Classification of Stages of a High Impedance Fault Using Sequential Learning Algorithms

High Impedance Fault (HIF) detection is a demanding task for distribution power system operators and an ongoing research challenge. Recent developments in pattern recognition and data analytics has motivated researchers to develop detection algorithms based on Deep Learning (DL). Inherently, an HIF generates an arbitrary and non-linear signal. The varying nature of the fault is an ideal match for a DL sequential algorithm as the basis for a pattern classification technique. Here we present two hybrid DL models to classify the progression of an HIF observed in a 400V network. The hybrid models were developed with a Convolutional Neural Network (CNN) and two variants of the Recurrent Neural Network (RNN) algorithm. The models have been trained on real-world Giant Magneto-Resistive (GMR) Sensor data collected via a purpose-built 400-Volt test facility. The preliminary testing of models demonstrated 99.48% accuracy in classifying different states of the HIF. The robustness of the model on a large and more varied data-set is the subject of ongoing work.