Online Detection of Events With Low-Quality Synchrophasor Measurements Based on $i$Forest

Abstract

In this paper, we propose an online data-driven approach that leverages the isolation mechanism for fast event detection with low-quality data measurement. The proposed adaptive and online isolation forest (iForest)-based detection (AOIFD) method adopts a hierarchical subspace feature selection scheme to design two levels of detectors. As such, it is capable of differentiating events from low-quality data measurements, preventing false alarms in the presence of low-quality data measurements. We further propose a data augmentation method to address the training data imbalance, which is caused by the rare occurrence of events. Moreover, we propose an adaptive training process to update the AOIFD method so that it can adapt to the time-varying operating conditions of power systems. The proposed AOIFD algorithm is practical in the sense that it is a fast-response method that requires no system modeling information and no global communications. Case studies with both synthetic and realistic PMU data are conducted to validate the effectiveness of the proposed method.