基于数据驱动的供应链库存鲁棒优化模型及运作策略研究

As an effective approach for dealing with uncertainty in supply chain operations, robust optimization has been drawn much attention by both academic and business. However, the traditional studies on robust optimization of supply chain inventory problems are mostly based on the assumption of the pre-defined uncertain parameters information, which limits the application of robust optimization. In this project, only the historical data of the uncertain parameters in supply chain inventory operations are assumed to be known, the data-driven method and robust optimization approach will be combined to develop some data-driven robust optimization models for supply chain inventory management which take the inventory operation performances as objective functions, various kinds of operation settings as constraints, and many operations as decision variables. Firstly, some approaches including test of goodness of fit, φ-divergency, likelihood estimation, and so on are applied by combining statistic inference theory to build uncertainty sets for unknown parameters under some certain confidence levels; Secondly, the data-driven based supply chain single period and multi-period inventory control problems will be studied; At last, concerning the serial and the general supply chain inventory systems, the robust operation problems of the supply chain inventory with considering some complex situations will be investigated. Aiming at the above models, the data-driven robust inventory strategies of the supply chain inventory will be proposed, and the factors which impact the inventory operation performances will be analyzed. The research results of this project will provide some supports for the enterprise managers to make decisions based on the exiting business data to against uncertain disturbances in operation, and then can be widely used in practice.

作为处理不确定性的有效手段，鲁棒优化引起学术界和企业界的普遍关注。传统供应链库存鲁棒优化问题研究大多基于对不确定参数信息的事先指定，限制了其应用性。本项目在仅知供应链库存运作中的不确定参数历史数据条件下，将数据驱动思想和鲁棒优化方法结合，建立以供应链库存绩效为目标、各种运作环境为约束，若干运作为决策的基于数据驱动的供应链库存鲁棒优化模型。首先，在获取供应链库存运作中的不确定参数数据基础上，结合统计推断理论，综合运用拟合度检验、φ-散度和似然估计等方法，构建满足一定置信度的关于未知参数的不确定集；其次，在鲁棒建模准则下，研究基于数据驱动的供应链单周期和多周期库存控制问题；最后，针对串行和一般供应链库存系统，研究考虑复杂情形的供应链库存鲁棒运作问题。针对所建模型，求解基于数据驱动的供应链鲁棒库存策略，分析影响库存绩效的因素。本项目研究成果将对企业依据已有数据制定应对不确定性扰动的决策提供支持。

本项目研究基于数据驱动的供应链库存鲁棒优化问题。与传统供应链库存鲁棒优化问题研究大多事先指定不确定参数信息不同，本项目在仅知供应链库存运作中的不确定参数历史数据条件下，将数据驱动思想和鲁棒优化方法结合，建立以供应链库存绩效为目标、各种运作环境为约束，若干运作为决策的基于数据驱动的供应链库存鲁棒优化模型。通过执行该项目，主要进行了如下研究工作：首先，在企业运作实践基础上对供应链运作中的不确定性进行了分析，明确供应链库存运作中的不确定性来源及属性，并从建模角度对不确定性因素进行了归类和参数化，实现了数据的收集、分析和初步处理。在此基础上，从连续和离散属性角度研究了不同鲁棒建模准则的数学描述方法；其次，在获取供应链库存运作中的不确定参数数据基础上，结合统计推断理论，综合运用拟合度检验、φ-散度、Wasserstein度量等方法，构建了满足一定置信度的关于未知参数的不确定集，以此作为供应链库存鲁棒优化模型的输入；再次，在鲁棒建模准则下，研究了基于数据驱动的供应链单周期和多周期库存控制问题。针对单周期库存控制问题，重点研究了需求以及供需不确定性下基于数据驱动的供应链库存鲁棒优化问题。针对多周期库存控制问题，重点研究了需求以及需求和提前期不确定性下基于仿射可调节鲁棒优化的供应链多周期控制问题。最后，针对供应链网络库存系统，研究了考虑复杂情形的供应链库存鲁棒运作问题。针对所建模型，运用对偶理论求解了基于数据驱动的供应链鲁棒库存策略，分析了影响供应链库存运作绩效的因素。上述研究具有重要理论意义与实践价值。在理论层面，丰富了不确定环境下的供应链库存鲁棒运作问题的研究，构建了依靠数据进行库存决策的框架和方法体系；在实践层面，有助于企业明确如何在不确定环境中对相关经营数据进行处理，从而为管理者在不确定经营环境中依据已有运营数据制定具有鲁棒性的库存策略提供有效的决策支持。