Fault section location method of active distribution network based on GAF and improved dual-channel CNN-BiLSTM parallel network

Abstract Aiming at the problems of incomplete analysis of single channel data, weak extraction of fault characteristics and low accuracy of section location when ground fault occurs in active distribution network, a parallel network fault section location method based on Gramian Angular Field (GAF) and improved dual-channel CNN-BiLSTM is proposed. Firstly, the zero-sequence current at both ends of the fault line section is collected, and the amplitude difference of the zero-sequence current at both ends is made. The amplitude difference of the zero-sequence current of each section is composed of a signal, and the zero-sequence current difference signal is processed by GAF transformation to obtain the GAF feature map. Then, the zero-sequence current difference signal is input into the bidirectional long short-term memory network (BiLSTM) branch to mine the timing characteristics, and the GAF feature map is input into the convolutional neural network (CNN) branch to extract the feature map information. Finally, the features extracted from the two channels are fused, and the multi-head attention mechanism is introduced to enhance the features. The simulation results show that the proposed section location method can effectively identify the fault section under different fault conditions. The accuracy of section identification is 100%, and it has strong anti-noise interference ability. This study is valuable for fault sector localization methods in active distribution networks.