Scholarlink

<p>Cyber-physical systems (CPS) encompass intricate and expansive networks equipped with an array of sensors and actuators, resulting in the generation of substantial volumes of multivariate time series (MTS) data. Anomaly detection in MTS is an effective way to guarantee system reliability. However, the high-dimensional nature of MTS and the complex time correlation make it difficult for the existing anomaly detection methods to capture temporal features efficiently. In addition, existing methods can not locate the root cause of the anomaly, which is more practical for fault restoration in Water Treatment CPS. Anomalies may originate from targeted attacks on specific sensors. To address the two challenges, we propose a Matrix Concatenation Feature Fusion-based MTS Anomaly Detection and Diagnosis Algorithm, named MCMADD, with two special components: (1) a data embedding and matrix concatenation feature fusion module for capturing temporal features; (2) an anomaly diagnosis module for root cause identification. We evaluate MCMADD on two publicly available datasets of smart waterworks. Experimental results show that MCMADD outperforms state-of-the-art approaches in both anomaly detection efficiency and anomaly diagnosis performance.</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

Matrix concatenation feature fusion-based multivariate time series anomaly detection and diagnosis algorithm in water treatment cyber-physical systems

  • 期刊论文

  • 发表日期：

    2026年1月17日

查看全文
  • Shiming He,
  • Keyao Feng,
  • Kaixuan Meng,
  • Meng Guo,
  • Diqing Liang,
  • Pradip Kumar Sharma

摘要

Cyber-physical systems (CPS) encompass intricate and expansive networks equipped with an array of sensors and actuators, resulting in the generation of substantial volumes of multivariate time series (MTS) data. Anomaly detection in MTS is an effective way to guarantee system reliability. However, the high-dimensional nature of MTS and the complex time correlation make it difficult for the existing anomaly detection methods to capture temporal features efficiently. In addition, existing methods can not locate the root cause of the anomaly, which is more practical for fault restoration in Water Treatment CPS. Anomalies may originate from targeted attacks on specific sensors. To address the two challenges, we propose a Matrix Concatenation Feature Fusion-based MTS Anomaly Detection and Diagnosis Algorithm, named MCMADD, with two special components: (1) a data embedding and matrix concatenation feature fusion module for capturing temporal features; (2) an anomaly diagnosis module for root cause identification. We evaluate MCMADD on two publicly available datasets of smart waterworks. Experimental results show that MCMADD outperforms state-of-the-art approaches in both anomaly detection efficiency and anomaly diagnosis performance.