KDI: Synergistic and Decentralized Decision Making in Complex Stochastic Systems

该项目将使用制造环境作为一个范例来建模和研究不确定条件下的分布式和分级控制中的问题。这个项目的主要目标是调查并提出解决方案，以解决由于技术的激增而产生的计算挑战，这些技术导致了具有前所未有复杂性的随机大规模系统。第二个目标是开发解决方案和方法，不仅满足制造业供应链协调问题的需要，而且具有足够的普遍性，能够探索它们对其他科学技术领域的影响。制造企业是一个由多个高度相互关联的层次组成的企业。制造系统的规模、非线性和随机性使其变得有效，更不用说最优的集中决策了。过去的方法是将系统的状态与许多变量联系起来，例如随机性、决策向量、成本，然后在分散的决策框架中求解解耦的组件。这产生了易于处理但效率低下的决策规则，因为所做的假设允许解决问题。本研究的目标是在不影响问题的可处理性的情况下，实现分散决策实体的系统协调。这些问题涉及从随机控制和大偏差理论到离散事件动态分析、神经动态规划和并发模拟等方法论。由于许多待开发的技术涉及优化和随机系统，因此还将审查经济学和统计物理学的方法论研究的相关性。与行业的紧密联系将使在实际制造环境中的部署、实施和测试成为所开发研究方法的传播和转移计划的一部分。

9873339CaramanisThe project will use the manufacturing environment as a paradigm for modeling and studying problems in distributed and hierarchical control under uncertainty. The primary objective of this project is to investigate and propose solutions to computational challenges that are the result of the proliferation of technologies which give rise to stochastic large-scale systems of unprecedented complexity. The second objective is to develop solutions and methodologies that not only meet the needs of the manufacturing supply chain coordination problem, but are general enough to be able to explore their impact on other areas of science and technology.The manufacturing enterprise is one consisting of multiple, but highly interconnected, hierarchical layers. The size, nonlinearity, and stochasticity of manufacturing systems renders effective, let alone optimal, centralized decision making intractable. The approach followed in the past has been to relate the system's state to a number of variables, e.g., randomness, decision vector, cost, and then solve for decoupled components in a decentralized decision framework. This produces tractable, but inefficient, decision rules because of the assumptions made to allow solutions.The goal of this research is to achieve systematic coordination of decentralized decision entities without compromising tractability of the problem. The issues involved span methodologies ranging from StochasticControl and Large Deviations Theory, to Discrete Event Dynamic Analysis, Neuro-dynamic Programming, and Concurrent Simulation.Since many of the techniques to be developed involve optimization and stochastic systems, the relevance of the methodological research on economics and statistical physics will be also examined.Strong ties with industry will allow deployment, implementation, and testing in an actual manufacturing environment to be included as part of the dissemination and transfer plan for the research approaches developed.