Approximate optimal control with εerror bound

近似动态规划方法求解非线性系统最优控制,需要迭代无限步才能得到最优控制律．本文提出了一种ε-近似最优控制算法,选择ε误差限,通过自适应迭代不断逼近哈密顿-雅可比-贝尔曼（HJB）方程的解,应用神经网络实现在有限步迭代后得到带ε误差限的近似最优控制律．计算机仿真结果表明了该算法的有效性．     

尺度与颜色不变性图像特征描述

尺度不变特征变换是目前公认的鲁棒性最强的图像特征描述方法之一,在尺度不变性和几何不变性方面具有较好的特性,但该方法主要适用于灰度图像,对图像颜色的区分能力不强,因此,一些对象可能会因为颜色的不同而被错误的区分.另外,尺度不变特征变换对关键点局部范围内描述子主方向的依赖性非常强,直接决定了匹配的正确率,但是研究表明,主方向分配产生的误差仅有三分之二左右能控制在[-20。,+20。]范围内,因此部分特征会有三分之一的概率因为主方向分配的误差较大而不能正确匹配.针对以上两个问题,本文提出了一种具有颜色和尺度不变性的局部特征描述方法,颜色不变性通过将RGB图像转换到高斯颜色模型下实现,特征描述过程中不再分配主方向,而用局部相对方向,尺度不变性通过构建高斯金子塔实现.实验选取阿姆斯特丹数据集图像进行了测试,结果表明本文方法比传统尺度不变特征变换方法,在特征点的数目、分布均匀性以及匹配精度方面均有所提高.     

Occlusion Culling Algorithm Using Prefetching and Adaptive Level of Detail Technique

Recent yearsin graphics area many rendering al-gorithms have been developed to support a newset ofdemanding applications.An efficient way to speed uprenderingis to avoid rendering geometry that cannotbe seen from current viewpoint,such as geometryoutside the current viewfrustumand obscured by ge-ometrythat is closer tothe current viewer.The hard-est part of the visibility culling problem is how toavoid rendering geometry whichis obscured by closergeometry.Our approach will focus on this.In thi…     

多尺度法的设解形式之探讨

当采用多尺度法研究非线性振动问题时，经常会遇到不同情形下系统响应的设解形式不同的问题．文中针对一类具有平方和立方非线性项的单自由度和多自由度非线性系统，得到不同设解形式下的一次近似解和二次近似解，并与数值解相比较，发现2种设解形式的近似解与数值仿真解的解曲线吻合很好，表明传统的各种设解形式在研究弱非线性系统时都有很好的近似性．     

基于椭圆函数求非线性系统的近似解

应用平均化方法求解了非线性自治系统x＋αx＋βx=εf（x,x），中当β不是小参数时的近似解。通过坐标变换，系统的解用Jacobian椭圆函数来摄动，而不是采用传统的三角函数摄动。最后给出一个算例，由椭圆函数积分求得了非线性微分系统的近似极限环，其结果与数值解进行比较，具有较高的精度，为求解非线性系统提供了另一种有效途径。     

自动推理技术发展的回顾与展望

介绍了国内外自动推理技术研究的历史,给出了自动推理的分类方法,阐述了各种自动推理技术的逻辑基础和基本思想,对各种推理模型的优缺点进行了系统的比较,并探讨了自动推理技术的发展趋势。     

时滞系统问题的一种近似解方法

本文对线性时滞微分方程边值函数问题提出一种级数近似解方法。其基本思想是利用正交多项式级数的近似性质把它近似地简化为一线性代数方程组的求解问题     

单晶生长中的三相界面的近似解析解

本文考虑了单晶生长中的三相界面问题，即研究了气——液新月形界面满足的Ｌａｐｌａｃｅ－Ｙｏｕｎｇ方程２ｙ＝β〔ｙ″（１＋ｙ′２）３／２－ｙ′ｘ（１＋ｙ′２）１／２〕在边界条件为ｙ（∞）＝０，ｙ′（∞）＝ｂ＜０时的解，其中β＝２σ０／ｇΔρ是正常数（Ｌａｐｌａｃｅ常数）．我们得到了新月形高ｈ应满足的关系式：ｈ＝βｓｉｎ２α０２ｃｏｓα０以及新月形轮廓线的一个近似解析解ｘ＝１－（Ｉ－１βｙ２＋ｓｉｎα０２ｒｈｙ２）２ｓｉｎα０ｒｈｙ其中α０是三相边界（ｒ，ｈ）处的切角，Ｉ＝１１＋ｂ２．     

Approximate Feedback Linearisation Using Multilayer Neural Networks

In this paper we introduce the approximate feedback linearisation using multilayer feedforward neural networks. We propose to approximate a basis of the one-dimensional codistribution of an affine nonlinear system with the derivative of a multilayer neural network [6] and form a change of coordinates with n multilayer neural networks [5]. In this paper we will prove that the transformation can define a local diffeomorphism, with which a local stabilising feedback law can be designed for a kind of non-linearisable nonlinear systems.     

Automated Meta-Control for Adaptable Real-Time Software

The software meta-controller is an online agent responsible for dynamically adapting an application's software configuration, e.g. altering operational modes and migrating tasks, to best accommodate varying runtime circumstances. In distributed real-time applications such adaptations must be carried out in a manner which maintains the schedulability of all critical tasks while maximizing some notion of system value for all other tasks. For large-scale real-time applications, considering all possible adaptations at the task-level is computationally intractable. This paper presents an automated aggregate approach to software meta-control, appropriate for large-scale distributed real-time systems. The aggregate automated meta-control problem is still NP-hard, but it has very practical approximate solutions. Introduced, here, are two very-effective approximation algorithms, QDP and GG, with very reasonable polynomial time complexity. Both algorithms also provide us with upper bounds for optimum system values, useful for deriving absolute, albeit somewhat pessimistic, measures of actual performance. Extensive Monte Carlo analysis is used to illustrate that expected performance for both algorithms is generally suboptimal by no more than a few percent. Our flexible software meta-control model is also shown to be readily applied to a wide range of time-sensitive applications.