神经式网络

A kind of on-line gradient learning algorithms based on signal flow graphs isproposed for neural networks on-line training. The signal flow graph and adjoint one of aneural network are used to compute the its gradients.

对于动态神经网络在线学习而言，该算法避免了传统梯度计算中复杂的链式求导过程，利用神经网络的信号流图及其线性的伴随流图，可以直接计算任何变量的梯度信息。

Based on the principle of rank consistency, it presents the ordering chain comparison method for determining weight vector and memberships.

在排序一致性原理的基础上，提出了确定目标权重向量与指标隶属度的有序链式比较法，进而结合神经网络BP学习算法的思想与模糊优选模型，建立了基于神经网络权重调节模型的多属性决策问题的模糊综合决策方法，使得权重的确定较为客观、准确，具有自学习功能。

A new knowledge acquirement way is presented based on the reliability data of product, then the fuzzy semantic net which consists of FEPM and fuzzy fault tree is regard as fault knowledge representation. The formalization of the fuzzy semantic net is described, and the relative concepts and relations are summarized. A stereo-knowledge representation which consisted of FFTF, rule and fuzzy neural network which make up of the shallow knowledge of the product is presented. The integrated fault diagnosis method based on the stereo-knowledge representation is suggested. The method is as follows. The fuzzy search inference is carried out on FFTF firstly. If the node being diagnosing can be decided if it failed or not in the fuzzy search inference, the rule inference to the node is carried out with the rules which being in symptom-fault way. An equivalent method, pattern recognization inference using fuzzy neural network, can fulfill the same work.

提出了以FFTF模糊语义网络为主、基于浅知识的产生式规则、模糊神经网络为辅的产品故障知识的立体知识表示方式，并提出了基于该多种知识表示的综合性故障诊断方法，即：首先对故障产品进行基于FFTF模糊语义网络的模糊搜索推理，如果在模糊搜索推理过程中不能确定当前诊断节点故障与否，则对当前诊断节点进行基于征兆-故障形式的验证式规则推理或者对包含该诊断节点的部件／子系统进行基于模糊神经网络的模式识别推理，如果在完成或者不再进行基于模糊语义网络的故障推理，仍然不能诊断出故障节点，则对基于模糊语义网络诊断过程中形成的故障诊断子图进行神经网络推理。

RBF network-based ball mill expert system theory and basic structure are discussed, and the author proposes its knowledge representation, knowledge acquisition, self-adapting training and parallel inference rule.

在Matlab环境下，充分利用Matlab强大的数值计算功能实现了球磨机神经网络专家系统开放式和模块式编程，建立了RBF网络和BP网络各种不同条件的训练程序库、神经网络推理程序库、神经网络权值和阈值库等各类知识库和数据库，用户通过特别设计窗口进行咨询。

On the basis of the data obtained from Cam Plastometer,an RBF neural network prediction model is established by Matlab neural network toolbox,which corresponds with the relationship between flow stress and chemistry elements,temperature,deformation strain and strain rate of carbon steel.

以凸轮式高速形变试验机得到的实验数据为基础，利用Matlab人工神经网络工具箱，建立了碳钢的变形抗力与其化学成分、变形温度、变形程度及变形速度对应关系的RBF神经网络预测模型。

Aiming at the disadvantage of conventional BP neural network, such as selecting parameter values by the empirical method, slow convergence speed and easy trap into local minimum points, this paper designs an improved BP neural network system. In order to improve the network convergence rate and reduce the training error, this paper optimizes the initial connecting weight value of neural network by the use of ant colony algorithm and trains artificial neural network by Levenberg-Marquardt algorithm.

针对常规BP神经网络参数的经验式取值方法以及收敛速度慢，容易陷入局部最小点等缺陷，设计了一种改进的神经网络系统，利用蚁群算法优化神经网络连接权初值，并采用LM算法对人工神经网络进行训练，提高了网络的收敛速度，降低了训练误差。

Neural networks based on local sequence and amino acid component with the best structures and linear equation are constructed. The results of testing to the networks indicate that the forming of cis-proline has distinct relationships with local sequence and amino acid component. A new method for peptidyl prolyl cis/trans conformation prediction based on the theory of support vector machines was introduced. The SVM represents a new approach to supervised pattern classification and has been successfully applied to a wide range of pattern recognition problems.

通过运用神经网络与线性拟合的方法，揭示了脯氨酸局部序列、肽链氨基酸组分对于顺式脯氨酸形成与否的影响，构造了基于局部序列和基于氨基酸组分的最优结构的神经网络，以及线性拟合方程，同时对顺式脯氨酸进行了筛选，结果表明，利用神经网络的方法可以提高顺式脯氨酸筛选时的信噪比，脯氨酸采取顺式还是反式构象与其局部的序列以及整条肽链的氨基酸组分均有较强的关联。

The performance of synergetic neural network is discussed.

摘 要：本文对一类全新的神经网络——协同式神经网络的性能进行了研究。

These smartened up networks were called neural networks.

这些变机灵了的网络被称为神经式网络。

Main points of the thesis are as follows:(1) The main function models and information relations between these models in the CAPP system in the integrated CAD/CAM project are explained, a new process plan design method is mentioned based on analyses of process plan design. This method is based on the process plan prototype, and develops the prototype little by little until the perfect process plan is completed, the prospect of this method is discussed;(2) The feature abstracting rules and methods are studied, a feature coding system is developed based on GT, all the defined feature models are described by the mathematical methods;(3) The integrated method of CAD/CAPP based on features is mentioned, this method defines a feature exchanging model, and develops an interface software to exchange design data into technology data base, this technology data base includes the design and technology information of the parts;(4) The process plan reasoning method based on features is created, this method uses Mycin algorithm to solve the reasoning of the feature process plan. A process plan making system is developed. In this system part process plan is made by the semi-intelligent method, this method uses the feature reasoning, feature process chain searching and man-computer talking together;(5) The main structure of the HOPE system is introduced, a shaft drawing and its process plan made by this system are shown, the data process problem in process drawing making is also discussed;(6) A new method of describing process rule is introduced, this method uses IfThen rules and neural nets weight values together to describe process rules, the BP algorithm is adopted in this method;(7) The optimal machining sequence problem of the process steps on the same fixturing is discussed, optimal algorithms SA is applied to solve this problem, the result shows that this method is more efficient than the traditional method;(8) All jobs in the thesis are summarized. In order to carry out the research in the future, some rational proposals are given.

全文论述的内容主要有下列几个方面：(1)较为系统地阐述了CAD／CAM一体化工程项目中CAPP系统的主要功能模块及其信息流程，在分析工艺设计过程的基础上，提出了基于特征原型的渐进式工艺设计方法，并探讨了该方法的应用前景；(2)研究了零件特征提取的原则和方法，在成组技术的基础上，开发了一套基于特征的分类编码系统，并在提取零件特征的基础上，定义了零件特征的数据模型，开发了基于轴类零件特征的参数化设计系统；(3)研究了基于特征的CAD／CAPP集成的方法，通过定义零件特征数据模型的存储结构，开发了将设计数据转换到工艺数据库中的接口程序，为后续的CAPP系统提供必要的零件形状和工艺信息；(4)研究了基于特征的工艺推理方法，将Mycin算法应用到工艺决策中，开发了基于特征的工艺推理及工艺方案生成系统，在系统中综合运用特征推理、特征加工工艺链查询及人机对话相结合的半智能化方法确定零件加工工艺；(5)介绍了整个HOPE系统的总体结构，通过运行HOPE系统绘出了一轴类零件的零件图，并生成了该零件的加工工艺方案，初步探讨了基于特征的工序图生成技术中数据处理问题；(6)研究了工艺设计过程中知识如何有效表示的问题，提出了将显式的IF-THEN规则表示与隐式神经网络权值表示相结合的工艺知识表示方法，并利用BP算法进行了工艺知识表示的初步研究；(7)研究了工艺设计过程中同一安装下工步优化排序问题，利用神经网络中基于模拟退火的理论来解决该工步优化排序问题，实践证明该方法较传统的算法具有更高的效率；(8)总结了本文所进行的研究工作，为今后进一步开展工作提出了较为合理的建议。

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。