A rise in traffic accidents has led to both traffic congestion and subsequent secondary accidents. Effectively addressing this issue requires rapid accident investigation and management. In this paper, we aim to improve the efficiency of traffic accident assessment and investigation with the aid of drone technologies. Our approach involves strategically pre-positioning drones, enabling traffic supervisory agencies to dispatch drones immediately upon receiving an accident report. Methodology-wise, we present a data-driven robust stochastic optimization (RSO) model, which encapsulates the uncertainty of traffic accidents within a scenario-wise Wasserstein ambiguity set. To the best of our knowledge, this is the first study that incorporates covariates, i.e., weather conditions, into the Wasserstein ambiguity set with the CVaR metric. We demonstrate that the proposed RSO model can be reformulated into a mixed-integer programming model, allowing an efficient solution approach. Via a real-world dataset of London traffic accidents, we validate the practical applicability of the RSO model. Across various parameter settings, our RSO model exhibits superior out-of-sample performance compared with various benchmark models. The numerical results yield valuable insights for traffic supervisory agencies.

中文翻译：

---

数据驱动无人机预定位交通事故快速评估

交通事故的增加导致了交通拥堵以及随之而来的二次事故。有效解决这一问题需要快速的事故调查和管理。在本文中，我们旨在借助无人机技术提高交通事故评估和调查的效率。我们的方法包括战略性地预先部署无人机，使交通监管机构能够在收到事故报告后立即派遣无人机。在方法论方面，我们提出了一种数据驱动的鲁棒随机优化（RSO）模型，该模型将交通事故的不确定性封装在场景方面的 Wasserstein 模糊集内。据我们所知，这是第一项将协变量（即天气条件）纳入具有 CVaR 指标的 Wasserstein 模糊度集中的研究。我们证明了所提出的 RSO 模型可以重新表述为混合整数规划模型，从而实现有效的解决方法。通过伦敦交通事故的真实数据集，我们验证了 RSO 模型的实际适用性。在各种参数设置下，与各种基准模型相比，我们的 RSO 模型表现出了卓越的样本外性能。数值结果为交通监管机构提供了宝贵的见解。