High Impedance Fault Detection in Distribution Networks using Unsupervised Learning Technique

Abstract

An algorithm is proposed in this article that detects High Impedance Faults (HIFs) in distribution networks using a novel approach. It is often difficult to detect HIFs in distribution grids because of the small magnitude of fault current and its non-linear variations. The purpose of this study is to investigate the possibility of automatically learning latent features from raw, unlabeled voltage and current data with the help of unsupervised feature learning. In order to identify transient behavior associated with HIFs and other typical faults, signal features are extracted using an autoencoder. A quadratic discriminant analysis (QDA) is then used to detect the presence of fault in the system. A fault detection accuracy of 98% is achieved by the proposed methodology. A modified IEEE 33 Bus benchmark system is used to validate the developed technique.