Fault Line Selection of Distribution Network Based on Modified CEEMDAN and GoogLeNet Neural Network
Xiaorong Cheng, Bao-Jing Cui, S. Hou
Abstract
Aiming at the difficulty of single-phase grounding fault line selection in a small current grounding system, a distribution network fault line selection method based on modified CEEMDAN and convolutional neural network is proposed. Firstly, the random forest and multiscale permutation entropy are used to modify the Complete Ensemble Empirical Mode Decomposition Adaptive Noise algorithm (CEEMDAN), and the zero-sequence current of each line is decomposed into a series of intrinsic mode functions through the modified CEEMDAN (MCEEMDAN) algorithm. Secondly, the intrinsic mode function is transformed into the image formation by using the signal-image conversion method, and the generated image is upgraded to a three-dimensional color image by combining pseudo-color coding technology. Finally, the color images converted from the signals of each line are fused as the input of the GoogLeNet network, and the fault line selection of the distribution network is realized in the form of probability output by the Softmax function. The experimental results show that the proposed method has not only strong feature extraction ability and high recognition accuracy but also has good anti-noise and robustness.