【报告人简介】

于国栋博士现为山东大学管理学院教授，院长助理，智能工程与管理创新实验班负责人，山东省泰山学者青年专家、省高校优秀青年创新团队负责人，曾在新加坡国立大学工业与系统工程系任博士后研究员。主要研究方向为复杂系统决策与韧性优化，聚焦不确定环境下制造与供应链系统决策问题，从事事前鲁棒性优化、事后响应决策敏捷性优化研究，融合系统工程、数据科学、运筹优化与机器学习方法求解最优方案。相关研究成果发表在领域代表性期刊如Manufacturing & Service Operations Management, INFORMS Journal on Computing, Production and Operations Management, Transportation Research Part B: Methodological, IISE Transactions, Naval Research Logistics 等。主持国家自然科学基金等各类纵向课题14项。提交的相关政策建议获得国家级采纳。

【摘要】

We study data-driven decision-making when direct samples from the target distribution are scarce, but related observational datasets and past operational decisions are available.  The challenge is that heterogeneous sources may be biased relative to the target, while operational decisions are not samples from the target law but optimization outputs that encode distributional information only indirectly.  We develop an operationally informed Bayesian DRO framework that learns the target distribution through a hierarchical multi-source model and incorporates decision data through an inverse-optimization likelihood.  The resulting target posterior predictive distribution defines the center of a Wasserstein ambiguity set, and the radius is calibrated from posterior uncertainty.  We provide both a parametric formulation and a nonparametric hierarchical Dirichlet process extension, and compare the proposed target-predictive construction with pooling, source-predictive barycenter, intersection, and posterior-averaged alternatives.  The analysis establishes finite-dimensional reformulations, finite-sample coverage, out-of-sample guarantees, posterior consistency, contamination robustness, and information-value characterizations for source and operational data.  Experiments on a multi-source newsvendor problem show that hierarchical posterior fusion mitigates source bias, operational decisions tighten the ambiguity set and shield against contaminated observational data, and Fisher-information scores provide an effective rule for acquiring informative decision records.

撰稿：王戈 审核：许建军 万丛颖 单位：现代供应链管理研究院