AI-Driven Detection/Classification of High Impedance Faults in Low Voltage Distribution Networks Using Sliding DFT and Deep Neural Networks

Abstract

High impedance fault (Hi-ZF) detection/classification in low voltage distribution networks remains a significant challenge due to the low fault current levels, which are often indistinguishable from normal load conditions using conventional overcurrent protection devices. Hi-ZF, typically caused by broken conductors contacting high-impedance surfaces such as soil or vegetation, pose serious public safety hazards, risk of fire ignition, equipment damage, and legal liability. This paper proposes a novel AI-driven detection/classification framework combining Sliding Discrete Fourier Transform (SDFT) and Deep Neural Networks (DNN) to accurately detect/classify both low impedance faults (LIFs) and Hi-ZFs using single-ended current measurements. The method first applies SDFT to extract frequency-domain features from local current signals. These features are then fed into a DNN classifier trained to distinguish between Hi-ZFs, LIFs, and non-fault transient events such as load/capacitor switching. The proposed scheme was validated using extensive simulations on the unbalanced IEEE 13-Bus distribution test system, employing ATP/EMTP and MATLAB/Simulink platforms. Results demonstrate that the scheme reliably detects Hi-ZFs and LIFs within 26.5 ms, achieving a classification accuracy of 99.1%, and the proposed AI-based methodology shows strong potential for enhancing protection and situational awareness in modern smart distribution grids.

Article Content

Keywords

Article Info