基于SDAE-SVM的高压电缆局部放电类型识别OA

A deep learning method based on an improved stacked denoising autoencoder(SDAE-SVM)is proposed for pattern recognization of partial discharge(PD)signals generated by different insulation defects in high-voltage cables.First,PD tests are conducted on five types of artificial defects in a high-voltage laboratory,and 3 500 sets of PD instantaneous pulses are extracted to construct 34 types of characteristic parameters.Then,the principles and network architecture of SDAE-SVM are introduced in detail.After that,the proposed model is used to recognize the PD signals of different types of defects and the pattern recognition accuracy of 93.56%is obtained.Moreover,the the layer-wise outputs of the SDAE-SVM are visualized using t-distributed stochastic neighbor embedding(t-SNE),illustrating the essence of layer-wise optimization of the deep neural network SDAE-SVM.Finally,the proposed method is compared with back propagation neural network(BPNN),support vector machine(SVM)and stacked de-noising autoencoders(SDAE).The results show that compared with BPNN,SVM,and SDAE,the overall recognition accuracy of SDAE-SVM has increased by 7.46%,6.70%,and 1.37%,respectively,demonstraing high engineering application value.