Low voltage multiloop series arc fault detection based on deep recurrent neural network

Abstract

In the low-voltage residential power distribution system, due to the diversification and variability of the load series and parallel connection forms of the load end in the actual distribution network, it is of great significance to detect the branch fault through the waveform change of the trunk line. Based on the periodicity and timing characteristics of the current signal, in this paper, a method of multi branch series arc fault detection based on deep recurrent neural network is proposed. The experimental platform is built to collect the trunk current signals under different series faults of branches. The experimental results show that the final detection result reaches 95.67%, which confirms the use of depth. The feasibility of using recurrent neural network to identify multi branch series arc fault in low voltage system is discussed, and provides a new idea and beneficial exploration for the accurate detection of low-voltage series arc fault.