网络函数

Department of Automation, University of Science and Technology of China, HeFei 230027）Abstract: This paper presents a method of neural network for interactive-force manipulator on the base of kinematics and analyses of the explosive force of human arm mode and neural network theory such as function approach, optimizing calculation. The method optimizes the whole servo-system, and ensures the system's robust.

中文摘要：本文根据关节机械臂的运动特性，在分析用于测试人体手臂运动爆发力机械臂模型设计的基础上，利用人工神经网络的函数逼近、最优化计算等方面理论，结合实际人－机力交互系统的控制特点，提出一种神经网络控制方法，对基于力控制策略的机械手伺服系统进行优化，同时保证了系统的鲁棒性。

In neural networks learning, the Sigmoid function is improved and the stability conditions of backpropagation algorithm is analyzed theoretically.

在神经网络学习中，改进了前向神经网络中Sigmoid作用函数表达式，并在理论上分析了反向传播修正权值的迭代算法稳定收敛的充分条件。

Feedforward neural networks of FEL control strategy was designed, using ADALINE neural networks, normalized LMS, plus characteristics of sinusoidal function.

利用ADALINE神经网络、归一化LMS算法、结合正弦函数的特点，设计FEL控制策略中的前馈神经网络。

The CCIM is used to evolve the structure and parameters of neural network. The proposed CCIN model has been implemented based on the actual data collected from the Eastern Slovakian Electricity Corporation and compared with the traditional RBF network method.

对斯洛伐克东部电力公司进行的电力系统短期负荷预测表明，与传统的径向基函数神经网络预测方法相比，协同进化免疫网络方法具有更高的预测精度。

Combined with the characteristic that biomass can vary in a wide range, the error function of the traditional BP network is improved. Meanwhile, optimal stopping rule is used to avoid over-fitting.

结合菌体浓度变化范围大这一特点，将传统BP网络的误差函数进行了改进，并利用最优停止法对网络进行训练，避免了过拟合现象。

For speeding up BP algorithm of learning of neural network, a BP algorithm based on the Taylor's series of the error function is studied, the method is effective in speeding up BP network training.

1在本文中，针对加速神经网络学习的BP算法进行研究，给出了基于目标函数的Taylor展开的BP算法。它对加快神经元网络学习速度效果明显。

The properties and expression of knowledge are discussed in brief and binary and multiple truth expression and inference of knowledge are studied focally. The neural networks which can realized the three valued logic are overviewed. The new idea of uncertain inference based on three valued logic is proposed for the first time The models of neural logic, which can realize three valued logic and uncertain three valued logic inference are proposed, too. The learning algorithm of coefficient of MVL's function based on the canonical realization of MVL network is investigated for the first time and the appropriate algorithm is proposed.

4简要地论述了知识的特征及其表现形式，重点研究了知识的真值二值和多值表示法以及基于知识真值表示法的逻辑推理；通过对目前可实现三值逻辑推理的逻辑神经元及其网络的简要回顾，指出了各自的局限性，首次提出了基于三值逻辑的不定性推理的新概念以及可实现三值逻辑和三值逻辑不定性推理的逻辑神经元模型，同时对基于这种逻辑神经元模型的推理系统的应用领域进行了探索性尝试；另外，首次对基于多值逻辑网络正规实现的多值逻辑函数的系数参数的学习算法进行了研究，并提出了相应的学习算法。

Based on percolation theory , truncated Weibull distribution is used as the pore throat distribution function, a two phase and 3 dimensional quasi-static network model is established to simulate the primary oil displacement and secondary water flooding of reservoir cores.

结合逾渗理论，采用截断威布尔分布作为孔喉特征分布函数，模拟储层岩心的初次油驱过程和二次水驱过程，建立了油水两相三维准静态孔隙网络模型；通过模型计算与试验结果的对比，验证了孔隙网络模型的有效性。

In a transportation network supported by the Advanced Traveler Information Systems,the drivers equipped with ATIS are directed by a control center to select their routes so as to minimize the total network travel time,while the unequipped ones make the route choice decisions on the base of minimizing their individual travel times.

考虑一个被路径诱导系统支持的交通网络，装有ATIS的出行者接受交通中心的指挥、按照系统最优的原则选择路径，没有该装置的出行者根据个人成本最小的原则择路，建立了刻画这种混合交通行为的变分不等式模型，推导了混合交通行为网络的效率损失上界，并以多项式路段成本函数为例给出了一些上界的特殊值。

Fault diagnosis architecture based on adaptive neural networks for a class of unknown nonlinear systems with unmeasured states is proposed.

针对一类模型未知及状态不可测的非线性系统，提出了基于自适应神经网络的故障诊断策略，不仅在线估计神经网络的矩阵权重，而且在线估计高斯函数的宽度和中心。

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。