Scholarlink

<p>Detecting abnormal events in multi-sensor signals is crucial for securing industrial production. Recently, association discrepancy-based anomaly detection models have emerged and demonstrated excellent performance. These models utilize the discrepancy between local and global associations of each timestamp for anomaly detection. However, they ignore the coupling between multiple sensors. To solve this issue, a GRU contrast discrepancy-based timestamp-level anomaly detection model, named GCDAD, is proposed for multi-sensor signals from both temporal and spatial perspectives. <i>Firstly</i>, a GRU-based contrast representation network is designed to generate a contrast dependency-discrepancy for each timestamp from a temporal perspective. <i>Secondly</i>, another contrast representation network is constructed to generate a contrast coupling-discrepancy for each timestamp from a spatial perspective. <i>Finally</i>, a customized scoring strategy is designed to merge temporal and spatial contrast dependencies using different weights, and thus generate an anomaly score for each timestamp. To the best of our knowledge, this is the first GRU variant model in the association discrepancy-based anomaly detection category. Extensive experimental results demonstrate that our GCDAD model outperforms several state-of-the-art (STOA) models. The source code of GCDAD can be downloaded from <a href="https://github.com/cxzing/GCDAD">https://github.com/cxzing/GCDAD</a>.</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

GRU contrast discrepancy-based anomaly detection for multi-sensor signals from temporal and spatial perspectives

  • 期刊论文

  • 发表日期：

    2025年11月11日

查看全文
  • Lei Chen,
  • Xinzhe Cao,
  • Tingqin He,
  • Haomiao Guo,
  • Xuxin Liu

摘要

Detecting abnormal events in multi-sensor signals is crucial for securing industrial production. Recently, association discrepancy-based anomaly detection models have emerged and demonstrated excellent performance. These models utilize the discrepancy between local and global associations of each timestamp for anomaly detection. However, they ignore the coupling between multiple sensors. To solve this issue, a GRU contrast discrepancy-based timestamp-level anomaly detection model, named GCDAD, is proposed for multi-sensor signals from both temporal and spatial perspectives. Firstly, a GRU-based contrast representation network is designed to generate a contrast dependency-discrepancy for each timestamp from a temporal perspective. Secondly, another contrast representation network is constructed to generate a contrast coupling-discrepancy for each timestamp from a spatial perspective. Finally, a customized scoring strategy is designed to merge temporal and spatial contrast dependencies using different weights, and thus generate an anomaly score for each timestamp. To the best of our knowledge, this is the first GRU variant model in the association discrepancy-based anomaly detection category. Extensive experimental results demonstrate that our GCDAD model outperforms several state-of-the-art (STOA) models. The source code of GCDAD can be downloaded from https://github.com/cxzing/GCDAD.