基于回声状态网络的时间序列预测方法研究

针对回声状态网络(Echo State Networks,ESNs)输入序列延迟时间(和嵌入维数D的选择以及储备池的适应性问题,利用自相关性分析法从被预测样本序列构建ESNs网络输入,并通过移动通信话务量的预测问题,采用实验分析的方法讨论了储备池参数选择对于时间序列预测性能的影响.与采用ARMA和BP神经网络的预测方法相比,新方法在保证预测精度和效率的情况下,具有更好的泛化能力.     

基于回声状态网络的OTHR海杂波抑制方法

在短相干积累时间(CIT)情况下,天波超视距雷达(OTHR)中低速目标检测很困难:低速目标靠近强大的海杂波频谱;短CIT导致多普勒分辨率低,使目标信号与海杂波更难区分。传统方法一般利用海杂波与目标信号的时频特征不同来抑制海杂波,目标速度较高时这些方法很有效,然而在短CIT、低速目标情况下目标与海杂波信号的时频特征的区分度很小,使得传统方法难以有效抑制海杂波。针对上述问题,考虑到海杂波与目标信号具有不同的动力学特征,提出一种基于回声状态网络的天波雷达海杂波抑制方法。该方法首先用海杂波参考信号训练回声状态网络,使该神经网络"记住"海杂波的混沌动态特征;然后用前述训练好的网络重构和预测天波雷达回波中...     

基于回声状态网络的音频频带扩展方法

宽带音频通信系统对传输信号有效带宽的限制会降低重建音频的主观质量和自然程度.本文提出了一种基于回声状态网络的宽带向超宽带音频盲目式频带扩展方法.该方法借助回声状态网络来模拟音频信号高低频频谱参数间的映射关系,并依据网络模型中的时延递归结构连续更新系统状态来近似描述音频特征的时域演变过程,有效地估计了高频成分的频谱包络.同时,结合频谱复制方法得到的高频频谱细节,该方法实现了宽带向超宽带音频的有效扩展.测试结果表明,本文所提方法提升了宽带音频的听觉质量;对于多数测试数据,该方法在静态和动态失真方面获得了优于高斯混合模型扩展方法的扩展性能.     

模糊回声状态网络

针对基于梯度下降的模糊递归神经网络训练效率低、容易陷入局部极小的缺点,本文基于回声状态网络(Echo State Networks,ESNs)和TS模型提出一种新的模糊模型结构——模糊回声状态网络(Fuzzy Echo State Networks,FESNs).FESNs由多条TS类型的模糊规则组成,规则后件采用ES...     

基于回声状态网络的功耗曲线特征提取

在模板(Template Attacks,TA)攻击的研究中,如何利用功耗曲线信息,合理选择有效点,增强匹配效果是改进模板攻击的一个重要方向.文中分析了目前有关功耗曲线主要特征提取方法的优缺点,并提出了一种基于回声状态网络(Echo State Network,ESN)的功耗曲线特征提取方法.该方法针对ESN分类方法中的储备池参数选择问题,以时间预测序列精度为标准,采用网格法进行参数空间的优化搜索,并利用神经网络以数据样本形式作为定量知识自行处理的能力,对粗略对齐下的功耗曲线的特征提取能力进行了测试和评估.实验结果表明,基于ESN功耗曲线特征提取方法在功耗曲线数量相同条件下,通过合理选择内核参数,能够降低模板攻击对功耗曲线预处理技术的依赖,提高正确密钥的分类精度.     

基于小波回声状态网络的声源定位算法

为了解决传统声源定位算法存在定位误差大的问题,提出了一种小波回声状态网络的声源定位算法。首先建立声源定位的阵列信号处理模型,并采集声源信号数据,然后采用小波分解将声源信号分解成为高频与低频两部分,并采用回声状态网络分别对高频与低频数据进行建模,最后采用小波重构对高频与低频估计结果进行融合,并对算法性能进行仿真测试。结果表明,本文算法可以精确对声源进行定位,相对于其它声源定位算法,具有更好的适用性和可靠性。     

混沌背景中微弱信号检测的回声状态网络方法

对复杂非线性系统的相空间重构理论进行了研究分析,提出了混沌背景中微弱信号检测的回声状态网络方法。针对回声状态网络模型参数选取困难这一问题,采用遗传算法对其模型参数进行优化。将回声状态网络模型参数作为遗传算法的个体,混沌时间序列预测均方根误差的倒数作为适应度函数,通过选择、交叉、变异等操作获得适合数据特点的最优模型参数。根据回声状态网络强大的学习和非线性处理能力,利用得到的回声状态网络模型最优参数建立混沌背景噪声的单步预测模型,将淹没在混沌背景噪声中的微弱瞬态信号和周期信号从预测误差中检测出来。以Lorenz系统和实测的海杂波数据作为混沌背景噪声进行仿真实验,仿真结果表明,本文所提方法在预测精度和训练速度方面均优于支持向量机和神经网络模型,能够有效地检测出混沌背景噪声中的微弱目标信号,且具有较小的预测误差。      

基于l2正则化回声状态网络的模拟电路故障诊断

为了提高模拟电路故障诊断的准确率,提出了一种基于l2正则化回声状态网络（l2-RESN）的模拟电路故障诊断方法。l2-RESN在ESN的约束优化函数中引入l2正则化因子,推导带正则化因子的ESN输出权重计算公式,避免传统的ESN算法因矩阵奇异而降低模型泛化能力。实验结果表明,相比于支持向量机（SVM）和标准ESN,l2-RESN的诊断准确率分别提高1.11%和18.34%。证明l2-RESN能够有效提高模拟电路诊断的准确性。     

基于混沌理论与改进回声状态网络的网络流量多步预测

网络流量预测是网络管理及网络拥塞控制的重要问题,针对该问题提出一种基于混沌理论与改进回声状态网络的网络流量预测方法。首先利用0-1混沌测试法与最大Lyapunov指数法对不同时间尺度下的网络流量样本数据进行分析,确定网络流量在不同时间尺度下都具有混沌特性。将相空间重构技术引入网络流量预测,通过C-C方法确定延迟时间,G-P算法确定嵌入维数。对网络流量时间序列进行相空间重构之后,利用一种改进的回声状态网络进行网络流量的多步预测。提出一种改进的和声搜索优化算法对回声状态网络的相关参数进行优化以提高预测精度。利用网络流量的公共数据集以及实际数据进行了仿真,结果表明,提出的预测方法具有更高的预测精度以及更小的预测误差。     

Fault Diagnosis of Analog Circuits Using Bayesian Neural Networks with Wavelet Transform as Preprocessor

We have developed an analog circuit fault diagnostic system based on Bayesian neural networks using wavelet transform, normalization and principal component analysis as preprocessors. Our proposed system uses these preprocessing techniques to extract optimal features from the output(s) of an analog circuit. These features are then used to train and test a neural network to identify faulty components using Bayesian learning of network weights. For sample circuits simulated using SPICE, our neural network can correctly classify faulty components with 96% accuracy.     

An Analog Circuit Fault Characterization Methodology

A methodology for diagnosing and characterizing multiple faults in analog circuits, and results from applying this methodology to a real circuit is presented. Our method is a novel combination of a Simulation Before Test (SBT) and Interpolation After Test (IAT) methodology. Our method uses the classical SBT concept of a fault dictionary database constructed before test. It also uses a method of IAT that consists in using the measurements to guide an interpolation algorithm to effectively increase the local resolution of the fault dictionary database and thereby yield the most likely test parameter value. Our methods underlying principle is to characterize the fault-free and faulty circuit cases by their impulse responses obtained by simulation and subsequently stored in a fault dictionary database. The method uses the technique of Lagrange interpolation to resolve the faults between the fault dictionary database entries and the actual measurements. Our experimental results reveal that the method is effective for characterizing faults when the simulations match the measurements sufficiently. Consequently, the methods effectiveness depends highly on the quality of the models used to build the dictionary as well as on the accuracy of the measurements.Yvan Maidon was born in Bordeaux, France. He received the M.Sc degree in (electronics) applied physics from the University of Bordeaux, in 1980. He is currently Head of the Department for Applied Sciences in Electrical and Electronic Engineering at the University of Bordeaux 1. His special research interests include failure analysis and relaibility of analog circuits. He has also developed original BICS for mixed circuits and SoC testing.Thomas Zimmer is currently Professor at the University of Bordeaux 1. He received the M.Sc. degree in physics from the University of Würzburg, Germany, in 1989 and the Ph.D. degree in electronics from the University of Bordeaux 1, France, in 1992. His research interests include characterization and modeling of high frequency bipolar devices. He has authored and co-authored about 70 scientific and technical publications including several book chapters. He is also co-founder of the start-up company XMOD.André Ivanov is Professor in the Department of Electrical and Computer Engineering, at the University of British Columbia. Prior to joining UBC in 1989, he received his B.Eng. (Hon.), M. Eng., and Ph.D. degrees in Electrical Engineering from McGill University. In 1995–96, he spent a sabbatical leave at PMC-Sierra, Vancouver, BC. He has held invited Professor positions at the University of Montpellier II, the University of Bordeaux I, and Edith Cowan University, in Perth, Australia. His primary research interests lie in the area of integrated circuit testing, design for testability and built-in self-test, for digital, analog and mixed-signal circuits, and systems on a chip (SoCs). He has published widely in these areas and holds several patents in IC design and test. Besides testing, Ivanov has interests in the design and design methodologies of large and complex integrated circuits and SoCs. Ivanov has served and continues to serve on numerous national and international steering, program, and/or organization committees in various capacities. Recently, he was the Program Chair of the 2002 VLSI Test Symposium (VTS 02) and the General Chair for VTS 03 and VTS 04. In 2001, Ivanov co-founded Vector 12, a semiconductor IP company. He has published over 100 papers in conference and journals and holds 4 US patents. Ivanov serves on the Editorial Board of the IEEE Design and Test Magazine, and Kluwers Journal of Electronic Testing: Theory and Applications. Ivanov is currently the Chair of the IEEE Computer Societys Test Technology Technical Council (TTTC). He is a Golden Core Member of the IEEE Computer Society, a Senior Member of the IEEE, a Fellow of the British Columbia Advanced Systems Institute and a Professional Engineer of British Columbia.     

模拟电路故障实时诊断的滤波器方法

本文讨论了一种模拟电路故障实时诊断的滤波器方法,该方法首先用状态空间模型来描述模拟电路,然后利用强跟踪滤波器对模型进行状态与参数联合估计,再利用修正的Bayes分类算法对故障进行检测与分离,从而实现故障的实时诊断.实验结果证明了本方法是有效的.     

基于BP神经网络的大规模电路模块级故障快速诊断方法

根据大规模电路故障诊断网络撕裂法和交叉撕裂搜索方法,采用基于误差反向传播算法的多层前向神经网络(BP神经网络)记载多次撕裂信息,提出了一种新型基于BP神经网络的大规模电路模块级快速诊断方法。该方法能快速有效地并行处理定位故障模块,具有测前工作量小,实时诊断性强等优点。     

模拟电路故障诊断的结点电压灵敏度比值法

在证明线性电路中结点电压变化量比值等于结点电压灵敏度比值的基础上,提出了结点电压灵敏度比值法,通过结点电压变化量比值和结点电压灵敏度比值的比对确定电路的故障元件。理论分析和实验结果表明,该方法算法简单、诊断速度快,在可测点受限条件下具有较高的诊断精度,特别适合大规模线性模拟电路的故障诊断和测试。     

电路故障诊断的神经网络方法

分析了神经网络的容错能力、分类能力以及高速神经计算能力在电路故障诊断中的优势。举例说明了应用神经网络方法进行数字电路及模拟电路故障诊断的基本方法和步骤,着重突出了实际应用中存在的故障特征选取、神经网络建立、多故障处理等难点,提出了具体解决方案,指出了进一步的发展方向和理论创新点。     

基于径向基函数网络的模拟电路故障诊断设计

模拟电路的固有特点使其故障诊断较数字电路困难.相对于BP网络,RBF神经网络具有最佳逼近性能且收敛快、无局部极小,可引入解决上述困难.根据具体电路,定义故障,选定测试点,确定网络结构,用Pspice获得训练样本,经过训练得到RBF网络.网络的输入为从测试点得到的输入向量,输出为对应的故障.为了验证网络的泛化性能,对每种...