基于差分灵敏度信息的排队系统性能优化

项目名称： 基于差分灵敏度信息的排队系统性能优化

项目编号： No.61203039

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 自动化学科

项目作者： 夏俐

作者单位： 清华大学

项目金额： 26万元

中文摘要： 在实际工程应用中，基于排队模型的系统性能分析与优化往往存在如下困难：对于结构复杂的排队网络，传统的排队理论无法给出闭合形式的解析解，难以直接进行分析优化；另一方面，由于问题结构和参数的内在约束，这类问题往往无法建模成为标准的马氏决策过程（MDP），使得MDP优化理论无法直接运用。本项目拟从新的研究角度出发，即基于差分灵敏度信息的摄动分析，研究排队系统中的经典控制与优化问题，包括服务率控制问题和准入控制问题等，给出快速有效的优化方法，并研究优化算法的在线化实现及其与学习类算法的结合。该方法的优点在于，相对于传统的梯度优化方法，差分信息能够提供更多的性能灵敏度信息，基于差分灵敏度信息的优化方法能够充分发掘排队系统的结构特点，为解决此类排队系统优化问题提供了一个全新的研究思路。该项目的研究成果将推进排队系统优化理论在实际工程系统中的实施与应用。

中文关键词： 排队系统；随机优化；马氏决策过程；灵敏度优化；离散事件动态系统

英文摘要： The performance analysis and optimization method of queueing models usually has the following difficulties in the practical engineering applications. First, for the queueing networks with complicated structure, traditional queueing theory cannot give a closed-form solution for the system performance. Thus, we cannot analyze and optimize the system performance directly based on formulas. Second, because of the constraints of problem structures and parameters, such type of problems usually cannot be modeled as a standard Markov decision process (MDP). Thus, we cannot directly apply the MDP optimization theory to solve these problems. The goal of this proposal is to study some classical control and optimization problems of queueing systems from a new research perspective, i.e., perturbation analysis based on performance difference sensitivity information. The targeted problems include service rate control problem and admission control problem in queueing systems. We plan to propose an effective and efficient optimization method for these problems. We will further study the online implementation of optimization algorithms and the combination with learning algorithms. Our method has the following advantages. Compared with the traditional gradient-based optimization methods, the performance difference can provide more

英文关键词： queueing system；stochastic optimization；Markov decision process；sensitivity-based optimization；discrete event dynamic system