Collaborative Research: Emerging Optimization Methods for Planning and Operating Shared Mobility Systems under Uncertain Budget and Market Demand

网络通信系统的进步使得包括汽车共享和拼车在内的共享移动形式得以使用。在预算和市场需求不确定的情况下，参与规划共享出行系统的政府和私营部门都必须在引入新的共享出行项目和扩大现有项目之间做出选择。在这个项目中，PI将在考虑共享车队规模、类型、位置设计和运营活动（如实时车辆路线、再分配和汽车共享系统收费）的情况下，解决共享移动策略设计中的需求不确定性问题；除了服务区域规划和共享移动。该项目的成功将：(i)推进用于解决新运输问题的优化方法的理论和计算前沿；（ii）影响与关键民用基础设施、供应链和物流以及其他服务行业相关的共享移动应用。教育计划，包括促进女性和未被充分代表的少数群体在科学、工程和管理方面的发展，将通过pi参与密歇根大学和普渡大学的各种教育计划来合作实施。这项合作研究的目标是推导出高保真度、数据驱动的数学模型和可证明有效的数值算法，创新地将优化和强化学习结合起来，用于共享移动系统的设计和运营。具体而言，我们将共享出行需求响应描述为一个多阶段的信息披露过程，并将相应的决策过程抽象为连续的资源规划、分配和任务优先级，并在后期的不同决策中进行调整。我们推导了基于单阶段、两阶段和多阶段随机优化的模型和解决方法，并依赖于对需求分布的充分了解。我们还研究了数据驱动的分布鲁棒优化方法和机器学习方法，以解决模棱两可的需求分布和预算不确定性。本研究的推导、验证和校准旨在(i)制定适当的基于优化的模型，以表征复杂共享移动系统中决策数据的相互依赖性；（ii）在新兴的共享移动服务中部署数据驱动的、无分布的方法来处理多源不确定性的分布模糊性；（iii）整合学习方法以动态适应系统内生性信息，并从多阶段优化过程中增强解决方案；（iv）设计具有求解质量保证的高效计算方法，使模型能够实际应用。

Advances in networked communication systems have enabled the use of shared mobility forms including carsharing and ridesharing. Both government and private sectors engaged in planning shared mobility systems must choose between introducing new, shared mobility programs versus expanding existing ones, under uncertain budget and market demands. In this project, the PI will address the issue of demand uncertainty in design of shared mobility strategies considering shared fleet size, type, location design, and operational activities such as real-time vehicle routing, redistribution, and charging for carsharing systems; in addition to service region planning and shared mobility. The success of this project will: (i) advance both the theoretical and the computational frontiers of optimization methods for use in solving new transportation problems; and (ii) impact applications of shared mobility that relate to critical civil infrastructures, supply chain & logistics, and other service industries. Education plans, including promoting female and underrepresented minority groups in science, engineering, and management, will be collaboratively undertaken through the PIs' involvements in various education initiatives at the University of Michigan and Purdue University.The objective of this collaborative research is to derive high-fidelity, data-driven mathematical models and provably efficient numerical algorithms that innovatively combine optimization and reinforcement learning for shared mobility system design and operations. In specific, we characterize shared mobility demand response as a multi-stage information revealing process, and abstract the corresponding decision process as sequential resource planning, allocation, and task prioritization, adjusted by varying decisions in later stages. We derive models and solution methods based on single-, two- and multi-stage stochastic optimization and dependent on full knowledge of demand distributions. We also investigate data-driven distributionally robust optimization methods and machine learning approaches to address ambiguous demand distributions and budget uncertainty. The derivation, validation, and calibration of this study aim at (i) formulating appropriate optimization-based models to characterize decision-data interdependence in complex shared mobility systems; (ii) deploying data-driven, distribution-free approaches for handling the distribution ambiguity of multi-sourced uncertainties in emerging shared mobility services; (iii) integrating learning approaches to dynamically adapt to endogenous system information and enhancing solutions from multi-stage optimization processes; (iv) designing efficient computational methods with solution quality guarantees to enable practical use of the models.