DDDAS-SMRP: Robustness and Performance in Data-Driven Revenue Management

这笔赠款为收入管理中不确定性的数据驱动模型的开发和分析提供资金，该模型将纳入决策者的风险偏好，并将实验测量动态集成到计算方法中。第一套模型将集中在基于数量的收入管理，其中的需求分布是独立的决策者的行动。第二套模型将把数据驱动的范例应用于基于价格的收入管理，其中需求分布以及数据受到决策变量的影响。具体而言，本研究项目将调查以下三个问题：（i）如何将额外的数据纳入算法以产生鲁棒的解决方案，特别是当部分历史数据现在可能过时时？(ii)优化模块如何不仅在测量频率方面，而且在监测量方面引导数据收集过程？(iii)如何以有效的方式处理丰富的可用数据，从而使系统状态的清晰度不会以牺牲计算的易处理性为代价？分析将涉及金融风险管理、控制理论、稳健回归和数学规划等领域的技术，以及大规模模拟，以评估该方法的稳健性和性能。如果成功，这项研究将提供一个全面的数学模型，收入管理下的不确定性，将更紧密地连接到现有的数据，而不过分依赖于任何一个假设的需求。它将允许从业者以更直观的方式考虑随机性，因此将促进在工业中使用最先进的优化模型和计算技术。此外，在这项工作中开发的模型和算法将提供一种新的方法来决策的不确定性下的利益，整个运筹学界和广泛的应用领域。

This grant provides funding for the development and analysis of data-driven models of uncertainty in revenue management, which will incorporate the decision-maker's risk preferences and dynamically integrate experimental measurements into the computational approach. A first set of models will focus on quantity-based revenue management, where the demand distribution is independent of the decision-maker's actions. A second set of models will apply the data-driven paradigm to price-based revenue management, where the demand distribution, and hence the data, is affected by the decision variables. Specifically, this research project will investigate the following three questions: (i) How should additional data be incorporated into the algorithm to yield robust solutions, especially when part of the historical data might now be obsolete? (ii) How can the optimization module steer the data collection process, not only in terms of measurement frequency, but also with respect to the quantities monitored? (iii) How can the wealth of available data be manipulated in an efficient manner, so that the added clarity about the state of the system does not come at the expense of computational tractability? The analysis will involve techniques from the fields of financial risk management, control theory, robust regression and mathematical programming, as well as large-scale simulations to assess the robustness and performance of the approach. If successful, this research will provide a comprehensive mathematical model of revenue management under uncertainty that will be more tightly connected to the available data without overly depending on any one assumption about demand. It will allow practitioners to take into account randomness in a more intuitive manner, and hence will promote the use of state-of-the-art optimization models and computational techniques in industry. Moreover, the models and algorithms developed in this work will offer a new approach to decision-making under uncertainty of interest to the whole operations research community and with applications to a wide array of domains.