基于深度卷积生成对抗网络与知识蒸馏的聚丙烯电缆缺陷局部放电模式识别方法OA

Polypropylene(PP)cables,as a new type of environmentally friendly material,face challenges in safe and stable operation due to partial discharge.To address this issue,this study proposes a pattern recognition method based on deep convolutional generative adversarial networks(DCGAN)and knowledge distillation(KD)techniques.Firstly,four typical PP cable defects were designed,and a 20 kV withstand voltage test was conducted to obtain 400 phase resolved partial discharge(PRPD)data samples.High-quality PRPD image datasets were then expanded using DCGAN.ResNet-110 was employed as the teacher model for training,and the learned feature knowledge was transferred to the lightweight student model ResNet-20 through KD.Experimental results show that the Frechet inception distance(FID)metric between DCGAN-generated images and original samples reaches as low as 13.22,indicating high similarity.With the introduction of KD,the student network achieves a 63.25％reduction in model parameters while maintaining a classification accuracy of 91.25％and improving inference speed by 4.16 times.The results demonstrate that the proposed recognition for PP cables but also significantly enhances model efficiency,providing theoretical and technical support for intelligent cable fault diagnosis.