题   目：Survival-Analysis-Driven Robust Inventory Management for Automotive Aftermarket

嘉　宾：周明龙 副教授 复旦大学

主持人：曾智宇 助理教授 上海交通大学安泰经济与管理学院

时　间：2026年9月16日（周三）14:00-15:30

地　点：上海交通大学安泰经济与管理学院包兆龙图书馆A507

内容简介：

Balancing the competing demands of part availability and cost efficiency is a critical challenge in spare part inventory management in the automotive aftermarket. In this paper, we focus on a multiperiod spare parts inventory management problem with stochastic demands and supply lead time. Recognizing that spare part demand is driven by component lifetimes and customer heterogeneity, we employ survival regression models to estimate demand distributions. However, such estimates are often inaccurate due to intermittent demand patterns and insufficient data. To address distributional ambiguity, we propose a novel survival-analysis-based robust satisficing model, where the nominal demand distribution is informed by survival analysis and customer-side covariates. We model this multiperiod inventory model as a two-stage model. First-stage decisions represent immediate order quantities, while the second-stage captures recourse actions and future costs under uncertainty. The complexity of the second-stage problem necessitates approximating the future cost function, which depends on initial conditions such as on-hand inventory and orders en route. We can adopt existing methods to approximate the second-stage cost; however, it is often a complicated function of uncertain parameters. We propose approximating it using a regression tree model, which yields an interpretable and tractable piecewise-affine second-stage cost function. We introduce a hierarchical tree branching structure to avoid the decision-dependent uncertainty that can arise from general tree structures. We provide a tractable reformulation for the resulting robust inventory management problem and extend to multi-item and multi-supplier settings. Extensive numerical studies, using both synthetic and real-world data from an automotive partner, demonstrate that our model achieves considerable cost reductions compared to benchmark methods, including empirical optimization and sample-based robust optimization. Finally, we provide insights into the benefits of integrating survival analysis into this problem context.

演讲人简介：

周明龙，复旦大学管理学院副教授，博士毕业于新加坡国立大学商学院，曾于新加坡国立大学运营与分析中心进行博士后工作。获得国家级青年人才、上海市高层次人才称号。主要研究数据驱动的鲁棒优化方法及其在供应链管理和提升供应链韧性等问题中的实践，研究工作发表于Operations Research, Manufacturing & Service Operations Management, Production and Operations Management。积极开展业界合作，为多家医疗机构、企业提供排程、调度、供应链管理等咨询建议。入选华人学者管理科学与工程协会早期职业研究员项目（Early Career Fellows Program），担任中国运筹学会数据科学与运筹智能分会（筹）理事。

欢迎广大师生参加！