High-Impedance Fault Detection in Electric Power Distribution Network Using Local Pattern Transformation Methods and Bidirectional Long Short–Term Memory

Abstract

The high-impedance fault (HIF) occurring in medium voltage (MV) distribution networks is dangerous to livestock and personnel due to its arcing nature. The untimely detection of the fault can endanger lives and destroy equipment. Identifying the occurrence of HIF in a power system is a cumbersome task, as fault current falls within the normal current range. The paper analyses current signals from radial and mesh distribution networks and features extracted during HIF and non-HIF conditions by using the local binary pattern (LBP), local neighbor gradient pattern (LNGP), local neighbor descriptive pattern (LNDP), and local gradient pattern (LGP). In the proposed algorithm, 1D signal analysis for HIF detection in the MV distribution system is performed for the first time for fault analysis. The Kruskal–Wallis test was carried out to get the best feature sets from the extracted features. HIF and non-HIF were classified by bidirectional long short–term memory (Bi-LSTM) for the selected feature sets. Among the four algorithms, LGP attains the best accuracy for both networks; hence, the paper recommends that LGP with Bi-LSTM is more effective for detecting HIF occurrence.