Wavelet-based extreme learning machine for distribution grid fault location

Precise knowledge of faults is very exigent to reduce the outage duration as most of the customer minute losses in distribution grids occur due to longer period of interruptions caused by faults. This study proposes a fault location technique combining advanced signal processing and machine learning tools for distribution grids. The proposed technique decomposes three-phase currents measured from sending end employing wavelet transform (WT) and collects useful features to fetch them as inputs of extreme learning machine (ELM). Satisfactory values of the selected statistical performance measures validate the efficacy of proposed fault location technique. Besides, the efficacy of support vector regression (SVR) and artificial neural network (ANN) are also tested employing the WT extracted features. The presented results show the superiority of ELM-WT technique over SVR-WT and ANN-WT techniques in terms of the selected performance measures and training times. Additionally, the proposed technique is independent of fault resistance, inception angle, the presence of measurement noise, thermal expansion/contraction of the distribution line and pre-fault loading condition. Furthermore, the hybrid method detects and classifies different types of faults before locating them with different machine learning tools.