Consistent Stochastic Inventory Routing Management (COSIMA)

该项目集成了供应链中通常独立或顺序调查的两个核心物流决策：库存管理和运输路线优化。虽然这种整合并不新鲜，在文献中被称为库存路径问题，但对于不确定性下的现实问题，很少有方法决策支持。因此，我们建议扩展关于几个重要的现实世界特征的现有知识，并为集成开发新的方法。零售商的随机需求通常是非平稳且相关的，允许紧急发货，并且必须满足服务水平要求。此外，我们还结合了一致性，即某一零售商的交货应该总是在相同的时间间隔内进行。这导致了一个两阶段的多级库存问题，包括路线、随机交货时间和非等距评审周期。从两个单独领域的分析开始，包括上述现实世界的问题特征，我们将各自其他领域的决策作为给定的参数，并开发和研究两个领域集成的新的预期和迭代方案。为了进行基准测试，我们使用随机动态规划方法开发了精确的解决方案。为了解决现实世界规模的问题，我们将开发并特别集成来自库存和路由领域的启发式方法。不确定性将使用情景和基于抽样的方法进行建模和表示。启发式解决方法将建立在近似动态规划、基于模板的自适应大邻域搜索和分支与后悔的基础上。所得到的方法将使用人工问题实例和真实世界的测试用例进行验证。

The project integrates two core logistics decisions in supply chains typically investigated independently or sequentially: inventory management and transportation routing optimization. Although this integration is not new and known as the Inventory Routing Problem in the literature, there is little methodological decision support for real-world problems under uncertainty. We therefore propose to extend the existing knowledge with regard to several important real-world characteristics and develop new approaches for the integration.Stochastic demands at the retailers are typically non-stationary and correlated, emergency shipments are allowed and service level requirements have to be met. Moreover, we incorporate consistency, i.e. delivery at a certain retailer should always take place within the same time interval. This leads to a two-stage multi-echelon inventory problem with routing, stochastic lead times and non-equidistant review periods.Starting from an analysis of both individual domains and including the above real-world problem characteristics, we take the decisions of the respective other domain as given parameters and develop and investigate new anticipatory and iterative schemes for the integration of both domains.For benchmarking purposes, we develop exact solutions using stochastic dynamic programming approaches. For the solution of real-world sized problems, we will develop and in particular integrate heuristic approaches from the inventory and routing domains. Uncertainty will be modeled and represented using scenarios and sampling-based methods. Heuristic solution methods will build on approximate dynamic programming, template-based adaptive large neighborhood search, and branch & regret. The resulting approaches will be validated using artificial problem instances and real-world test cases.