神经网络

The operating principles of the monolayer neural network, Adeline network and B-P network in the system identification are introduced in this paper.

主要研究了单层神经网络、Adaline网络及B-P网络用于系统辨识中的工作原理，提出了把B-P网络权值转换为传递函数的方法。

That is that information processing of neural networks is viewed as the composition of the two steps: field organization and field action, in which, field organization transformation converts input pattern into relevant structure representation, extracting the underlying architecture features and forms a architecture representation in action transformation space; The field action transformation converts the features representation into expected output. In strategies, we use the dual affine coordinates in dual flat manifold as way of representation of field organization system and field action system, which established a theoretical framework of analysis the coupling mechanism of information processing from the view of point of structure representation and model generation.

在具体人工神经网络结构描述的策略上，利用对偶流形空间上的对偶仿射坐标概念分别表示场组织变换系统和场反应变换系统，建立起从问题的结构表示和模型生成两种观点来分析信息处理的理论框架；分析了有监督和非监督学习人工神经网络模型在神经场变换机理下的形式化描述。

Because of appearing empty aggregates in the time zone (the time zone do not include the information), we only establish the indeterminate grey differential equation according to the approximate condition of the differential equation.

利用灰色微分方程中的最小二乘估计参数：进而得到灰色神经网络初始的权值及阈值，这样使初始的权值和阈值更接近准确值，克服了以往BP网络中初始权值的随机性和不确定性，减少了网络的训练次数，提高了网络收敛速度和输出的精度，为避免网络训练陷入局部极小值也起到了重要的作用。

This paper makes use of improved BP algorithm compared to the traditional one and the new algorithm brings us higher speed in constringency and more preciseion in detection.

BP学习算法是神经网络中的一种重要的方法，反向传播网络是典型的前馈型网络，结构上它属于多层网络，分为输入层、隐层和输出层

In accordance with discrete, non-symmetry and time-delay feedback neural networks, we provide some theoretical criteria to judge the stability for designing the networks. On continuous feedback networks, we built an algorithm to estimate the convergence region. On the HNN feedback networks, an effective optimization algorithm based on the annealing strategy and the chaotic neural network is presented, and some disciplinary conclusions are provided to guide the selection of parameters. Simulation results on benchmark problems demonstrate that the performances of our algorithm are distinctly superior to those of the classic HNN based on gradient descend and stochastic optimization methods based on probabilistic distribution.

针对离散、非对称和时延等三种类型的动态反馈网络，给出了可用于网络设计的稳定性判别准则；针对连续动态反馈网络，建立了其收敛域估计的一个算法；针对以HNN模型描述的动态反馈网络，基于退火策略和混沌神经网络提出了一种高效的优化算法，建立了可用于指导参数选择的一些规律性结论，对典型算例的仿真研究显示，改进算法在各类性能上相对基于梯度下降的传统HNN模型和基于概率分布的随机优化算法，具有明显的优越性。

Bring forward a kind of learning algorithm fitting on the base of spline function, because the spline function has good flexibility and quadratic smoothing property, the search problem of network weight functions training can be converted into the resoluting extreme value problem of the plurality function through using spline function to represent network input and link weight functions, thus the PNN can be trained with the help of the existing neural network learning algorithms;④build a kind of learning algorithm base on the alternate of Walsh function, the complexity of network computing can be decreased through using the complete orthogonality of Walsh function corollary.

应用结果表明该算法收敛速度快，稳定性好；③提出了一种基于样条函数拟和的学习算法，由于样条函数具有很好的柔韧性和二次光滑性，将网络输入函数和连接权函数用样条函数的形式表示，可把网络权函数训练的函数寻优问题转换为多元函数求极值的问题，从而可借助于现有的神经网络学习算法训练过程神经元网络；④建立了一种基于Walsh函数变换的学习算法，利用Walsh函数系的完备正交性，可大大降低网络计算的复杂度。

For two examples, result testifies that it is feasible to use in nonlinear restriction optimization. Chapter 3 is important part in the paper. Analyze and research BP network, concealed node number of network former frame newly. Give new supplement and advice about How to select study velocity and activation function in BP algorithm.

第3章是论文中最重要的一章，通过实例分析研究了BP网络，对网络模型结构的隐层节点数进行分析和新的探讨；对BP算法中的学习速率和激活函数的选取进行了新的补充和建议；对网络的拓扑结构进行了新的构造；对人工神经网络收敛依据提出了新的判断原则。

A network with different topologic configurations is needed for flatness pattern recognition when strip width is changing, in this case the large learn assignment, slow convergence and local minimal in the network are observed, in order to solve these problems, first, a flatness pattern recognition GA BP network model with only 6 input and 3 output signals has been established based on Legendre orthodoxy polynomial, which is not only simple in structure but also is of clear physical meaning.

针对板宽变化时需要不同拓扑结构的神经网络才能完成板形模式识别任务，网络学习工作量大，网络存在收敛速度慢，易陷入局部极小等结构性能不佳的问题，首次建立了以勒让德正交多项式为基模式的只用 6个输入信号、3个输出信号的板形模式识别 GA- BP网络模型。

For existing problems in robot planning, disadvantages and difficulties in self-organization NN applied to robot track planning, and the features in self-organization rival network, a theory model of the TSPSONN is provided in order to resolve some problems producing easily fuzzy or uncertainness of track reconstruction Consequently, theory evidence and applied basis are offered for complicated track control of telerobot control system.

针对机器人轨迹规划中存在的问题，以及目前用于机器人轨迹规划的自组织神经网络存在的缺陷和困难，并结合自组织竞争网络的特点，提出一个自组织神经元网络时序处理算法的理论模型，用来解决具有重复／共享状态易于产生轨迹重构时的模糊性或不确定性问题，从而为网络远程机器人控制系统进行复杂的轨迹控制提供了理论依据和应用基础。

This paper utilizes the nonlinear mapped competence and consilient approach to faculative function of Artificial Neural Networks and discusses the application of Back Propagation Networks model in integrated bid evaluation in order to set up a scientific bid evaluation method based on impersonality and experts knowledge. With the unceasing development of computer network technology, this method has a bright perspective of being conducted on the computer network.

本文利用人工神经网络的非线性映射能力以及对任意函数的一致逼近性能，对BP (Back Propagation，多层前向网络)网络模型在综合评标中的应用进行了探讨，目的是建立一种科学的基于客观和专家知识的评标方法；并且随着计算机网络信息技术的不断发展，可将该方法拓展为网上在线评标。

We have no common name for a mime of Sophron or Xenarchus and a Socratic Conversation; and we should still be without one even if the imitation in the two instances were in trimeters or elegiacs or some other kind of verse--though it is the way with people to tack on 'poet' to the name of a metre, and talk of elegiac-poets and epic-poets, thinking that they call them poets not by reason of the imitative nature of their work, but indiscriminately by reason of the metre they write in.

索夫农 、森那库斯和苏格拉底式的对话采用的模仿没有一个公共的名称；三音步诗、挽歌体或其他类型的诗的模仿也没有——人们把&诗人&这一名词和格律名称结合到一起，称之为挽歌体诗人或者史诗诗人，他们被称为诗人，似乎只是因为遵守格律写作，而非他们作品的模仿本质。

The relationship between communicative competence and grammar teaching should be that of the ends and the means.

交际能力和语法的关系应该是目标与途径的关系。