IMF energy moments and LS-SVM based fault section location method for distribution network

Abstract

In China, the method of neutral resonant grounding and neutral ungrounded are widely used in distribution networks. When grounding fault occurred in distribution net work, the fault current is small, and vulnerable to the outside interferences, so the fault features are difficult to be detected. To solve this problem, a new method of fault section location based on intrinsic mode function (IMF) energy moments and least squares support vector machines (LS-SVM) was proposed. Firstly, the fault current signals were decomposed into several IMFs based on ensemble empirical mode decomposition (EEMD), then the fault feature vectors of IMF energy moments were obtained by integrating the IMF components with time. Secondly, the IMF energy moments with high correlation coefficients were taken as learning samples, then they were inputted to LS-SVM classifier to obtain fault selection location model. Finally, the unknown fault samples were inputted to the LS-SVM classifier trained before to achieve fault section location results. The simulation results show that this method can recognize features of fault signals accurately and identify the fault sections correctly.