Scholarlink

To address insufficient time-frequency feature extraction, limited samples, and poor generalization in partial discharge (PD) pattern recognition, this study proposes a hybrid method for oil-immersed transformers. Adaptive Optimal-Kernel (AOK) time-frequency representation is used to suppress cross-term interference, generating high-resolution (300 × 375) time-frequency matrices that preserve dynamic frequency characteristics of PD signals. Denoising Diffusion Probabilistic Models (DDPM) augment data, expanding each PD class from 100 to 200 samples while retaining original feature distributions. A modified ResNet combined with SVM is adopted for classification, achieving 99.6% accuracy. This method effectively distinguishes similar PD types (e.g., SD, SD-p, SD-p-w) and enhances model generalization, providing a reliable solution for PD pattern recognition.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Partial Discharge Pattern Recognition in Oil-Immersed Transformers via DDPM-Based AOK Time-Frequency Data Augmentation

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Rui Liu,
  • Kai Zhou,
  • Yi Ao,
  • Lei Jin,
  • Yangze Lu,
  • Zihan Teng,
  • Zhixian Zhang,
  • Xingang Chen

摘要

To address insufficient time-frequency feature extraction, limited samples, and poor generalization in partial discharge (PD) pattern recognition, this study proposes a hybrid method for oil-immersed transformers. Adaptive Optimal-Kernel (AOK) time-frequency representation is used to suppress cross-term interference, generating high-resolution (300 × 375) time-frequency matrices that preserve dynamic frequency characteristics of PD signals. Denoising Diffusion Probabilistic Models (DDPM) augment data, expanding each PD class from 100 to 200 samples while retaining original feature distributions. A modified ResNet combined with SVM is adopted for classification, achieving 99.6% accuracy. This method effectively distinguishes similar PD types (e.g., SD, SD-p, SD-p-w) and enhances model generalization, providing a reliable solution for PD pattern recognition.