Partial discharge pattern recognition of DC XLPE cables based on convolutional neural network

In order to deal with the limitations on the feature extraction of strong random signals in DC XLPE cables, this paper proposes a self-adaptive pattern recognition method based on convolutional neural network (CNN). Convolutional Architecture for Fast Feature Embedding (Caffe) has great performance on image recognition using CNN. Four typical insulation defects are designed and PD signals are collected for pattern recognition. Four different Caffe frameworks are constructed to analyze the impact of the network structures and solver parameters on training effect. Compared with Quick-CIFAR-IO and original Alexnet network, the modified Alexnet network proposed by this paper has great adaptability to pattern recognition of partial discharges in DC XLPE cables.