EAGER: Integrated Simulation Models, Experiments, and Optimization

EAGER奖将开发和展示新的仿真优化算法，该算法集成了分析，实验设计和建模方法，以利用仿真系统的许多独特特性。目前的仿真实验与传统的真实系统实验几乎没有区别。他们很少或根本没有利用在计算机实验中一切都是可控的这一事实。与传统的“黑盒”实验设计不同，不同的场景应用于不同的实验单元，模拟可以在同一处理器上同时运行许多场景。其基本概念是让相互竞争的模拟系统“竞赛”，而不是像目前这样顺序运行它们（通常每个系统都在一个处理器上运行）。这项研究的基础是模拟代码是“透明的盒子”，至少对它们的创造者来说是这样。人们不仅可以看到这个盒子的内部，还可以在模拟实验运行的同时到达里面并改变它。模拟影响了现代社会的几乎所有方面，有助于优化系统设计和政策。保证模拟项目的问题通常是高影响和紧迫的。本研究之前的初步（特设）综合模拟实验/模型包括一个涉及生物制药的紧急库存，使其供应链更强大的突然中断（如地震）。这表明集成的仿真模型/实验可以比在仿真上运行的常规实验设计有效得多。这允许探索更多的选择，最终导致更多的决策制定。

This EAGER award will develop and demonstrate new simulation optimization algorithms that integrate methods of analysis, experimental design, and modeling to exploit the many unique characteristics of simulated systems. Current simulation experiments are almost indistinguishable from conventional real-system experiments. They take little or no advantage of the fact that in a computer experiment everything is controllable. Unlike conventional "blackbox" experimental designs, where different scenarios are applied to distinct experimental units, simulations can simultaneously run many scenarios on the same processor. The basic concept is to have competing simulated systems "race" rather than the current approach of running them sequentially (usually each on a single processor). The foundation for this research is that simulation codes are "clear boxes", at least to their creators. One not only can look inside this box, they can reach inside and change it while the simulation experiments are running.Simulations impact almost all aspects of modern society to help optimize system designs and policies. Problems that warrant simulation projects are typically high-impact and pressing. Preliminary (ad-hoc) integrated simulation experiment/models prior to this research include one involving emergency inventories of biopharmaceuticals to make their supply chains more robust to sudden disruptions (like earthquakes). This indicates that integrated simulation model/experiments can be considerably more efficient than conventional experimental designs run on simulations. This allows exploring many more options, ultimately leading to moreinformed decision making.