Deep Neural Networks(DNN)is widely used for fault classification using Partial Discharges(PD)to assess the insulation level of various electrical equipment,but there is a risk of false positives for untrained PD fault data.Based on this,a deep ensemble model based on improved CNN is proposed.First-ly,UHF sensors are used to collect PD signals for seven kinds of GIS insulation defect faults in the field,and PRPD spectra are formed and analyzed.Secondly,the collected data is imported into the model for un-certainty estimation and the confidence value and the threshold value of the model is determined.The influ-ence of the scale of the CNN deep ensemble model on the classification performance is studied again.The results show that the proposed model has good detection performance for unknown partial discharge faults ex-perimental,and has certain engineering practice value.
关键词
气体绝缘开关设备/故障诊断/局部放电检测/卷积神经网络/深度集成模型
Key words
gas insulated switchgear/fault diagnosis/partial discharge detection/convolutional neural net-works/deep integration model
维普
万方数据