Faulty Feeder Identification in Low Current Grounding Active Distribution Networks Based on Dilated Convolution and Attention Mechanism

Abstract

Due to the unclear feature information of single-phase grounding faults in the active distribution network and the influence of fault conditions and environmental noise on existing route selection methods, this paper proposes a route selection method for active distribution network faults based on dilated convolution and the attention mechanism. Firstly, form two-dimensional time-frequency graphs mapped from the zero-sequence current by using wavelet transform. Then, construct a dilated convolutional network model based on the attention mechanism, combine it with the softmax classifier, and train the network model. Finally, the effectiveness of the proposed method is verified by comparison of simulation and experiment. The results show that, when the accuracy reaches 95%, the network constructed in this paper takes 179s less than ordinary convolution, and the accuracy increases by 3.3% under the same training time.