Scholarlink

<p>Accurate prediction of road accidents remains challenging due to data sparsity, dynamic interference, and heterogeneous multi-modal inputs in intelligent transportation systems. To address these issues, this study develops a Multi-Modal Grey Markov Adversarial Optimization (GM-AO) framework that integrates grey relational analysis, dynamic Markov modeling, and adversarial meta-learning into a unified architecture. The grey-relational module adaptively allocates modality weights among traffic flow, weather, and semantic data, while an improved modularity-based dynamic partitioning algorithm captures temporal transitions with high structural consistency. The adversarial meta-learning component enhances cross-scenario adaptability and robustness under noisy or perturbed conditions. Experiments on multi-city traffic networks verify that GM-AO achieves higher accuracy, faster recovery, and better scalability than state-of-the-art baselines, demonstrating reliable real-time performance under complex urban conditions. Specifically, the model reduces prediction error by nearly 40%, maintains a structural consistency index above 0.85, and recovers from strong adversarial attacks within 60&#xa0;s. Overall, this study proposes a hybrid framework integrating multi-modal grey Markov chains and adversarial meta-learning, with dynamic weight allocation and collaborative optimization mechanisms that enhance robustness, interpretability, and large-scale deployability in traffic-accident prediction.</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

Research on robust prediction model for road accidents based on multi modal grey Markov chain—collaborative optimization with adversarial meta-learning and dynamic state partitioning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月28日

查看全文
  • Jingli Jia,
  • Jiqiang Zhang,
  • Yongfang Zhu

摘要

Accurate prediction of road accidents remains challenging due to data sparsity, dynamic interference, and heterogeneous multi-modal inputs in intelligent transportation systems. To address these issues, this study develops a Multi-Modal Grey Markov Adversarial Optimization (GM-AO) framework that integrates grey relational analysis, dynamic Markov modeling, and adversarial meta-learning into a unified architecture. The grey-relational module adaptively allocates modality weights among traffic flow, weather, and semantic data, while an improved modularity-based dynamic partitioning algorithm captures temporal transitions with high structural consistency. The adversarial meta-learning component enhances cross-scenario adaptability and robustness under noisy or perturbed conditions. Experiments on multi-city traffic networks verify that GM-AO achieves higher accuracy, faster recovery, and better scalability than state-of-the-art baselines, demonstrating reliable real-time performance under complex urban conditions. Specifically, the model reduces prediction error by nearly 40%, maintains a structural consistency index above 0.85, and recovers from strong adversarial attacks within 60 s. Overall, this study proposes a hybrid framework integrating multi-modal grey Markov chains and adversarial meta-learning, with dynamic weight allocation and collaborative optimization mechanisms that enhance robustness, interpretability, and large-scale deployability in traffic-accident prediction.