北航经管学院“经济统计论坛”系列讲座
（2026年第8期，总第45期）
讲座题目：Doubly Robust Airline Fleet Assignment over Time-Series Demand with Price Shift
讲座时间：2026.09.14（周一）10:00-11:30
讲座地点：新主楼A618
讲座嘉宾：王曙明，中国科学院大学长聘教授
讲座嘉宾 简介
中国科学院大学经济与管理学院长聘教授。主要从事预测驱动最优化、分布鲁棒优化、统计学习、模型不确定性的理论及应用研究。研究成果发表于《Management Science》、《Operations Research》、《Manufacturing & Service Operations Management》、《Production and Operations Management》、《INFORMS Journal on Computing》以及《Transportation Science》等领域顶刊。主持国家自科基金青A-C项目及面上项目等。目前担任运筹学著名期刊《Computers & Operations Research》的领域主编 (Area Editor) 以及决策科学旗舰期刊《Decision Sciences》的副主编 (Associate Editor).
讲座概要
This paper studies an airline fleet assignment (AFA) problem under demand uncertainty, aiming to maximize expected profit by optimally assigning aircraft types to flight legs and seat capacities to passenger itineraries. We consider a data-driven setting where historical time-series demand observations and associated contextual covariates are available. The problem exhibits several key features: aircraft assignments endogenously influence passenger demand; demand dynamics show both serial and cross-sectional dependencies; and a price-shift phenomenon exists, i.e., a discrepancy between the price estimates used in the planning stage and the price realizations during operations. To systematically address these issues, we propose a structured predict-then-robust-optimize framework. The framework integrates a time-series demand prediction model that incorporates fixed design variables (e.g., fleet assignment and price), random covariates (e.g., weather conditions), and lagged demands, thereby capturing decision-dependent demand and covariate extrapolation. In a decoupled manner, the framework hedges against prediction uncertainty in the conditional demand distribution using a Wasserstein-metric regularity term, and addresses the price-shift effect via a $\phi$-divergence ball, leading to a computationally viable reformulation. We establish finite-sample and consistency performance guarantees for the model solution, explicitly characterizing the influence of price shift. The practical efficacy of our framework is demonstrated through extensive numerical experiments using real-world airline operational data.
