A Series Arc Fault Identification Method Based on Improved Gramian Angular Field

The AC series arc fault of low-voltage power lines is easy to lead electrical fires and cause personal and property losses. The detection of series fault arc in low-voltage lines is mainly based on the extraction of arc fault features of current signals, which cannot make full use of all the information contained in the current waveform. In order to avoid the loss of valuable feature information in the current waveform, this paper proposes a 2D image coding method of current time series based on Gramian angular field (GAF), and uses computer vision technology for time series classification. Aiming at the problem that the arc feature information contained in the image is not obvious due to the limitation of computing resources and the pixel size of the image should be as small as possible, this paper improves the traditional GAF coding method. Based on the zero-current period and power features of arc current, the second color coding is performed on the image encoded by GAF in HSV color space to better express the original signal features. A series arc fault experimental platform was built, and the normal operation current and arc fault current of various electrical appliances were collected. The convolutional neural network and the improved image coding method are used to classify and identify the collected current. The results show that the average of the improved GAF coding method is 99.7 %, which has better accuracy than the traditional GAF.