小神经

By means of the projection method,adding a small disturbing function to the objective function, the penalty method, andthe gradient method, this dissertation proposes several neural networks for solvingsaddle point problems and minimum radius of ball-coverings problems, and investi-gates the stability of the proposed networks by the LaSalle invariance principle andthe Lyapunov method, so as to design neural networks which avoid getting stuck inthe local minimum.

本文针对约束鞍点问题和球覆盖最小半径的计算问题，分别利用投影法、对目标函数加上很小"扰动函数"的逼近法、罚函数法和梯度法等建立神经网络模型进行求解，并基于LaSalle不变集原理和Lyapunov直接法等工具，对模型的动态特性进行研究，从而设计出避免陷入局部极小的优化计算神经网络模型。

It has become one of the important tasks in the study on modern electric system. At first, this text has briefly introduced meanings, the current development and the main methods which the electric load forecasting are used. This paper aimed at the floatability which the load data had and the nature that neural networks model is easy to fall into part infinitesimal value and low iterative speed, this paper proposed a method based on neural networks and SVM. SVM can dispel the less summation and the larger single point errors of neural networks, and the neural networks could dispel the simplification of the model of the vector support machine.

本文首先介绍了电力系统负荷预测的意义、发展现状和主要方法；针对负荷数据具有浮动的特点以及神经网络模型算法容易陷入局部极小和迭代速度较慢的缺点，提出了一种基于神经网络和支持向量机的混合负荷预测方法，此方法能通过支持向量机消除神经网络的总和较小但单点误差较大的不利现象，而神经网络消除了支持向量机的模型简单问题。

Aimed at the characteristic of multiple types of faults possibly happened in nuclear power plant, large scale of training sample, and requirement of quick and accurate diagnosing, after the analysis of large sample number generated by large fault type exceeds limitation of 64K paragraph in DOS and large number of local minmum in error surface, measurements of same format of data file, Win95/NT operating system platform, resetting weight learning rate, dynamic training set in quick learning algorithm and improvement on quick learning algorithm using homotopy method which can avoid local minmum points in error surface have been adopted to ensure quickly and effective process of the course of neural network's training and testing.

针对核动力装置可能发生的故障种类多，训练样本规模大，故障诊断需要快速准确的特点等，分析了样本多和故障种类多产生的训练样本量超出DOS段大小和网络误差曲面上局部极小值多的情况，采取了使用相同格式数据文件、Win95/NT操作系统平台，对快速学习算法采用重置神经网络权值学习率和动态训练集、并采用能够有效克服网络误差曲面上局部极小点的同伦方法对学习算法进行改进等措施保证神经网络训练测试过程的快速、有效地进行；针对装置发生的故障须快速、准确诊断的需要，分析了故障的产生对装置参数变化的影响及操纵员对故障诊断的基础，在核动力装置发生故障时参数曲线的变化量与正常运行时参数曲线的变化量存在明显差异的基础上，提出采用参数曲线的变化量作为神经网络的输入，并围绕参数变化量的方法采用二次曲线拟合滤波求变化量和阈值技术来保证神经网络得到精确的装置参数变化量，从而得到准确的诊断结果。

A fault diagnosis based on wavelet and PNN was introduced, which firstly uses Mexican Hat wavelet to transform diagnosis signal and extract energy normalized vector, then uses PNN to output fault as a classifier.

针对煤矿主通风机故障与征兆对应关系复杂的特点，以及利用传统BP网络进行故障诊断存在训练速度慢、易陷入局部极小的缺点，提出基于小波和概率神经网络的故障诊断方法：先利用时频两域有紧支撑能力的MexicanHat小波变换故障信号并提取能量归一化故障特征向量；然后将概率神经网络作为诊断决策分类器，输出故障模式。

RESULTS The degenerative changes of nerve ending in Meissner's corpuscles were observed after 1 month of denervation, and the basic structure of the corpuscles had no obvious changes. After 3 months, the axons of corpuscles were disappeared, and the volume of corpuscles was shrunk. The basic structure of nerves was disappeared, and the lemmocyte and neurolemma plate were changed after 5 months.

结果 失神经1个月时触觉小体内的神经末梢已有溃变，触觉小体的基本结构无明显改变；3个月时轴突消失，触觉小体的体积开始萎缩；5个月时神经结构消失，膜细胞及其膜板亦开始改变；8个月时触觉小体内胶原纤维含量逐渐增多；12个月时触觉小体内膜结构与膜间基质完全消失，被胶原纤维代替。

This method consists of two steps: first, the image is divided into numbers of nonoverlapping blocks and the GMRF parameters are computed for each of these blocks, then a clustering method is used to merge these blocks, a coarse segmentation and texture types of the image are obtained. Second, a deterministic relaxation neural network is constructed, the parameters of the model estimated from the first step are used as input data, through neural network computation, a fine segmentation is obtained.

提出一种基于高斯-马尔可夫随机场和神经网络的无监督纹理分割方法，方法分为两步：第一步先将图像分为不重叠的小块，在小块中计算GMRF参数，并将此参数和该块的均值、方差作为该块的特征向量，然后进行聚类，得到原图像的一个初始分割和图像中所包含的类别数；第二步构造一个决定性松弛的神经网络，将第一步得到的结果作为初始输入，经过神经网络计算，得到一个精确的分割结果。

Further, a hemagglutination elution test revealed that the 6593S virus was unable to release from hemagglutinated RBC. Through the protein structure modeling, it was suggested that the change in amino acid residue 443 of neuraminidase might alter the neuraminidase catalytic sites.

而进一步透过蛋白质结构模拟分析，可以知道6593S小溶斑病毒神经胺酸酶第443个胺基酸残基的改变，使得小溶斑病毒神经胺酸酶酵素催化位置和唾液酸及神经胺酸酶抑制药物之间结合的键结有所改变，此结果可和实验结果相互印证。

Three kinds of arm's robust control theorys are designed: the torque control based on FNN controller, the computed torque control with sliding mode compensation and the computed torque control with WNN compensation. FNN syncretizes the reasoning ability of fuzzy control and parameters' self-learning ability of neural networks. It does not depend on the precision of the mathematical model, and can overcome the impact of the uncertainty effectively. So FNN is regarded as the manipulator controller and used to the trajectory tracking of arm control. The computed torque control with sliding mode compensation is designed.

在此基础上，研究了机械手臂的鲁棒控制法，分别设计了基于模糊神经网络的机械手臂力矩控制方法、基于滑模变结构控制补偿的机械手臂力矩控制方法以及基于小波神经网络补偿的机械手臂力矩控制方法，具体内容如下：模糊神经网络融合了模糊控制的推理能力和神经网络的参数自学习能力，它不依赖于对象精确的数学模型，能有效地克服被控对象存在的不确定部分的影响，本文把模糊神经网络作为机械手臂的关节伺服控制器，通过对网络参数的学习训练来调整机械手臂关节的控制力矩，实现对机械手臂的轨迹跟踪控制。

Before transecting inject 1ml Sodium Chloride in the epineurium. Annular tensile the proximal of epineurium with microscopic forceps annular shear the epineurium 7mm before broken ends with microscopic shear and stretch the epineurium until the two ends distance is 5mm.Use 8-0 suture line to bridge the nerve gap and make a 5 mm nerve tube at the proximal nerve end. Interrupted suture the epineurium tube and the distal nerve end make a 5 mm nerve mamuro for the injury nerve.Results: Postperative observation lasted for 6 weeks regeneration nerve through the 5 mm nerve mamuro.

在大鼠的坐骨神经距股方肌下缘15mm处，用1ml注射器向神经外膜下注入1ml生理盐水，然后于此处切断坐骨神经，用显微镊环形拉伸近端神经外膜，在距断端约7mm处用显微剪环形剪开神经外膜，并继续拉伸外膜使外膜两断端间距约5mm停止，用8-0的无损伤缝线桥接神经外膜两端，在神经近端形成约5mm的神经外膜管，将神经外膜管与神经远端间断缝合，使断伤的神经束间留有5mm的小间室。

Positive nerve fibers are in pedal nerve cords and visceral nerve cords. There are no neuronal nitric oxide synthase or inducible nitric oxide synthase positive cells or fibers in the nervous system. A few of endothelial nitric oxide synthase strongly positive nerve cells are identified in pedal ganglia and visceral ganglia. Weakly positive cells and nerve fibers of eNOS can be seen in all of the ganglia and nerve cords.

免疫组化定位表明，神经型一氧化氮合酶和诱导型一氧化氮合酶在整个神经系统内均呈阴性；足神经节和脏神经节内有少量神经细胞呈内皮型一氧化氮合酶强阳性；各神经节和神经索内的部分小细胞和神经纤维呈eNOS弱阳性。

The split between the two groups can hardly be papered over.

这两个团体间的分歧难以掩饰。

This approach not only encourages a greater number of responses, but minimizes the likelihood of stale groupthink.

这种做法不仅鼓励了更多的反应，而且减少跟风的可能性。