Online stochastic optimization for dynamic operations management systems

这项研究计划旨在通过开发新的数据驱动的离散优化方法，创造应用于动态运营管理系统的在线随机优化的最新知识。这些问题中的许多都出现在服务方面，比如叫车、送餐、诊所预约、库存管理和劳动力管理。更准确地说，当请求(例如，搭车)进入系统时，运营商将分配一些资源(司机)来满足服务。在云计算和移动应用时代，这些运营问题变得至关重要，因为所有服务都必须提出符合客户需求的实时应用，才能在数十亿美元的不断扩大的市场中具有竞争力。我的研究重点是在线随机优化，用数据驱动算法来解决这些动态问题。在线随机优化提出了一种通过定期重新优化决策计划来有效解决这些问题的范例，以考虑新请求和未来请求到达时的几种情况。我的研究计划的长期目标是开发为动态运营管理问题量身定做的离散优化算法，并在行业内传播在线随机优化技术。特别是，这个五年计划计划1)开发基于改进的积分单纯形的随时数据驱动算法，2)利用在线随机优化来实时解决工业应用。这个项目将帮助培训6名研究生，他们将有机会开发出解决实际工业问题的最先进的方法。例如，这些方法有可能成为改变游戏规则的技术，因为它们将为乘车服务或医院库存管理引入新一代实时优化系统。

This research program aims at creating state-of-the-art knowledge on online stochastic optimization applied to dynamic operations management systems by developing new data-driven discrete optimization methods. Many of these problems appear for services such as ride sourcing, food delivery, clinic appointments, inventory management, and workforce management to name a few. More precisely, when a request (e.g., a ride) enters the system, the operator will allocate some resources (a driver) to fulfill the service. These operational problems have become critical at the era of cloud computing and mobile applications, as all the services must propose real-time applications that fit the needs of their clients in order to be competitive in multi-billion dollar expanding markets. My research focuses on online stochastic optimization to solve these dynamic problems with data-driven algorithms. Online stochastic optimization proposes a paradigm to efficiently solve these problems by regularly reoptimizing a decision plan to take into account the new requests and several scenarios on the arrival of future requests. The long-term objectives of my research program aim at developing discrete optimization algorithms tailored for dynamic operations management problems and at disseminating online stochastic optimization techniques in the industry. In particular, this five-year program plans 1) to develop an anytime data-driven algorithm based on the improved integral simplex and 2) to solve industrial applications in real-time with online stochastic optimization. This program contributes to the training of 6 graduate students who will get the opportunity to develop state-of-the-arts methods for real industrial problems. These methods have the potential to be game changing technologies as they will introduce a new generation of real-time optimized systems for ride-sourcing services or hospital inventory management for example.