Series arc fault detection and identification based on improved residual network by PSO

With the development of industrialization, the requirements for power safety are getting higher and higher. Low voltage series arc fault is the main cause of electrical fire, and it has high concealment because it is directly connected to the circuit and the current is lower than the loop current. Aiming at the problems of difficult feature extraction and poor recognition accuracy in low voltage series arc fault detection and recognition method, this paper proposes a low voltage series arc fault recognition method based on improved residual network. The improvement of the traditional residual network includes the following three aspects : First, the particle swarm optimization ( PSO ) algorithm is used to optimize the residual network, so as to quickly and accurately fit the fault arc characteristic quantity ; secondly, the dual attention network module ( DANET ) is added to the convolutional neural network detection algorithm model. Finally, the RELU activation function of the traditional ResNet network is replaced by the RELU-Softsign activation function to improve the convergence speed of the network and strengthen batch standardization. Through comparative experiments, the results show that the recognition accuracy of the improved residual network is 98 %, and the recognition accuracy and convergence speed are better than other models.