关键词: 深度学习, 卷积神经网络, 闪络实验, 紫外放电, 特征提取, 特征融合, 无损检测, 绝缘子

Abstract:

Currently, it is difficult for the electric power transmission and distribution network to detect the insulator's insulation deterioration in operation. In view of this problem, on the basis of the analysis of the insulation deterioration mechanism, this study proposes an intelligent method based on the feature-fused CNN to detect whether the insulator produces the insulation deterioration. Firstly, the flashover experiment is conducted on the insulator to produce three states as follows: no discharge, weak discharge and strong discharge. Subsequently, the ultraviolet imager is applied to collect the CNN insulator's ultraviolet images in different discharge state to form the sample library. Then, the CNN is adopted to extract the sample's features and fuse the extracted shallow and deep features. Finally, the fusion feature map is identified and classified to judge whether the insulator is degraded. The result reveals that this method has a high detection rate, and the highest correct rate reachs 97.4%. Compared to the AlexNet algorithm, this method has a higher accuracy and quicker convergence speed.

Key words: deep learning, CNN, flashover experiment, ultraviolet discharge, feature extraction, feature fusion, nondestructive testing, insulator