神经网络

A novel method is proposed, which is based on orthogonal least square algorithm. This method offers a systematic way for the selection of input variables.

神经网络是进行负荷预测的一种成功方法，输入变量的选择是影响神经网络预测效果的重要因素，本文提出了基于正交最小平方方法的神经网络输入变量选择方法。

Based on analyzing regression analysis, neural network and the least square support vector machines that the three soft computing methods fitting for data fitting, in the paper try to use nonlinear regression analysis, the BP neural network based on Levenberg-Marquardt training algorithm, radial basis function neural network on orthogonal least square training algorithm and least square support vector machines for tracking the position of glass plate.

本文在分析适于数据拟合的回归分析、神经网络、最小二乘支持向量机这三种"软计算"方法的基础上，尝试将非线性回归分析方法，基于Levenberg-Marquardt算法的BP神经网络，基于正交最小二乘训练算法的径向基函数神经网络和最小二乘支持向量机应用于玻璃板位的跟踪中。

The main works include five parts:(1) A modified K-means algorithm for optimizing the RBPNNs' structures was proposed;(2) The RBPNNs' structure optimization method based on the orthogonal least square algorithm was proposed that can greatly reduce the computation expense for the structure optimization;(3) Using the GA to perform the whole structure optimization of the RBPNNs was proposed in this thesis, which include simultaneous optimization of the hidden centers and the corresponding controlling parameters of the kernel functions.

主要工作体现在如下三个方面：(1)提出一种改进的 K-Means 算法；(2)提出了基于递推正交最小二乘算法的结构优化算法来训练径向基概率神经网络，从而大大减小了用于结构优化的计算开销；(3)提出使用遗传算法来实现径向基概率神经网络的全结构优化，即隐中心矢量和核函数控制参数同时优化，通过新设计的编码方式、新构造的适应度函数，充分发挥了 GA 的全局搜索性能，使得所优化的径向基概率神经网络的结构趋于最简。

Based on the relationship between matrix and symmetric matrix global exponential stability of the discrete-time neural networks model and the result of exponential convergence rate were obtained by using the characteristics of eigenvalues of a positive definite matrix and introducing a proper factor.

针对带有边界约束的凸二次规划问题，利用离散神经网络模型的建模原理，构造了一个神经网络模型。利用矩阵与对称矩阵的关系和正定矩阵特征值的性质，通过引入一个适当的因子，得到了该离散型神经网络模型是全局指数稳定性和指数收敛率的结果。同时分析了该结果的优越性和存在的不足，提出了解决的3种方法，最后给出了实例说明本方法取得结果的实用性

The multi neural networks ensemble algorithm was primary discussed and studied, A neural network ensemble algorithm used in automatic target recognition is proposed,.

5对多神经网络集成算法进行了初步的探讨研究，给出了一种用于自动目标识别的神经网络集成算法，并结合算法设计了一个基于多CPU并行结构的多目标识别神经网络系统。

In this paper, the forecast model of scouring depth is conducted by neural network and BP neural networks.

利用神经网络和一些实测数据建立BP神经网络模型，进行冲刷深度的预测，用收集到的桥墩局部冲刷数据样本训练并测试BP神经网络模型。

Self-generating neural network is a self-organization neural network, whose network structures and parameters need not to be set by users, and its learning process needs no iteration.

自生成神经网络是一类自组织神经网络，它不需要用户指定网络结构和学习参数，而且不需要迭代学习，是一类特点突出的神经网络。

On this basis, according to historical data, apply ANN and differential simulation method to get the quantitatively correlative relations between each production and its own influence factors, and introduce the new methods of prediction for dynamic indexes with gas-field development (The combinatorial prediction method based on fuzzy comprehensive evaluation, the method of ANN to select optimally combinatorial prediction models and the ANN prediction method based on genetic algorithm).(2) Base on mathematical programming, combine with quantitative economics and techno-economics, introduce economical indexes to establish production"s distribution optimal model, production"s constitution optimal model and measured production"s constitution optimal model, including multi-objective models and five-years models. Upon this, the optimal project for all gas field and each gas-collected factory can be got. Also, introduce the time value of capitals to improve on these models.(3) Base on the optimal solution theory and algorithm theory for the nonlinear programming problem, introduce the SUMT algorithm and genetic algorithm to study how to solve the models, and on the basis of normal genetic algorithm, make use of auto-adaptively modulating method to improve on normal genetic algorithm; Base on algorithm"s convergence theory and calculation"s complexity theory to analyze seriatim SUMT algorithm"s convergence and genetic algorithms convergence, and compare performance with each other.

在此基础上，利用神经网络方法和微分模拟方法根据历史数据得到各分项产量与其影响因素之间的定量关联关系，并引入气田开发动态指标新的预测方法(基于模糊综合评判的组合预测方法、神经网络优选组合预测模型预测方法以及基于遗传优化的神经网络预测方法)；(2)以数学规划为基础，结合数量经济学和技术经济学，引入经济指标建立产量分配优化模型、产量构成优化模型、措施产量构成优化模型、气田开发多目标规划模型以及五年规划模型，进而获得全气田及各采气厂的最优方案，并引入资金时间价值对五年规划模型进行改进；(3)以非线性规划问题的最优解及算法理论为基础，引入SUMT算法以及遗传算法对模型的求解进行研究，并在原有的遗传算法基础上，引入自适应调整方法对遗传算法进行改进；以算法的收敛性理论和计算复杂性理论为基础，逐一分析SUMT算法以及遗传算法的收敛性，并比较三种算法的优劣性。

Aimed at the typical defects of sample resource lacking and easy local optimal convergence, in this paper, a new way of inserting ANN into genetic algorithm is implemented, which make full use of the nonlinear solution calculation ability of ANN and the full field optimal solution seeking ability of GA. The method optimized the tube stagger spinning parameters.

针对神经网络中样本资源紧张，易于收敛于局部优解的问题，本文将神经网络模型嵌入遗传算法中，充分利用了神经网络的非线性求解能力和遗传算法的全局寻优能力，实现了对错距旋压工艺参数的智能优化。

For further study to improve performance, an adaptive control algorithm based on Neural Network is tried in stabilization control. A new method of selecting the initial weights of Neural Network is put forward, which is that the coefficients of discrete equivalent of continuous transfer function in traditional control are used as the initial weights of Neural Network. Adaptive Neural Network control is combined with traditional control by using different method in different segment.

作为进一步提高稳定系统性能的探讨，对基于神经网络的参数自适应调整控制方法在稳定控制中的应用进行了研究，提出了将传统校正方法经离散化后所得到的数字控制算法的系数作为神经网络权值初值的新方法，并采用分段控制，将传统校正方法和神经网络控制方法相结合，取得了较好的控制效果。

Singer Leona Lewis and former Led Zeppelin guitarist Jimmy Page emerged as the bus transformed into a grass-covered carnival float, and the pair combined for a rendition of "Whole Lotta Love".

歌手leona刘易斯和前率领的飞艇的吉他手吉米页出现巴士转化为基层所涵盖的嘉年华花车，和一双合并为一移交&整个lotta爱&。

This is Kate, and that's Erin.

这是凯特，那个是爱朗。