Research Initiation: Unconstrained Optimization in DiscreteEvent Simulation

在许多现实世界的系统中，它们是如此复杂，以至于系统参数与其性能之间的函数关系没有得到很好的理解。仿真通常是唯一可以用来评估系统性能的工具。标准的数学规划技术不能用来优化系统的性能。这个项目将进行研究，旨在开发关于连续决策变量的随机系统性能优化的有效方法。该项目涉及三个问题：1)几种新的随机逼近算法在什么条件下收敛，2)这些条件在模拟中什么时候满足，以及3)随机逼近算法在经验上如何比较？

In many real world systems, they are of such complexity that the functional relationship between the parameters of the system and its performance is not well understood. Simulation is often the only tool that can be used to evaluate the performance of the system. Standard mathematical programming techniques cannot be used to optimize the performance of the system. This project will conduct research aimed at developing efficient methods for optimizing the performance of stochastic system with respect to continuous decision variables. The project is concerned with three issues: 1) under what conditions do several new stochastic approximation algorithms converge, 2) when are these conditions satisfied in simulation, and 3) how do stochastic approximation algorithms compare empirically?