Arc fault detection and identification via non-intrusive current disaggregation

Abstract

The series arc faults causing increasing incidences of household fires are difficult to be detected due to the low fault current amplitude. The existing research on arc fault detection is at the power socket level with simple appliances circumstances, but hardly on the whole residential household. As series arc fault current waveform holds unique information of downstream appliance in operation, this paper proposes an arc fault detection and identification method via supervised non-intrusive current disaggregation for low-voltage consumers in the service entrance. In the proposed method, a NILM module initially identifies the appliances in operation in the residential household while an arc fault detection module screens the simultaneous total current signal for an arc fault. In case of arc fault detected, an arc current disaggregation module will find the best-fit arc fault waveform of one or more specific appliances by total current harmonic disaggregation. Then combining the prior knowledge of household power line topology, the consumer can be alerted with the arc resided branch indicated by the operation appliances. By collecting current signals of appliances with and without arc fault, a set of features is extracted from the measurements for total current feature threshold selection and appliance-specific template acquisition. Besides arc fault detection at the service wire entrance, it also infers the specific circuit branch on which arc faults occur. This approach contributes to timely location and elimination of fire hazards. Experiments in consumer scenarios are carried out with an arc fault generator, and the results confirm the effectiveness of the proposed method.