Identification method for low-voltage Arc fault based on the loose combination of wavelet transformation and neural network

According to statistics, the low-voltage arc fault has become one of the primary factors leading to electrical fires. It´s the fact that traditional circuit protecting devices cannot detect arc fault effectively. Therefore, there are great practical significance and prospect of application in research on arc fault detecting technology. In this paper, the identification method for arc fault based on the loose combination of wavelet transformation and neural network is proposed. In order to realize the recognition of the testing samples of diverse loads, the high-frequency energy in each layer is obtained through decomposing the acquired current waveforms by wavelet, and then these properties are inputted into back-propagation (BP) neural network to constitute a loose wavelet neural network. The adaptive learning rate and momentum term are used to improve the learning speed. Two selecting schemes of nodes in input layer are compared, and the accuracy rate of better scheme reaches 95 percent. By using mean impact value method, the validity of the extracted characteristics in input layer is verified.