Some reports reveal that a number of home fires are caused by electric arc faults on low voltage electric power lines. The arc fault circuit interrupter (AFCI) is a device to detect arc faults. However, the accurate rate of detection by commercial AFCIs is still imperfect. In this paper, an intelligent method is intended to improve the detection ability. A wavelet based stationary non-stationary filter (WTST-NST) is employed to eliminate the impulse components of measurement data. Then, the discrete wavelet transform (DWT) and the multi-resolution are used to obtain the signal energy in the corresponding frequency band. A back propagation neural network (BPNN) is trained by using the signal energy. After the training procedure, the BPNN is able to detect the occurrence of serial arc faults. Finally, the detection results of the system under several loading conditions and operation conditions are compared with a commercial AFCI.

Included in

Engineering Commons