Designing and Controlling Processing Networks with Parameter Uncertainty

这项拨款为研究结合排队网络动力学和随机规划决策框架的模型提供了资金。在运筹学等相关领域，对排队网络和具有参数不确定性的模型的研究已经有很长的历史，但对同时包含这两种特征的模型进行系统分析的工作相对较少。在研究的第一部分中，分析了流体处理网络，其中流体到达率，加工率和初始库存不是完全确定的。控制器感兴趣的是最小化预期库存成本，因为必须在上述随机元素暴露之前做出某些结构操作决策。然后，这个流体模型将用于开发具有离散作业的随机处理网络的设计和控制策略，其中存在参数不确定性。在研究的最后阶段，控制策略将在半导体晶圆制造设施的全尺寸详细仿真模型中进行测试。仿真模型将与国际SEMATECH合作开发。作为这项工作教学贡献的一部分，研究结果将通过随机规划和随机处理网络的教程在随机规划社区主页上传播。如果成功，该项目将对排队理论、随机规划和制造研究界做出贡献。在队列社区中，历史上的焦点通常放在性能分析和控制上，要么假设所有系统参数都是已知的，要么假设控制器可以即时适应不断变化的参数。目前，采用系统方法将参数不确定性纳入这些模型的研究还很少。在许多随机规划模型中，目标是在更详细操作的粗粒度子模型的指导下做出战术级或战略级决策。本项目研究的排队网络是此类子模型的理想候选者。在工业环境中，该研究将允许制造业的网络设计师和控制器在面对不确定性时做出更好的决策。由于该项目包括与半导体行业的深入接触，该研究将对该行业产生最直接的影响。然而，在许多需要随机网络控制的部门，如电信和交通管理，有很大的潜力产生广泛的影响。

This grant provides funding to examine models which combine queueing networkdynamics and a stochastic programming decision framework. The study of both queueing networks and models with parameter uncertainty have a long history in operations research and related areas, but there has been relatively little work on systematically analyzing models which contain both features. In the first part of the research, a fluid processing network is analyzed in which the fluid arrival rates, processing rates, and initial inventory are not known with complete certainty. The controller is interested in minimizing expected inventory costs, given that certain structural operating decisions must be made before the aforementioned random elements are revealed. This fluid model will then be used to develop design and control policies for stochastic processingnetworks, with discrete jobs, in which there is parameter uncertainty. In the final phase of the research the control policies will be tested in full-scale, detailed simulation models of semiconductor wafer fabrication facilities. The simulation models will be developed in collaboration with International SEMATECH. As part of the pedagogical contribution of this work, the research results will be disseminated on the Stochastic Programming Community Homepage via a tutorial on stochastic programming and stochastic processing networks.If successful, this project will make contributions to the queueing theory, stochastic programming, and manufacturing research communities. In the queueing community the historical focus has usually been on performance analysis and control, under either the assumption that all system parameters are known or that a controller can instantaneously adapt to changing parameters. At present, there is little research which takes a systematic approach to incorporating parameter uncertainty into these models. In many stochastic programming models, the goal is to make tactical- or strategic-level decisions with the guidance of a coarse-grain submodel of more detailed operations. The queueing networks studied in this project are ideal candidates for such submodels. In the industry setting, the research will allow network designers and controllers in the manufacturing sector, and beyond, to make better decisions in the face of uncertainty. Since this project includes in-depth involvement with the semiconductor industry, the research will have the most immediate impact in this sector. However, there is strong potential for broad impact in many sectors, such as telecommunications and traffic management, where stochastic network control is necessary.