Scholarlink

<p>As vehicles become increasingly connected to external networks, communication between vehicular networks and external systems grows more frequent. Detecting anomalies in vehicular networks is critical for ensuring the safety of vehicles and collaborative traffic systems. However, existing anomaly detection methods often fail to account for semantic relationships at the physical mechanism level, leading to high false-positive rates and poor interpretability. To address these challenges, we propose a Causality-Aware Graph Convolutional Network (CA-GCN) framework for vehicular network anomaly detection. The model integrates three key components: a semantic extractor, a causality-aware module, and a temporal graph convolutional network. The semantic extractor captures semantic features from messages, while the causality-aware module evaluates causal relationships between semantics using Normalized Transfer Entropy (NTE), filters redundant relationships via Conditional Mutual Information (CMI), and dynamically optimizes the graph structure. The temporal graph convolutional network then aggregates node features and predicts semantic temporal changes. By learning causal relationships among semantics in vehicular network messages, CA-GCN dynamically updates and refines the causal graph. In experiments on real-world CAN-FD datasets, CA-GCN outperformed baseline methods, accurately identifying attack messages causing semantic temporal anomalies. Additional experiments on SWaT and WADI datasets further validate the model's general applicability.</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

Anomaly Detection in Vehicular Networks Using Causality-Aware Graph Convolutional Networks (CA-GCN)

  • 期刊论文

  • 发表日期：

    2025年7月18日

查看全文
  • Feng Luo,
  • Cheng Luo,
  • JiaJia Wang,
  • Zhihao Li,
  • Zeqi Liao,
  • Qin Liu

摘要

As vehicles become increasingly connected to external networks, communication between vehicular networks and external systems grows more frequent. Detecting anomalies in vehicular networks is critical for ensuring the safety of vehicles and collaborative traffic systems. However, existing anomaly detection methods often fail to account for semantic relationships at the physical mechanism level, leading to high false-positive rates and poor interpretability. To address these challenges, we propose a Causality-Aware Graph Convolutional Network (CA-GCN) framework for vehicular network anomaly detection. The model integrates three key components: a semantic extractor, a causality-aware module, and a temporal graph convolutional network. The semantic extractor captures semantic features from messages, while the causality-aware module evaluates causal relationships between semantics using Normalized Transfer Entropy (NTE), filters redundant relationships via Conditional Mutual Information (CMI), and dynamically optimizes the graph structure. The temporal graph convolutional network then aggregates node features and predicts semantic temporal changes. By learning causal relationships among semantics in vehicular network messages, CA-GCN dynamically updates and refines the causal graph. In experiments on real-world CAN-FD datasets, CA-GCN outperformed baseline methods, accurately identifying attack messages causing semantic temporal anomalies. Additional experiments on SWaT and WADI datasets further validate the model's general applicability.