Multiple Comparisons for Optimization via Simulation

研究将集中在衍生的统计有效的多重比较程序，优化通过随机模拟。当目标是最大化或最小化相对于离散的有限决策空间的预期系统性能时，结果将适用。在现有预算或技术的限制下，由于设备、时间表或设施的组合而产生了一些可供选择的系统设计时，就会遇到这种问题。在这类设计问题中，多重比较程序是进行优化的自然方法。两个主要的努力集中在多重比较的基本方法上：(1)多重比较推理，从而能够区分预期系统性能的微小差异；(2)导出稳态模拟实验的多重比较程序。这项研究的结果将有助于使蒙特卡罗模拟这一分析复杂系统的最重要工具更加有效和有用。

The research will focus on derivations of statistically efficient multiple-comparison procedures for optimization via stochastic simulation. The results will be applicable when the objective is to maximize or minimize expected system performance with respect to a discrete, finite decision space. Such problems are encountered when a number of alternative system designs arise from combinations of equipment, schedules or facilities, subject to constraints on the available budget or technology. Multiple- comparison procedures are the natural approach for optimization in such design problems. Two major efforts concentrate on basic methods for multiple comparisons: (1) multiple-comparison inference, and thus be able to distinguish small differences in expected system performance and (2) derivations of multiple- comparison procedures for steady-state simulation experiements. Results of this research should aid in making Monte Carlo simulation, a most important tool for analysis of complex systems, more efficient and useful.