影响因子：9.3
DOI码：10.1016/j.tre.2026.105300
教研室：TURBO LAB - 交通与城市运行实验室
发表刊物：Transportation Research Part E: Logistics and Transportation Review
摘要：Vehicle routing problems (VRP) and their variants, such as team orienteering problems and prize-collecting VRPs, typically assume that service locations and prizes are known before optimization. However, in observation-driven applications such as environmental inspection, emergency response, and humanitarian relief, candidate locations and their uncertain prizes often need to be inferred from massive and noisy observation data. Here, prize refers to the expected benefit of visiting a node, such as inspection value, risk-reduction value, or service benefit, depending on the application context. To this end, this paper develops an observation-driven data pipeline and integrated solution framework with a two-stage distributionally robust prize-collecting vehicle routing problem (DR-PCVRP). The first stage devises an adaptive dynamic clustering algorithm to aggregate spatially concentrated observations into candidate nodes with empirical prize samples, achieving expected amortized 𝑂(log4 𝑚) per-update complexity for fixed-dimensional spatial data, where 𝑚 denotes the current clustering data size. The second stage characterizes prize uncertainty through a Wasserstein ambiguity set and develops a hybrid solution framework that combines deep-reinforcement-learning-based warm starts with neighborhood-search refinement. Experiments on standard test datasets demonstrate the effectiveness of the proposed clustering algorithm and hybrid routing algorithm in terms of solution quality and computational efficiency, particularly for large-scale instances. The results show that the framework can efficiently construct candidate nodes from large-scale observations and generate robust routes evaluated under the exact Wasserstein DRO objective. Finally, a real-world case study of urban environmental management involving more than 100 million GPS records in Chengdu confirms the framework’s scalability and practical applicability, with the Exact DRO solution achieving the lowest mean out-of-sample loss compared with stochastic programming and robust optimization.
论文类型：期刊论文
论文编号：105300
是否译文：否
发表时间：2026-10-16
收录刊物：SCI
发布期刊链接：https://www.sciencedirect.com/science/article/pii/S1366554526006381