Waveform Measurement Unit-Based Fault Location in Distribution Feeders via Short-Time Matrix Pencil Method and Graph Neural Network

Abstract

This article proposes the use of the Short-Time Matrix Pencil method (STMPM) and Graph Neural Network (GNN) for fault location in active distribution feeders based on an emerging class of sensors, known as Waveform Measurement Units (WMUs). WMUs record synchronized voltage and current waveforms in the time domain with high sampling rates. The proposed fault location framework consists of two stages. In the first stage, STMPM is adopted to capture the dominant modes of the transient changes of WMUs' sinusoidal signals due to faults in different locations of the distribution grid. The second stage is to use a grid-informed GNN model to identify the fault location and type using the captured features of the signal before, during, and post-fault with STMPM. GNN can capture the spatial-temporal relationship between data from different sensors in different locations to enhance situational awareness and fault location accuracy. The proposed method is examined on a modified IEEE network with distributed energy resource (DER) and for transient symmetrical and asymmetrical faults under different loading, DER generation level, noises, and sensors' sampling rate conditions. The results show the merits of the proposed two-stage fault location framework compared to the conventional approaches; while a challenging problem is addressed in active distribution grids.