A novel MODWT–local pattern transformation feature fusion approach for high-impedance fault detection in medium voltage power distribution networks

Due to the arcing nature, the high-impedance fault (HIF), which typically occurs in medium voltage (MV) distribution networks, poses a risk to equipment, personnel, and livestock. Early fault detection can save lives and prevent equipment destruction. Since the fault current in a power system is contained within the normal current range, identifying the occurrence of HIF is difficult. Using maximal overlap discrete wavelet transform (MODWT) and a combination of MODWT–local pattern transformations (local binary pattern, local gradient pattern, local neighbor gradient pattern, and local neighbor descriptive pattern), the paper analyses the current signals from radial and mesh distribution networks and features extracted during HIF and non-HIF (line-to-ground (LG), double line (LL), double line-to-ground (LLG), and triple line-to-ground (LLLG) fault conditions). For the first time in power distribution networks for fault analysis, the suggested algorithm performs the HIF detection in the MV distribution system. To determine the best feature sets from the extracted features, the Kruskal–Wallis test was performed. Bidirectional long short-term memory (Bi-LSTM) was used for the chosen feature sets to classify the data as HIF or non-HIF. MODWT–LGP fusion achieves the highest accuracy for both networks among the four algorithms.