随机逼近

According to the study of the stochastic approximation theory, a new optimization method of yield is given.

在对随机逼近理论进行研究的基础上，提出了一种新的合格率优化算法。

Using the stochastic approximation theory, a rigorous analysis of the convergence properties of the algorithm was presented.

然后，采用随机逼近理论，严格分析了算法的收敛性。

The dissertation focuses on performance gradient estimation and SRO algorithm of DES aswell as other related topics,such as modeling of DES and discrete event simulation,performancemeasure Hessian matrix estimation,Stochastic Approximationand so on.General Semi-Markov Processis adapted to describe behaviour of DES and discrete eventsimulation.A general result is obtained and applied to analyze strong consistency of performancegradient Perturbation Analysisestimator.Conditions for effectiveness of InfinitesimalPerturbation Analysisare also discussed.New PA algorithms,such as server PA,PA usingDirac generalized function,job sequence PA etc.,are proposed to deal with discreteparameters.Furthermore,strategies of combining PA algorithms with certain iterative rules areanalyzed widely in order to optimize DES parameters and some SRO algorithms are constructed toslove DES optimization problems.

论文将DES性能分析领域中性能梯度估计和单运行仿真优化作为主要的研究对象，探讨了与性能梯度估计有关的DES和离散事件仿真过程建模、高阶导数估计以及随机逼近等理论和方法；建立了具有一般性的DES和离散事件仿真的广义半Markov过程模型；获得了分析性能梯度估计量强相合性的理论结果和无穷小摄动分析算法应用的条件；提出了诸如服务台数目的摄动分析、奇异摄动分析、串联生产线和加工车间中工件加工序的摄动分析等一系列摄动的思想和算法，对连续参数和离散参数都进行了深入的研究；对摄动分析算法同优化规则结合实现仿真优化的策略也进行了广泛的分析，构造了新型的变界截尾的随机逼近算法，并用于DES的单运行仿真优化研究；最后，给出了DES性能分析方法的实证性研究结果。

Finally, it has been proved that a step learning function, which satisfies the condition of LMS algorithm, could make network weight converge with probability 1 to the feature vector.

并证明了SOFM网络的学习存在一表征特征矢量的稳定解，最后证明了一个满足LMS随机逼近收敛算法条件的步幅函数可使网络权值以概率1收敛于特征矢量。

Adopt the arithmetic of synchronization disturbance random approach presented by Spall to train the Neural Network Controller by the output error of system.

采用Spall［1］提出的同步扰动随机逼近的算法，通过系统的输出误差对神经网络控制器进行训练。

Then, the stochastic approximation method is applied to determine the optimal control parameters.

应用扰动分析法得到性能指标对控制参数的灵敏度估计，并证明了估计的无偏性，又利用随机逼近技术优化控制参数。

1Because some conditions can not be fulfilled,one can not applythe known results in stochastic approximation directly.

1由于有些条件不能满足，现有的随机逼近结果无法直接利用，我们巧妙地克服了这一困难，在较弱的条件(文中H.4.1—H.4.3)下，给出了这类算法的严格收敛性分析(定理4.3.2，定理4.3.3)。

Then, an online optimization algorithm that combines policy gradient estimation by learning and stochastic approximation is derived.

利用此模型的动态结构特性，结合在线学习估计梯度与随机逼近改进策略，提出适应带宽配置在线优化算法。

Then by utilizing the features of this model an online optimization algorithm that combines policy gradient estimation and stochastic approximation is derived.

利用此模型的动态结构特性，结合在线学习估计梯度与随机逼近改进策略，提出动态电源管理策略的在线优化算法。

A new structure of neural network control system is introduced, its availability to accomplish feedbaek linearization and pole placement in the sense of stochastic approximation is proved, and the accuracy estimate is also discussed.

提出了一种新的神经网络控制系统的结构模型，证明了在随机逼近意义下实现反馈线性化和极点配置的有效性，讨论了逼近精度估计方法；（2）。

This is not an ordinary box, which is used as a background picture of the dialog box, very pretty.

详细说明：这可不是一个一般的对话框，它是用图片作为背景的对话框，非常好看。

Conceal me what I am,and be my aid for such disguise as haply shall become the form of my intent.

遮掩我的身份，帮助我，我的面具将成为我的目的。