Supply Chain Analytics under Disruptive Uncertainty

COVID全球大流行的影响已被证明对多个行业的供应链具有高度破坏性。例如，在零售背景下，零售商经历了多个品类产品的恐慌性抢购的影响。在航空航天行业，运营停顿和需求突然下降造成了生产和运营的冲击和中断。关键零部件面临重大且不可预测的延误。供应链中以低成本实现效率的流程和操作变得越来越复杂，因此往往容易出现极具破坏性的情况。在这种情况下，主要依赖于供应链决策过程中常用的大量历史数据的工具通常不能适应其运作环境的突然变化。这一研究计划的独特之处在于，它与具有高度挑战性的大流行和破坏情况具有实际意义，并整合了运筹学和机器学习技术，以在数据质量和数量不足的情况下增强分析。我们的研究计划利用预测性和规范性分析方法，通过两个重要的对策来增强供应链的稳健性和弹性，即(I)中断前的风险缓解计划，以主动为供应链做好准备，使其在中断之前具有足够的健壮性和弹性，以及(Ii)中断后的恢复行动，以迅速检测、评估和应对手头的情况。我们将制定规范的方法，在供应链的战略、战术和运营层面确定风险缓解计划，更具体地说，在稳健的供应链网络设计、弹性生产和分销规划以及稳健的运输规划方面。为了使决策者能够在中断发生后进行处理，我们将把精力集中在三个重要的应用上，即零售中的需求异常检测和在线补货、制造和生产计划中的动态调度以及实时车辆调度和交付。分析工具的开发将基于用于规范能力的多阶段随机和稳健优化方法、高级分解方法、启发式方法和强化学习算法，这些方法通过用于生成对抗性(破坏性)情景、降低维度和构建情景不确定性表示(歧义集)的统计和机器学习方法得到加强。这项研究计划的大部分将与我们的零售和工业合作伙伴合作完成。这一研究计划战略性地补充了我们加拿大供应链分析研究主席下的现有研究，以及SCALE.AI倡议下的大规模产业合作。

The impact of COVID global pandemic has proven to be highly disruptive to the supply chain across multiple industries. For example, in the retail context, retailers experienced the impact of panic buying in multiple categories of products. In the aerospace industry, operations halts and sudden declines in the demand have generated shocks and disruptions in production and operations. Critical parts and components faced significant and unpredictable delays. The processes and operations in supply chains which are made to achieve efficiency at a low cost have become increasingly complex, and thus are often prone to highly disruptive scenarios. In such situations, the tools which rely mainly on a large amount of historical data commonly used in decision making processes in supply chains cannot generally be adapted to deal with sudden shifts in the environments they operate in. The unique aspects of this research program are its practical relevance to the highly challenging pandemic and disruption situations, and integration of operations research and machine learning techniques to enhance analytics in the context where quality and amount of data are insufficient. Our research program leverages the predictive and prescriptive analytics approaches to enhance the robustness and resilience of supply chains through two important countermeasures, i.e., (i) pre-disruption risk mitigation plans to proactively prepare the supply chain to be sufficiently robust and resilient prior to the disruptions, and (ii) post-disruption recovery actions to promptly detect, assess and react to the situations at hand. We will develop prescriptive approaches to determine risk mitigation plans at the strategic, tactical and operational levels of the supply chains, more specifically, in the areas of robust supply chain network design, resilience production and distribution planning, as well as robust transportation planning. To allow the decision maker to deal with a disruption after it occurs, we will focus our effort on three important applications, namely demand anomaly detection and online replenishment in retails, dynamic scheduling in manufacturing and production planning, and real-time vehicle dispatching and delivery. The developments of the analytics tools will be based on multi-stage stochastic and robust optimization methods, advanced decomposition methods, heuristics and reinforcement learning algorithms for the prescriptive capabilities, which are enhanced through statistical and machine learning methods used to generate adversarial (disruptive) scenarios, reduce dimensionality, and construct scenario-wise uncertainty representations (ambiguity set). The majority of this research program will be done in collaboration with our retail and industrial partners. This research program strategically complements existing research under our Canada Research Chair in Supply Chain Analytics as well as the large-scale industrial collaboration under the SCALE.AI initiative.