如今的商业环境瞬息万变，日常的商业运作正频繁地被恶劣天气甚至自然灾 害所中断，而诸如定制化生产等先进技术也刺激了更加快速变化的需求。因此，企 业需要新的供应链网络设计方法来同时应对供给与需求的不确定性。得益于信息 技术的发展，供应链运营所产生的数据正在不断累积，其数量与复杂度都迅速增 长，这也为供应链管理带来了新的机遇与挑战。本论文旨在探索从数据中提取有 效信息的方法，为供应链网络的设计提供决策支持，并为实际运营管理提供指导 与启示。具体地，我们利用不同来源且可能包含不同信息量和规律的数据，分析 来自供应链不同环节、具有不同目标与决策层次的实际问题。 首先，本论文研究一个全渠道零售商如何进行实体零售和服务设施的选址、选 品与库存的联合决策问题。面对供应链下游的顾客，零售商可以收集到大量高维 度的交易数据和商品与用户信息数据，并据此进行选址以及诸如选品和库存等运 营层面的决策。我们采用一般的混合多项选择模型从数据中估计顾客对商品的偏 好，基于此建立联合选址、选品与库存优化模型，并提出高效求解方法。进一步， 我们将该模型应用于实际的数据，来研究实际的全渠道零售运营问题并给出在运 营管理方面的指导与启示。 接着，本论文研究一个运输商在考虑到系统中可能存在随机发生的中断事件 时，如何在计划层面设计可靠的具有轴辐结构的运输网络。我们利用网络中各节 点上发生中断事件的历史记录数据，融合数据分析与组合优化方法建立数据驱动 的可靠轴辐网络优化模型。我们通过挖掘模型的结构性质，提出高效的约束生成 算法与启发式算法来进行求解。我们进一步利用实际数据验证了模型与算法的有 效性，并给出了可靠轴辐网络设计的规律和准则。 最后，本论文研究如何利用有限的边际或辅助信息来进行战略层的柔性供应 链设计问题，以应对可能同时出现的供给与需求不确定性。我们采用基于鲁棒分 析的决策模型，针对最不利情况下的供给与需求分布设计足够稀疏且可以满足预 设服务水平的柔性网络结构。我们创新性的提出拓展概率扩张图的概念，给出这 种柔性结构的高效生成算法，并证明该种结构的渐进最优稀疏性。通过数值实验， 我们分析了所提出柔性结构与几种经典柔性设计方法相比既在可靠性与应对极端 需求方面所具有的优势以及更广泛的适用性。
Businesses today are operating in increasingly turbulent environments, with natural disasters and social disorders disrupting normal operations on a more frequent basis. Technological advancements such as mass customization manufacturing also stimulate more volatile demand. As a result, firms need new solutions for more reliable and flexible supply chain networks to cope with both availability uncertainties and demand uncertainties. Due to the rapid development of information technology, the data collected from the supply chain operations, with fast growth in both size and complexity, has provided new opportunities as well as challenges for this issue. This dissertation aims to develop new methodologies to extract information from data and provide effective solutions and managerial insights for the design of supply chain networks. We analyze real applications with different level of objectives that cover different sectors of supply chains, by using data with several degrees of information richness. This dissertation first considers an integrated location, assortment and inventory planning problem faced by an omni-channel retailer, whose offline operational decisions are based on big and high-dimensional transaction data as well as information of items and customers. We use the powerful Mixed Multinomial Logit (MMNL) choice model to capture customer preferences, and develop an integrated location, assortment and inventory optimization model under the MMNL choice model. We derive an efficient solution approach, further obtain solutions using real data, and investigate the implications for the operations in omni-channel retail. The second part of this dissertation studies the reliable hub location problem under random network disruptions. We use the historical data of network disruptions, and develop a reliable hub location model based on data-driven methods and combinatorial optimization. By exploiting structural properties of the model, we propose a mixed-integer linear programming reformulation as well as a constraint generation approach for exact solutions and several heuristics for approximate solutions. Furthermore, we numerically test the efficacy of the proposed model and solution approaches, and provide insights into the design of reliable hub-and-spoke networks. The last part of this dissertation focuses on the reliable flexibility design problem based on marginal or side information of supply and demand uncertainties. We propose a flexibility design model, aiming to determine sparse flexibility structures that satisfy the given service level under the worst-case distribution of supply and demand. We propose a novel concept of flexibility structures, i.e., the extended probabilistic expander, provide an efficient algorithm for structure generation, and prove the asymptotically optimal sparsity of the proposed structure. Numerical results demonstrate that our design has not only a wide range of applications, but also better performance than a variety of well-known flexibility structures.
