Dc series arc fault detection based on random forest combined with entropy weight method

Abstract

In this paper, a fault detection method for DC series arc based on entropy weight method and random forest algorithm is proposed, which can be effectively applied to dc series arc fault identification for series resistive, capacitive and inductive loads. Firstly, the short-time Fourier transform (FFT) is used for frequency domain analysis of the data collected by accessing different loads. By comparing and analyzing the spectrum graphs under normal and fault conditions, the spectrum segment with the strongest frequency influence is selected for analysis. Time domain feature selection peak-to-peak value, mean value and standard deviation using entropy weight method to determine the weight, to form a comprehensive time domain feature, to avoid the instability of a single index; Spectrum standard deviation and mean value are selected for frequency domain features, and finally the time domain criterion and frequency domain criterion are taken as the input of random forest, and the random forest algorithm is used to achieve accurate detection of arc faults. The experimental results show that the proposed method can effectively distinguish the current characteristics of arc fault from those of normal operation, and the accuracy is higher than that of single criterion.