of the proposed method is demonstrated using simulated case studies, constructed using DigSilent Power Factory, and real data case studies, acquired from the UK and Irish Power System.