• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>Annual Conference and Expo of the Institute of Industrial Engineers >Control of Non-Conventional Queueing Networks: A Parametric and Approximate Dynamic Programming Approach
【24h】

Control of Non-Conventional Queueing Networks: A Parametric and Approximate Dynamic Programming Approach

机译：非常规排队网络的控制：参数和近似动态规划方法

获取原文
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

In this paper, we investigate the control of non-conventional queueing networks, where multiple concurrent state transitions following non-exponential and general sojourn distributions are allowed. Two approximation schemes are discussed that produce an approximated Markovian model. We further propose to model the problem as an uncertain Markov decision processes (MDP) by considering the induced approximation error. A new simulation-based approach is investigated here, to give an overall optimal policy beyond the classic approach to such problems, e.g., a robust formulation. In particular, we approach the problem as one of finding a best overall control policy, via exploration and exploitation within a heuristic policy set. An approximate dynamic programming algorithm is then used, in connection with a parametric cost function, for efficiently learning and finding policies in this set. We show that the problem of finding the best control policy within this new policy set can be understood, equivalently, as one of finding the best set of parameters for one-stage cost function of the problem. Later, an integrated framework, denoted as extended actor-critic, is proposed to give a comprehensive treatment for those types of problems. Results of a case study are also presented and discussed.
机译：在本文中，我们研究了非常规排队网络的控制，其中允许遵循非指数分布和一般sojourn分布的多个并发状态转换。讨论了两种近似方案，它们产生了近似的马尔可夫模型。我们进一步建议通过考虑诱导近似误差，将问题建模为不确定的马尔可夫决策过程（MDP）。这里研究了一种新的基于仿真的方法，以提供一种超越经典方法的总体最优策略，例如一种健壮的公式。尤其是，我们通过在启发式策略集中进行探索和开发，将问题视为找到最佳总体控制策略之一。然后，结合参数成本函数使用近似动态规划算法，以有效地学习和查找该组中的策略。我们表明，在此新策略集中找到最佳控制策略的问题可以等同地理解为为该问题的一阶段成本函数找到最佳参数集的问题。后来，提出了一个综合框架，称为扩展的参与者评论家，以对这些类型的问题提供全面的解决方案。还介绍并讨论了案例研究的结果。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号