Scholarlink

<p>Aiming at the problems of low accuracy of the existing anomaly detection models and the setting of anomaly thresholds relying on prior experience. This thesis proposes an adaptive threshold anomaly detection method based on GAT-Diffusion-VAE (Graph Attention Networks-Diffusion-Variational AutoEncoder). This method introduces the Graph Attention Networks (GAT) and Variational AutoEncoder (VAE) modules in the Diffusion model, and mines the spatio-temporal dependencies among device features through GAT. The combination of VAE and diffusion model strengthens the modeling ability of the model for complex data distribution, thereby improving the accuracy of anomaly detection. Meanwhile, this method also designs an adaptive threshold calculation mechanism based on K-means clustering (K-means), which can achieve dynamic adjustment of abnormal thresholds and effectively solve the problem of reliance on human experience in traditional methods. To verify the effectiveness of this method, experiments were conducted on the SWaT public dataset and the dataset of vacuum dry pumps of a certain company in this thesis. The results show that the F1-score of the adaptive threshold unsupervised anomaly detection method based on GAT-Diffusion-VAE proposed in this thesis is superior to the mainstream comparison algorithms.</p>

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

An industrial equipment anomaly detection model with adaptive thresholding based on graph attention networks-diffusion variational autoencoder

  • 期刊论文

  • 发表日期：

    2025年7月24日

查看全文
  • Fengyi Fang,
  • Cheng Wang,
  • Lincong Chen,
  • Xiaobin Liao,
  • Zhaoji Huang

摘要

Aiming at the problems of low accuracy of the existing anomaly detection models and the setting of anomaly thresholds relying on prior experience. This thesis proposes an adaptive threshold anomaly detection method based on GAT-Diffusion-VAE (Graph Attention Networks-Diffusion-Variational AutoEncoder). This method introduces the Graph Attention Networks (GAT) and Variational AutoEncoder (VAE) modules in the Diffusion model, and mines the spatio-temporal dependencies among device features through GAT. The combination of VAE and diffusion model strengthens the modeling ability of the model for complex data distribution, thereby improving the accuracy of anomaly detection. Meanwhile, this method also designs an adaptive threshold calculation mechanism based on K-means clustering (K-means), which can achieve dynamic adjustment of abnormal thresholds and effectively solve the problem of reliance on human experience in traditional methods. To verify the effectiveness of this method, experiments were conducted on the SWaT public dataset and the dataset of vacuum dry pumps of a certain company in this thesis. The results show that the F1-score of the adaptive threshold unsupervised anomaly detection method based on GAT-Diffusion-VAE proposed in this thesis is superior to the mainstream comparison algorithms.