Engineering Initiation Award: Time-Average Markov Decision Processes: Theory Algorithms and Applications

工程启动奖：时间平均马尔可夫决策过程：理论算法和应用

• 批准号: 8707620
• 负责人: Keith Ross
• 金额: --
• 依托单位: University of Pennsylvania
• 依托单位国家: 美国
• 项目类别: Continuing grant
• 财政年份: 1987
• 资助国家: 美国
• 起止时间: 1987-09-01 至 1990-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=8707620&HistoricalAwards=false
• 关键词: Engineering; Initiation; Award; Time; Average

Engineering Initiation Abstract - Ross This research will investigate Markov Decision Processes (MDPs), particularly as they relate to telecommunications. MDPs have been proposed for a variety of fields in which decisions must be made sequentially in the presence of uncertainty: communication networks (routing, flow control, multiplexing); operations management (inventory control, machine replacement, highway maintenance, hospital admission scheduling); and statistics (sequential hypothesis testing), among others. However, industrial implementations of MDPs are surprisingly few, particularly in the telecommunication industry. The principal investigator will study a new formulation for time-average MDPs that could overcome some of the apparent reasons for the lack of their application. It should lead to simpler and more elegant proofs for the classical results, and should also be a powerful technique for obtaining new results and algorithms with a wide range of applicability. Regarding telecommunication systems, the study will address the application of time-average MDPs to a variety of queueing models, based on the researcher's previous work with optimal flow control and optimal multiplexing for integrated communication networks. It is also anticipated that the new formulation will broaden substantially the range of applicabability of MDPs to telecommunications. Another important aspect of the research is the simultaneous study of concrete applications of MDPs along with the mathematical theory of MDPs, thereby increasing synergistically the quality and quantity of the results.

工程启动摘要-罗斯 本研究将探讨马尔可夫决策过程 （MDPs），特别是与电信有关的问题。 MDPs 已经被提出用于各种领域， 必须在存在不确定性的情况下依次进行： 通信网络（路由、流量控制、复用）; 运营管理（库存控制，机器更换， 公路维修、医院入院时间表）;以及 统计（序列假设检验），等等。 然而，MDP的工业实现令人惊讶地少， 特别是在电信行业。 主要研究者将研究一种新的配方， 时间平均MDP，可以克服一些明显的 原因是没有应用。 它应该会导致 更简单和更优雅的证明经典的结果，和 也应该是获得新结果的强大技术， 具有广泛适用性的算法。 关于 电信系统，该研究将解决应用 时间平均MDP的各种建模模型，基于 研究人员以前在最佳流量控制方面工作， 用于综合通信网络的最佳复用。 它 预计新的提法也将扩大 MDP的适用范围基本上 电信。 研究的另一个重要方面是 沿着同步研究MDP的具体应用 MDP的数学理论，从而增加 协同地提高结果的质量和数量。