基于神经网络的微分对策控制器设计

采用伴随－BP技术，将微分对策的两点边值求解问题转化两个神经网络的学习问题，训练后的两个神经网络分别作为对策双方的最优控制器在线使用，避免了直接求解复杂的两点边值问题，对追逃微分对策问题的仿真结果表明，该方法对初始条件和噪声具有较好的鲁棒性。     

Stochastic neural networks

The first purpose of this paper is to present a class of algorithms for finding the global minimum of a continuous-variable function defined on a hypercube. These algorithms, based on both diffusion processes and simulated annealing, are implementable as analog integrated circuits. Such circuits can be viewed as generalizations of neural networks of the Hopfield type, and are called diffusion machines.Our second objective is to show that learning in these networks can be achieved by a set of three interconnected diffusion machines: one that learns, one to model the desired behavior, and one to compute the weight changes.This research was supported in part by U.S. Army Research Office Grant DAAL03-89-K-0128.     

Multi-class pattern classification using neural networks

Multi-class pattern classification has many applications including text document classification, speech recognition, object recognition, etc. Multi-class pattern classification using neural networks is not a trivial extension from two-class neural networks. This paper presents a comprehensive and competitive study in multi-class neural learning with focuses on issues including neural network architecture, encoding schemes, training methodology and training time complexity. Our study includes multi-class pattern classification using either a system of multiple neural networks or a single neural network, and modeling pattern classes using one-against-all, one-against-one, one-against-higher-order, and P-against-Q. We also discuss implementations of these approaches and analyze training time complexity associated with each approach. We evaluate six different neural network system architectures for multi-class pattern classification along the dimensions of imbalanced data, large number of pattern classes, large vs. small training data through experiments conducted on well-known benchmark data.     

Servo controller design using neural networks

The dynamics of a physical plant may be difficult to express as concise mathematical equations. In practice there exist uncertainties that cannot be modeled with the system equations. Hence, robustness against system uncertainties is essential in a control system design. In this article, multilayered neural networks (MNNs) are used to compensate for model uncertainties of a dynamical system. Neural network models are used along with a classical linear servo controller derived from the linear state space equations. These models are trained so that system uncertainties are compensated. The design of a servo system indicates the enhanced performance of the neural-network-based servo controller as compared to the classical servo controller.     

Clustering Search for the Berth Allocation Problem

This work presents a new approach to the Berth Allocation Problem (BAP) for ships in ports. Due to the increasing demand for ships carrying containers, the BAP can be considered as a major optimization problem in marine terminals. In this paper, the BAP is considered as dynamic and modeled in discrete case and we propose a new alternative to solve it. The proposed alternative is based on applying the Clustering Search (CS) method using the Simulated Annealing (SA) for solutions generation. The CS is an iterative method which divides the search space in clusters and it is composed of a metaheuristic for solutions generation, a grouping process and a local search heuristic. The computational results are compared against recent methods found in the literature.     

Pruning backpropagation neural networks using modern stochastic optimisation techniques

Approaches combining genetic algorithms and neural networks have received a great deal of attention in recent years. As a result, much work has been reported in two major areas of neural network design: training and topology optimisation. This paper focuses on the key issues associated with the problem of pruning a multilayer perceptron using genetic algorithms and simulated annealing. The study presented considers a number of aspects associated with network training that may alter the behaviour of a stochastic topology optimiser. Enhancements are discussed that can improve topology searches. Simulation results for the two mentioned stochastic optimisation methods applied to non-linear system identification are presented and compared with a simple random search.     

Training neural networks by stochastic optimisation

We present a stochastic learning algorithm for neural networks. The algorithm does not make any assumptions about transfer functions of individual neurons and does not depend on a functional form of a performance measure. The algorithm uses a random step of varying size to adapt weights. The average size of the step decreases during learning. The large steps enable the algorithm to jump over local maxima/minima, while the small ones ensure convergence in a local area. We investigate convergence properties of the proposed algorithm as well as test the algorithm on four supervised and unsupervised learning problems. We have found a superiority of this algorithm compared to several known algorithms when testing them on generated as well as real data.     

按自然法则计算导论：一个颇具前途的学科

一、什么是按自然法则计算 Geoffrey．C．Fox在1991年[1]首次提出了“按自然法则计算”(Physical Computation)这一概念。按Fox的定义,按自然法则计算是“将大量的自然科学领域的思想、方法,用于其传统应用领域之外的其它领域”,将原思想、方法的本质提取出来,用于解决新领域的问题。 G．C．Fox例举了按自然法则计算的四大分支：将统计物理学中的退火思想应用于组合优化理论的模拟退火算法(简称SA);将生物学和神经细咆学思想用于“机器学习”和“模式识别”的“人工神经网络”;模拟牛顿力学方法的弹性网络方法;以及由模拟退火算法与启发式搜索算法相结合而演化来的确定性退火思想。     

自适应联想记忆细胞神经网络的优化设计

针对训练自适应联想记忆细胞神经网络(AM-CNN)过程收敛慢,设计出的网络抗噪性能不高的特点,通过融合蚁群优化算法和粒子群算法的思想,提出以目标网络对噪声模式的输出误差为目标函数,在目标函数的一个阈值分成的两个区间内,分别采取局部搜索和全局搜索策略,训练出AM-CNN的克隆模板的设计方法。数字模拟表明,与以往的设计方法相比,该算法能在细胞神经网络4~6次的迭代过程中稳定输出期望模式,收敛速度更快,设计出的AM-CNN性能比较稳定,并对噪声鲁棒,对高斯噪声N(0,0.8)准确率达到90%左右。     

Introducing the bounded derivative network—superceding the application of neural networks in control

The Bounded Derivative Network (BDN), the analytical integral of a neural network, is a natural and elegant evolution of universal approximating technology for use in automatic control schemes. This modeling approach circumvents the many real problems associated with standard neural networks in control such as model saturation (zero gain), arbitrary model gain inversion, ‘black box’ representation and inability to interpolate sensibly in regions of sparse excitation. Although extrapolation is typically not an advantage unless the understanding of the process is complete, the BDN can incorporate process knowledge in order that its extrapolation capability is inherently sensible in areas of data sparsity. This ability to impart process knowledge on the BDN model enables it to be safely incorporated into a model based control scheme.     

改进的小波神经网络在桥梁损伤中的预测研究

提出基于BP算法的小波神经网络改进算法。仿真结果表明它避免了BP 神经网络结构设计的盲目性和局部最优等非线性优化问题,简化了训练,具有较强的函数学习能力和推广能力。该算法成功应用于桥梁损伤预测,具有广泛的应用前景。     

Tabu search vs. simulated annealing as a function of the size of quadratic assignment problem instances

Performance comparisons between algorithms have a long tradition in metaheuristic research. An early example is comparisons between Tabu Search (TS) and Simulated Annealing (SA) algorithms for tackling the Quadratic Assignment Problem (QAP). The results of these comparisons are to a certain extent inconclusive, even when focusing on only these two types of algorithms. While comparisons of SA and TS algorithms were based on rather small-sized instances, here we focus on possible dependencies of the relative performance between SA and TS algorithms on instance size. In fact, our experimental results show that the assertion whether one algorithm is better than the other can depend strongly on QAP instance size even if one focuses on instances with otherwise same characteristics.     

无监督神经网络在游戏中的应用研究——解决游戏中的障碍物绕行问题

使用无监督神经网络解决了游戏中的障碍物绕行问题（obstacle avoidance）。使用遗传算法实现了无监督机制,该方法通过最优化适应度来改进神经网络的权值,使得神经网络得到最佳的输出值;利用以智能体（Agent）中心为出发点的5条射线模拟传感器（Sensor）,通过检测5条射线与障碍物边界的相交情况来感知环境。经过768代的进化,遗传算法种群最优适应度和平均适应度都有了明显提高,同时绕行成功率从12.5%上升到85%。     

Study on optimal design of pressure-adjusting spring using artificial neural networks

A preliminary discussion has been carried out on the traditional optimization design method for pressure-adjusting spring of relief valve. Based on the traditional optimization methods about the pressure-adjusting spring of the relief valve and combined with the advantages of neural network, this paper puts forward the optimization method with many parameters and a lot of constraints based on neural network in order to find the maximal inherent frequency. The object function of optimization is transformed into the energy function of the neural network and the mathematical model of neural network optimization about the pressure-adjusting spring of the relief valve is set up in this method which also puts forward its own algorithm. An example of application shows that network convergence gets stable state of minimization object function E, and object function converges to the utmost minimum point with steady function, then best solution is gained, which makes the design plan better. The algorithm of solution for the problem is effective about the optimum design of the pressure-adjusting spring. The specified technical performances of the relief valve are certified by experiments. The results of experiments showed that by configuring pressure-adjusting spring the dynamic performance and working stability of the relief valve are enhanced.     

基于神经网络的线性相位FIR滤波器设计

针对FIR滤波器的神经网络设计法,提出一种泛函连接人工神经网络的改进算法。通过设置不同的加权误差函数值来控制各个样本的学习率,改善了网络的学习效果;制定了神经网络训练集的选取规则,使用该规则选取样本对网络进行训练,可设计通带阻带截止频率指标精确可控的滤波器,克服了现有算法只能设计具有通带截止频率的滤波器和不能精确控制任意截止频率的不足。仿真结果表明所提出的方法能很好地满足设计要求。     

Solving fuzzy shortest path problems by neural networks

In this paper, we introduce the neural networks for solving fuzzy shortest path problems. The penalization of the neural networks is realized after transforming into crisp shortest path model. The procedure and efficiency of this approach are shown with numerical simulations.     

American Sign Language word recognition with a sensory glove using artificial neural networks

An American Sign Language (ASL) recognition system is being developed using artificial neural networks (ANNs) to translate ASL words into English. The system uses a sensory glove called the Cyberglove™ and a Flock of Birds^® 3-D motion tracker to extract the gesture features. The data regarding finger joint angles obtained from strain gauges in the sensory glove define the hand shape, while the data from the tracker describe the trajectory of hand movements. The data from these devices are processed by a velocity network with noise reduction and feature extraction and by a word recognition network. Some global and local features are extracted for each ASL word. A neural network is used as a classifier of this feature vector. Our goal is to continuously recognize ASL signs using these devices in real time. We trained and tested the ANN model for 50 ASL words with a different number of samples for every word. The test results show that our feature vector extraction method and neural networks can be used successfully for isolated word recognition. This system is flexible and open for future extension.     

Optimal design of neural networks for control in robotic arc welding

Robotic gas metal arc (GMA) welding is a manufacturing process which is used to produce high quality joints and has to a capability to be utilized in automation systems to enhance productivity. Despite its widespread use in the various manufacturing industries, the full automation of the robotic GMA welding has not yet been achieved partly because mathematical models for the process parameters for a given welding tasks are not fully understood and quantified. In this research, an attempt has been made to develop a neural network model to predict the weld bead width as a function of key process parameters in robotic GMA welding. The neural network model is developed using two different training algorithms; the error back-propagation algorithm and the Levenberg–Marquardt approximation algorithm. The accuracy of the neural network models developed in this study has been tested by comparing the simulated data obtained from the neural network model with that obtained from the actual robotic welding experiments. The result shows that the Levenberg–Marquardt approximation algorithm is the preferred method, as this algorithm reduces the root of the mean sum of squared (RMS) error to a significantly small value.     

基于深度卷积网络的多传感器信号故障诊断方法研究

针对传统故障诊断方法中多传感器数据融合技术难度大、特征提取困难等问题,提出了一种基于深度卷积网络的多传感器信号故障诊断方法,通过构建测量数据帧进行卷积计算实现多通道数据的自然融合,利用深度网络结构实现高层特征的自动提取和分类,从而高效地实现了故障分类诊断;经分别采用小规模数据集REF和大规模故障数据集BI02进行实验验证,均取得了较高的故障识别准确率,具有很强的工程应用价值。     

Aircraft design optimization

In this paper we describe briefly a set of procedures for the optimal design of full mission aerospace systems. This involves multi-physics simulations at various fidelity levels, surrogates, distributed computing and multi-objective optimization. Low-fidelity analysis is used to populate a database of inputs and outputs of the system simulation and Neural Networks are then designed to generate inexpensive surrogates. Higher fidelity is used only where is warranted and also to do a local exploration after global optimization techniques have been used on the surrogates in order to provide plausible initial values. The ideas are exemplified on a generic supersonic aircraft configuration, where one of the main goals is to reduce the ground sonic boom.