基于神经网络的时变非线性系统迭代学习辨识

时变神经网络结构可简单地取为常规神经网络连接形式,但连接权却是时变的.如何确定时变权是应用时变神经网络时的难题.迭代学习方法是一种合理的选择,它不同于将时变连接权展成Taylor级数,通过训练多项式系数的处理方法.而且,后者的处理方式不可避免地存在截断误差.对于有限区间连续时变非线性系统的神经网络建模与辨识,借助于重复运行过程,以迭代学习算法调整权值,进行网络训练.不计逼近误差,提出的学习算法能够使得辨识误差在整个区间上渐近收敛于零.为处理非零但有界的逼近误差,采用带死区的迭代学习算法.逼近误差界值已知时,文中证明带死区修正的迭代学习算法使得辨识误差在整个区间上渐近收敛于由死区界定的邻域内.对于逼近误差界值未知的情形也进行了讨论.     

一种有实用价值的小波包的构造

小波分析是八十年代发展起来的新数学分支,小波基和小波包的构造不但在理论上,而且在应用中具有实际意义上,本文构造了一个具有较快衰减性,较好光滑性及对称性且能实现数值计算的正交小波包。     

基于信息压缩矩阵特征值映射的时变系统UD分解辨识算法

在系统辨识领域遗忘因子UD分解算法(一种通过对系统数据矩阵进行UD分解的在线辨识算法)具有对时变系统阶次和参数同步估计的优异性能,但传统的遗忘策略不能从根本上解决信息压缩矩阵数据过饱和问题,为了拓展现有UD分解算法在时变系统的适用范围,同时针对数据空间分布不均匀性,提出一种基于信息压缩矩阵特征值映射的UD分解辨识算法....     

一类时变仿射非线性系统的完全线性化

本用微分几何方法讨论时变仿射非线性系统的完全线性化问题,即同时线性化状态方程和输出方程,给出了一类时变仿射非线性系统完全线性化的充分条件。     

任意矩阵伸缩的正交小波包

1 引言 Coifman和Meyer引入L~2(R)中正交小波包,可以用张量积形式构造L~2(R~2)上的二维正交小波包;Chui和Li研究单变量非正交小波包和对偶小波包;Shen给出矩阵伸缩为2I时L~2(R~s)上非张量积小波包的构造算法;程正兴给出矩阵小波包的构     

一类时变仿射非线性系统的完全线性化

本文用微分几何方法讨论时变仿射非线性系统的完全线性化问题 ,即同时线性化状态方程和输出方程 ,给出了一类时变仿射非线性系统完全线性化的充分条件     

用小波网络辨识离散非线性系统

本文用具紧支集的尺度函数之张量乘积构成人工神经网络的基函数,再由这个小波神经网络辨识静态与动态的离散线性系统,并且证明了依所给的方法产生的模型是收敛的.最后,用一个仿真例子,说明如何实现算法及算法的效果.     

一类二维不可分非正交小波包

Zuowei Shen构造了L^2（R^s)空间的二进制小泡包,与伸缩矩阵M＝（1 1 1 －1）相关的小波适用于二维图像处理中的梅花状子取样,本文给出了一种构造非正交M－小滤包的方法。     

小波和小波包算法与图象压缩

在小波方法的实际应用中,利用尺度方程系数构造得到的小波和小波包变换算法即金字塔算法是非常重要的。考虑到应用的需要,本文结合数字图象数据压缩问题的讨论介绍了一维和二维的小波和小波包算法。最后,数字图象变换压缩实验的结果说明了小波和小波包分析在数字图象压缩研究中的潜在作用     

一类二元半正交小波包的构造

本文给出伸缩矩阵行列式为2的一类二元半正交小波包的构造算法.该小波包是以频域给出的,随着用于小波包分裂的滤波器选取的不同会得到L2(R2)中形态各异的Riesz基,这样使得L2(R2)中小波基的选择更灵活.     

Optimal input system identification for nonlinear dynamic systems

The estimation accuracy for nonlinear dynamic system identification is known to be maximized by the use of optimal inputs. Few examples of the design of optimal inputs for nonlinear dynamic systems are given in the literature, however. The performance criterion is selected such that the sensitivity of the measured state variables to the unknown parameters is maximized. The application of Pontryagin's maximum principle yields a nonlinear two-point boundary-value problem. In this paper, the boundary-value problem for a simple nonlinear example is solved using two different methods, the method of quasilinearization and the Newton-Raphson method. The estimation accuracy is discussed in terms of the Cramer-Rao lower bound.     

QRS complex detection based on multi wavelet packet decomposition

We present in this paper a wavelet packet based QRS complex detection algorithm. Our proposed algorithm consists of a particular combination of two vectors obtained by applying a designed routine of QRS detection process using ‘haar’ and ‘db10’ wavelet functions respectively. The QRS complex detection routine is based on the histogram approach where our key idea was to search for the node with highest number of histogram coefficients, at center, which we assume that they are related to the iso-electric baseline whereas the remaining least number coefficients reflect the R waves peaks. Following a classical approach based of a calculated fixed threshold, the possible QRS complexes will be determined. The QRS detection complex algorithm has been applied to the whole MIT-BIH arrhythmia Database to assess its robustness. The algorithm reported a global sensitivity of 98.68%, positive predictive value of 97.24% and a percentage error of 04.12%. Eventhough, the obtained global results are not as excellent as expected, we have demonstrate that our designed QRS detection algorithm performs good on a partial selected high percentage of the whole database, e.g., the partial results, obtained when applying the algorithm on 85.01% of the whole MIT-BIH arrhythmia Database, are 99.14% of sensitivity, 98.94% of positive predictive value and 01.92% of percentage error.     

Application of wavelet bases in linear and nonlinear approximation to functions from Besov spaces

Linear and nonlinear approximations to functions from Besov spaces B _{p, q} ^σ ([0, 1]), σ > 0, 1 ≤ p, q ≤ ∞ in a wavelet basis are considered. It is shown that an optimal linear approximation by a D-dimensional subspace of basis wavelet functions has an error of order D ^{-min(σ, σ + 1/2 ? 1/p)} for all 1 ≤ p ≤ ∞ and σ > max(1/p ? 1/2, 0). An original scheme is proposed for optimal nonlinear approximation. It is shown how a D-dimensional subspace of basis wavelet functions is to be chosen depending on the approximated function so that the error is on the order of D ^?σ for all 1 ≤ p ≤ ∞ and σ > max(1/p ? 1/2, 0). The nonlinear approximation scheme proposed does not require any a priori information on the approximated function.     

Exceptional sets and wavelet packets orthonormal bases

We give a partial positive answer to a problem posed by Coifman et al. in [1]. Indeed, starting from the transfer function m₀ arising from the Meyer wavelet and assuming m₀=1 only on [–/3, /3], we provide an example of pairwise disjoint dyadic intervals of the form I(n, q)=[2^qn, 2^q(n+1)), (n, q)EN×Z, which cover [0, +) except for a set A of Hausdorff dimension equal to 1/2, and such that the corresponding wavelet packets 2^q/2w_n (2^qx–k), kZ, (n, q)EN×Z form an orthonormal basis of L²(R).     

基于小波变换和人眼视觉系统特性的图像降噪方法

利用小波变换与人眼视觉系统（human visual system,HVS）的多通道特性相匹配的特点,提出一种基于人类视觉系统的图像降噪方法.该方法在P-M模型中引入小波变换与视觉敏感函数,并且结合视觉敏感函数的带通特性,提出一种新的扩散函数.实验结果表明,该方法得到的图像不论在客观评价（峰值信噪比）方面还是主观测评方面,都能达到较好的效果.     

Estimation of nonlinear systems without a priori structural information and its application in Hammerstein model identification with dynamic nonlinearities

Email: er-wei-bai{at}uiowa.edu Received on July 30, 2005; Accepted on July 23, 2006 In the paper, we discuss identification of a nonlinear systemwithout structural information and propose two methods, thekernel method and the orthonormal basis method. The convergenceresults are established for both methods without a priori structuralinformation. We then apply the results to identification ofHammerstein models with an unknown dynamic nonlinearity. Itis also shown that identification of the linear part in Hammersteinmodels is possible with no knowledge of the dynamic nonlinearity.     

A low-order active fault-tolerant state space self-tuner for the unknown sampled-data nonlinear singular system using OKID and modified ARMAX model-based system identification

In this paper, we present two control schemes for the unknown sampled-data nonlinear singular system. One is an observer-based digital redesign tracker with the state-feedback gain and the feed-forward gain based on off-line observer/Kalman filter identification (OKID) method. The presented control scheme is able to make the unknown sampled-data nonlinear singular system to well track the desired reference signal. The other is an active fault tolerance state-space self-tuner using the OKID method and modified autoregressive moving average with exogenous inputs (ARMAX) model-based system identification for unknown sampled-data nonlinear singular system with input faults. First, one can apply the off-line OKID method to determine the appropriate (low-) order of the unknown system order and good initial parameters of the modified ARMAX model to improve the convergent speed of recursive extended-least-squares (RELS) method. Then, based on modified ARMAX-based system identification, a corresponding adaptive digital control scheme is presented for the unknown sampled-data nonlinear singular system with immeasurable system state. Moreover, in order to overcome the interference of input fault, one can use a fault-tolerant control scheme for unknown sampled-data nonlinear singular system by modifying the conventional self-tuner control (STC). The presented method can effectively cope with partially abrupt and/or gradual system input faults. Finally, some illustrative examples including a real circuit system are given to demonstrate the effectiveness of the presented design methodologies.     

A new approach to fuzzy wavelet system modeling

In this paper, we propose simple but effective two different fuzzy wavelet networks (FWNs) for system identification. The FWNs combine the traditional Takagi–Sugeno–Kang (TSK) fuzzy model and discrete wavelet transforms (DWT). The proposed FWNs consist of a set of if–then rules and, then parts are series expansion in terms of wavelets functions. In the first system, while the only one scale parameter is changing with it corresponding rule number, translation parameter sets are fixed in each rule. As for the second system, DWT is used completely by using wavelet frames. The performance of proposed fuzzy models is illustrated by examples and compared with previously published examples. Simulation results indicate the remarkable capabilities of the proposed methods. It is worth noting that the second FWN achieves high function approximation accuracy and fast convergence.     

Identification of nonlinear system using extreme learning machine based Hammerstein model

In this paper, a new method for nonlinear system identification via extreme learning machine neural network based Hammerstein model (ELM-Hammerstein) is proposed. The ELM-Hammerstein model consists of static ELM neural network followed by a linear dynamic subsystem. The identification of nonlinear system is achieved by determining the structure of ELM-Hammerstein model and estimating its parameters. Lipschitz quotient criterion is adopted to determine the structure of ELM-Hammerstein model from input–output data. A generalized ELM algorithm is proposed to estimate the parameters of ELM-Hammerstein model, where the parameters of linear dynamic part and the output weights of ELM neural network are estimated simultaneously. The proposed method can obtain more accurate identification results with less computation complexity. Three simulation examples demonstrate its effectiveness.