<b>Robust Road Accident Prediction via Multi-</b><b>Multi modal fusion</b><b>Modal Grey Markov Chain and Adversarial Meta-Learning</b>

Rapid urbanization and the proliferation of intelligent transportation systems have amplified the complexity of road accident prediction, where challenges of data sparsity, multi-modal heterogeneity, and dynamic environmental interference persist. To address these issues, this study proposes a robust prediction framework that integrates multi-modal grey Markov chains with adversarial meta-learning. The model adaptively allocates modality weights through grey relational analysis, enhancing sensitivity to extreme weather and semantic risk factors, while dynamic state partitioning ensures real-time adaptability of large-scale road networks.快速城市化进程与智能交通系统的普及，进一步加剧了道路交通事故预测任务的复杂性，当前该领域仍面临数据稀疏、多模态异质性以及动态环境干扰等核心挑战。为解决上述问题，本研究提出一种融合多模态灰色马尔可夫链与对抗性元学习的鲁棒预测框架。该模型通过灰色关联分析自适应分配模态权重，提升了对极端天气与语义化风险因素的感知灵敏度；同时采用动态状态划分机制，保障了大规模道路网络的实时适配性。