Real-time Planning and Optimization under Uncertainty in Logistics and Transportation

物流和运输相关业务通常占项目总成本的很大一部分，并与主要污染排放有关。私营部门和公共部门都呼吁提高效率，减少交通运输，以缓解交通拥堵和污染排放。虽然数学优化是为这些领域提供有效规划解决方案的最成功工具之一，但传统的模型是基于对问题参数的预测。从不同来源获取和存储数据的信息技术能力的不断提高，为运筹学社区提供了一个机会，可以在实践中提高生产计划解决方案的性能。***在这个研究项目中，我重点关注物流和运输中的两个关键应用，即车辆共享系统的重新部署，特别是自行车共享系统和不久的将来自动驾驶（无人驾驶）出租车的场景，以及覆盖移动个人客户需求的移动设施的设施位置。这些系统在缓解上述问题方面具有特别大的潜力，因为通常可以获得数据（例如在电信网络中）。然而，他们往往会受到难以预测的事件（例如天气状况）的影响，这使得可靠地满足客户需求成为一个持续的挑战。我建议研究优化模型和解决方法，通过考虑（预测）输入数据的不确定性、优化模型中通常未考虑的外部数据以及在整个执行过程中发布的实时信息，为实际规模的规划问题提供解决方案。我们将解决三个主要挑战：(i)问题是组合的，由于应用程序的性质，必须大规模解决，因此即使有完美的信息也难以解决；（ii）问题是动态的，在整个规划过程中必须调整决策；（iii）问题的输入数据具有不确定性。***本研究项目中开发的模型和解决方案方法可能对物流和运输行业产生强烈影响，并为决策者提供指导，例如通过改善电信服务，增加车辆共享系统的效用，减少交通和车辆拥有量。该计划促进了我们对信息不确定性如何转化为最佳决策，来自外部来源的数据如何有助于改进决策，以及迄今为止很少探索运筹学与机器学习之间的相互作用的理解。学生将被训练成下一代数据科学家，专门从事规范分析，并将从良好的就业市场中受益。

Logistics and transportation related operations typically account for large parts of total project costs and are linked to major pollution emission. Both private and public sectors call for more efficiency and reduction of transportation to relieve traffic and pollution emission. While mathematical optimization has been among the most successful tools to provide efficient planning solutions for these domains, models have been traditionally based on forecasts of the problems' parameters. The ongoing increase of information technology capacity to capture and store data from different sources offer an opportunity to the Operations Research community to improve the performance of produced planning solutions in practice.***In this research program, I focus on two key applications in logistics and transportation, namely the redeployment in vehicle sharing systems, in particular bike sharing systems and the near-future scenario of autonomous (driver-less) taxis, as well as facility location with mobile facilities that cover customer demands of moving individuals. Those systems have a particularly high potential to relieve the above mentioned issues, as data is often available (e.g. in telecommunication networks). However, they tend to suffer from events that are difficult to forecast (e.g. weather conditions), making a reliable satisfaction of customer demands a constant challenge. I propose to investigate optimization models and solutions methods that provide solutions to realistically sized planning problems by taking into consideration the uncertainty in the (forecast) input data, external data that has typically not been considered in optimization models, as well as real-time information that is released throughout the execution process. We will tackle three main challenges: (i) the problems are combinatorial and, by nature of the applications, have to be solved on large scale and are therefore difficult to solve even when perfect information is available; (ii) the problems are dynamic and decisions have to be adjusted throughout the planning; (iii) the problems' input data is subject to uncertainty.***The models and solution methods developed in this research program may have a strong impact in logistics and transportation industries and provide guidance to decision-makers, e.g. by improving telecommunication services, increasing the utility of vehicle sharing systems and reducing traffic and vehicle-ownership. The program advances our understanding of how information uncertainty translates into optimal decisions, of how data from external sources can be useful to improve decision-making, and of the so-far little explored interplay between operations research and machine learning. Students will be trained as the next data scientist generation specialized in prescriptive analytics and will benefit from an excellent job market.