Real-Time Synchrophasor Data Anomaly Detection and Classification Using Isolation Forest, KMeans, and LoOP

Abstract

Power grid operators assess situational awareness using time-tagged measurements from phasor measurement units (PMUs) placed at multiple locations in a network. However, synchrophasor measurements are prone to anomalies which may impact the performance of phasor based applications. Anomalies include any deviation from expected measurements resulting from power system events or bad data. Bad data include data errors or loss of information due to failures in supporting synchrophasor cyber infrastructure. It is necessary to flag bad data before utilizing for an application. This work proposes a tool for the detection and classification of anomalous data using an unsupervised stacked ensemble learning algorithm. The proposed synchrophasor anomaly detection and classification (SyADC) tool analyzes a selected window of data points using a combination of three unsupervised methods, namely: isolation forest, KMeans and LoOP. The method classifies the data as anomalies or normal data with more than 99% recall. The method also provides a probability of the data to be an event or bad data with more than 99% recall. Results for the IEEE 14 and 68 bus systems with synchrophasor data obtained using Real-Time Digital Simulator and data of industrial PMUs highlight the superiority of the algorithm to detect and classify anomalies.