Approximate Dynamic Programming for Perishable Asset Management with Applications in Dynamic Pricing, Capacity Allocation and Revenue Management

许多行业面临着在对资产的需求是随机的并且不同资产相互作用的环境中管理多种类型的易腐实物资产的问题。例子很多。时尚商品零售商为作为替代品的服装做出库存决策。航空公司为数百万个行程产品定价，这些产品使用同一航空公司网络上的飞机容量。保健诊所为不同优先事项的病人分配预约和手术室的位置。尽管这些资产管理问题的经常性，用于同时管理多个易腐资产的库存的优化模型和算法在很大程度上限于完全忽略不确定性的确定性方法或忽略系统随时间的演变的静态方法。该基金将资助开发复杂的模型和算法，以解决各种环境中出现的大规模易腐资产管理问题。这项研究的结果将对定价、产能分配、产品组装和收入管理问题特别有用。我们的目标是开发优化模型和算法，（i）适当地捕捉系统中的随机性，（ii）易于扩展以处理现实的问题大小，（iii）独立于特定的应用程序设置应用，（iv）提供性能保证，以及（v）对数据可用性做出合理的假设。该计划旨在证明资产管理问题具有共同的特征，并且可以为它们开发统一的方法。不确定性下的决策课程将大大受益于这项研究。管理模拟游戏，让学生掌握一个现实的资产管理场景，将编程并纳入课程。高中和本科生将获得早期的研究经验。将为定价和收入管理问题编制数据集，并提供给研究界。

Many industries face the problem of managing multiple types of perishable physical assets in an environment where the demand for the assets is random and the different assets interact with each other. Examples are many. Fashion good retailers make stocking decisions for clothing items that act as substitutes. Airlines price millions of itinerary products that use the aircraft capacities over the same airline network. Health clinics allocate appointment and operating room slots to the patients with different priorities. Despite the recurrent nature of these asset management problems, the optimization models and algorithms for simultaneously managing the inventories of multiple perishable assets are largely limited to either deterministic approaches that ignore the uncertainty altogether or static approaches that ignore the evolution of the system over time. This grant provides funding for the development of sophisticated models and algorithms for large-scale perishable asset management problems that arise in a variety of settings.The results of this research will be particularly useful for pricing, capacity allocation, product assembly and revenue management problems. The goal is to develop optimization models and algorithms that (i) appropriately capture the randomness in the system, (ii) scale easily to handle realistic problem sizes, (iii) apply independent of the particular application setting, (iv) provide performance guarantees, and (v) make reasonable assumptions about data availability. The plan is to demonstrate that asset management problems share common features and one can develop a unified approach for them. The course offerings on decision-making under uncertainty will significantly benefit from this research. Management simulation games, which put the students in command of a realistic asset management scenario, will be programmed and incorporated into the courses. High school and undergraduate students will gain early research experience. Data sets for pricing and revenue management problems will be developed and made available to the research community.