Scholarlink

<p>Accurate risk prediction remains a critical challenge in reliability engineering and system safety, particularly in complex systems characterized by interdependent temporal progression and spatial co-occurrence patterns. While existing approaches predominantly focus on either temporal dynamics or co-occurrence relationships, this study introduces a novel spatio-temporal graph learning architecture. First, a dual-matrix graph construction mechanism simultaneously captures spatial risk correlations through co-occurrence frequency analysis and temporal progression patterns using transition probability modeling. Second, an adaptive subgraph extraction module generates system-specific topological representations that preserve both localized risk clusters and directed temporal pathways. Third, a dual-channel graph convolutional network with bilinear interaction fusion facilitates synergistic processing of spatial coexistence features and temporal progression patterns while preserving modality-specific characteristics. Empirical validation across medical diagnosis and vehicular risk domains demonstrates the model’s effectiveness in handling multi-risk coexistence scenarios and long-term progression patterns, significantly outperforming conventional single-modality approaches. The proposed methodology offers a generalizable solution for cross-domain risk prediction tasks. The source code is available at <a href="https://github.com/FanghuaX/HS-TGN">https://github.com/FanghuaX/HS-TGN</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

Hierarchical spatio-temporal graph network for risk prediction

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月30日

查看全文
  • Fanghua Chen,
  • Hong Jia,
  • Wei Zhou,
  • Liwei Zhu,
  • Lin Xiao

摘要

Accurate risk prediction remains a critical challenge in reliability engineering and system safety, particularly in complex systems characterized by interdependent temporal progression and spatial co-occurrence patterns. While existing approaches predominantly focus on either temporal dynamics or co-occurrence relationships, this study introduces a novel spatio-temporal graph learning architecture. First, a dual-matrix graph construction mechanism simultaneously captures spatial risk correlations through co-occurrence frequency analysis and temporal progression patterns using transition probability modeling. Second, an adaptive subgraph extraction module generates system-specific topological representations that preserve both localized risk clusters and directed temporal pathways. Third, a dual-channel graph convolutional network with bilinear interaction fusion facilitates synergistic processing of spatial coexistence features and temporal progression patterns while preserving modality-specific characteristics. Empirical validation across medical diagnosis and vehicular risk domains demonstrates the model’s effectiveness in handling multi-risk coexistence scenarios and long-term progression patterns, significantly outperforming conventional single-modality approaches. The proposed methodology offers a generalizable solution for cross-domain risk prediction tasks. The source code is available at https://github.com/FanghuaX/HS-TGN.