The reduction in the current level produced when a series arc fault occurs causes a common purpose circuit breaker protection to fail to identify series arcs. Continued arcing can cause a fire. This paper presents two ac series arc fault detection methods. One is a series arc fault identification method based on the gramian transform, which is called the gramian angular field arc network (GAF-ArcNet). Another is based on the discrete wavelet transform, which is called the discrete wavelet transform arc network (DWT-ArcNet). At the same time, an experimental platform capable of automatically recording normal and arc fault current waveforms is designed in order to collect data set. Experimental results show that under the condition of ignoring the load type, the accuracy of arc fault recognition of the GAF-ArcNet model is 99.97%, which is better than that of the DWT-ArcNet model, with an accuracy rate of 99.91%. If the load type is identified, the arc fault identification accuracy of the DWT-ArcNet model is 99.44%, which is better than that of the GAF-ArcNet model with an accuracy rate of 98.74%. Meanwhile, the DWT-ArcNet model has no false positives compared to the GAF-ArcNet model.

arc fault, gramain angular field, neural network, discrete wavelet transformation.