Scholarlink

<p>To analyze how dynamic operational conditions (e.g., environment changes, traffic surges) drive risk propagation in airport movement areas, this study proposes a model integrating complex network theory and reinforcement learning to decode causal relationships from historical incidents. We construct a risk propagation network and leverage causal convolutional reinforcement learning (CCRL) to dynamically quantify inter-node causal strength through temporal pattern mining. First, an enhanced grey relational model (with standardized discrimination coefficients) identifies 63 critical risk factors from multi-source data, addressing extremum-induced distortion in traditional methods. Second, a capacity-load propagation framework classifies nodes into impedance/cumulative types for heterogeneous risk modeling. Third, the CCRL framework facilitates dynamic adjustment of propagation weights through continuous updates of time-varying causal strengths. Experimental results show that our model outperforms Dynamic Bayesian Networks (DBN) by 4.1% in prediction accuracy under data scarcity (n = 1067 incidents) and strong time-varying conditions. The model objectively identifies 63 critical risk factors and enables targeted control strategies that reduce risk diffusion indices by 20%, providing a novel and effective approach for deconstructing time-varying risk propagation in airport movement areas.</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

Dynamic causal weighting-based risk propagation modeling for airport movement areas

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月14日

查看全文
  • Wei Wu,
  • Jiayi Lin,
  • Ming Wei,
  • Xinglong Wang,
  • Longfei Zhu

摘要

To analyze how dynamic operational conditions (e.g., environment changes, traffic surges) drive risk propagation in airport movement areas, this study proposes a model integrating complex network theory and reinforcement learning to decode causal relationships from historical incidents. We construct a risk propagation network and leverage causal convolutional reinforcement learning (CCRL) to dynamically quantify inter-node causal strength through temporal pattern mining. First, an enhanced grey relational model (with standardized discrimination coefficients) identifies 63 critical risk factors from multi-source data, addressing extremum-induced distortion in traditional methods. Second, a capacity-load propagation framework classifies nodes into impedance/cumulative types for heterogeneous risk modeling. Third, the CCRL framework facilitates dynamic adjustment of propagation weights through continuous updates of time-varying causal strengths. Experimental results show that our model outperforms Dynamic Bayesian Networks (DBN) by 4.1% in prediction accuracy under data scarcity (n = 1067 incidents) and strong time-varying conditions. The model objectively identifies 63 critical risk factors and enables targeted control strategies that reduce risk diffusion indices by 20%, providing a novel and effective approach for deconstructing time-varying risk propagation in airport movement areas.