A genetic approach to the identification of linear dynamical systems with static nonlinearities

This paper investigates the use of genetic algorithms in the identification of linear systems with static nonlinearitites. Linear systems with static nonlinearities at the input known as the Hammerstein model, and linear systems with static nonlinearities at the output known as the Wiener model are considered in this paper. The parameters of the Hammerstein and the Wiener models are estimated using genetic algorithms from the input-output data by minimizing the error between the true model output and the identified model output. Using genetic algorithms, the Hammerstein and the Wiener models with known nonlinearity structure and unknown parameters can be identified. Moreover, systems with non-minimum phase characteristics can be identified. Extensive simulations have been used to study the convergence properties of the proposed scheme. Simulation examples are included to demonstrate the effectiveness and robustness of the proposed identification scheme.     

Subspace based approaches for Wiener system identification

We consider the problem of Wiener system identification in this note. A Wiener system consists of a linear time invariant block followed by a memoryless nonlinearity. By modeling the inverse of the memoryless nonlinearity as a linear combination of known nonlinear basis functions, we develop two subspace based approaches, namely an alternating projection algorithm and a minimum norm method, to solve for the Wiener system parameters. Based on computer simulations, the algorithms are shown to be robust in the presence of modeling error and noise.     

Identification of IIR Wiener systems with spline nonlinearities that have variable knots

An algorithm is developed for the identification of Wiener systems, linear dynamic elements followed by static nonlinearities. In this case, the linear element is modeled using a recursive digital filter, while the static nonlinearity is represented by a spline of arbitrary but fixed degree. The primary contribution in this note is the use of variable knot splines, which allow for the use of splines with relatively few knot points, in the context of Wiener system identification. The model output is shown to be nonlinear in the filter parameters and in the knot points, but linear in the remaining spline parameters. Thus, a separable least squares algorithm is used to estimate the model parameters. Monte-Carlo simulations are used to compare the performance of the algorithm identifying models with linear and cubic spline nonlinearities, with a similar technique using polynomial nonlinearities.     

Convergence analysis of recursive identification algorithms basedon the nonlinear Wiener model

Recursive identification algorithms, based on the nonlinear Wiener model, are presented. A recursive identification algorithm is first derived from a general parameterization of the Wiener model, using a stochastic approximation framework. Local and global convergence of this algorithm can be tied to the stability properties of an associated differential equation. Since inversion is not utilized, noninvertible static nonlinearities can be handled, which allows a treatment of, for example, saturating sensors and blind adaptation problems. Gauss-Newton and stochastic gradient algorithms for the situation where the static nonlinearity is known are then suggested in the single-input/single-output case. The proposed methods can outperform conventional linearizing inversion of the nonlinearity when measurement disturbances affect the output signal. For FIR (finite impulse response) models, it is also proved that global convergence of the schemes is tied to sector conditions on the static nonlinearity. In particular, global convergence of the stochastic gradient method is obtained, provided that the nonlinearity is strictly monotone. The local analysis, performed for IIR (infinite impulse response) models, illustrates the importance of the amplitude contents of the exciting signals     

Consistent identification of Wiener systems: A machine learning viewpoint

Wiener system identification has been recently performed by adopting a Bayesian semiparametric approach. In this framework, the linear system entering the first block is given a finite-dimensional parametrization, while nonparametric Gaussian regression is used to estimate the static nonlinearity in the second block. In this paper, we study the asymptotic behavior of this estimator when the number of noisy output samples tends to infinity without assuming the correctness of the Bayesian prior models. For this purpose, we interpret Wiener identification under a machine learning perspective. This allows us to extend recent results on function estimation in reproducing kernel Hilbert spaces to derive a condition guaranteeing the statistical consistency of the identification procedure. We also discuss how the violation of such a condition can lead to useless estimates of the Wiener structure.     

Initializing Wiener–Hammerstein models based on partitioning of the best linear approximation

This paper describes a new algorithm for initializing and estimating Wiener–Hammerstein models which consist of two linear parts with a static nonlinearity in between. The algorithm makes use of the best linear model of the system, which is a consistent estimate of the systems dynamics for Gaussian excitations. The linear model is split in all possible ways into two sub-models. For all possible splits, a Wiener–Hammerstein model is initialized which means that a nonlinearity is introduced in between the two sub-models. The linear parameters of this nonlinearity can be estimated using least-squares. All initialized models can then be ranked depending on the fit. Typically, one is only interested in the best one, for which all parameters are fitted using prediction error minimization.The paper explains the algorithm in detail and consistency of the initialization is proven. Computational aspects are investigated, showing that in most realistic cases, the number of splits of the initial linear model remains low enough to make the algorithm useful. The algorithm is illustrated on an example where it is shown that the initialization is a tool to avoid many local minima.     

Comment on ‘Frequency domain identification of Wiener models’, by E.W. Bai, Automatica 39 (2003), 1521–1530

A frequency-domain identification method is proposed by Bai (2003) for Wiener systems. The key component is a phase estimator that gives estimates of the linear subsystem phase for different frequencies. These estimates are then used to identify the whole system. In this note, it is shown that the phase estimator is not generally consistent. Consequently, consistency of the overall identification method is in turn not generally guaranteed.     

A subspace-based identification of Wiener–Hammerstein benchmark model

This paper develops a subspace-based method of identifying the Wiener–Hammerstein system, where a nonlinearity is sandwiched by two linear subsystems. First, a state space model of the best linear approximation of it is identified by using a subspace identification method and the poles of the best linear model are allocated between two linear subsystems by a state transformation. Unknown system parameters and coefficients of a basis function expansion of the nonlinearity are estimated by using the separable least-squares for all possible allocations of poles, so that there is a possibility that many iterative minimization problems should be solved. Finally, the best Wiener–Hammerstein system that yields the minimum mean square error is selected. Numerical results for a benchmark model show the applicability of the proposed method.     

维纳模型的最小二乘支持向量机辨识

针对传统维纳模型辨识方法存在算法复杂、精度低的问题,通过对最小二乘支持向量机建模原理和维纳模型结构特点的分析,提出一种基于最小二乘支持向量机的维纳模型辨识新方法.该方法充分利用了维纳模型中具有线性环节这一先验知识,实现了线性和非线性环节参数的同时辨识.对于多变量维纳模型,该方法同样适用.给出并证明了该方法存在唯一解的约束条件 - 参数部分列满秩.仿真实验表明了该方法的有效性,与标准最小二乘支持向量机辨识方法相比,该方法具有更高的精度.     

Least-Squares Support Vector Machines for the identification of Wiener–Hammerstein systems

This paper considers the identification of Wiener–Hammerstein systems using Least-Squares Support Vector Machines based models. The power of fully black-box NARX-type models is evaluated and compared with models incorporating information about the structure of the systems. For the NARX models it is shown how to extend the kernel-based estimator to large data sets. For the structured model the emphasis is on preserving the convexity of the estimation problem through a suitable relaxation of the original problem. To develop an empirical understanding of the implications of the different model design choices, all considered models are compared on an artificial system under a number of different experimental conditions. The obtained results are then validated on the Wiener–Hammerstein benchmark data set and the final models are presented. It is illustrated that black-box models are a suitable technique for the identification of Wiener–Hammerstein systems. The incorporation of structural information results in significant improvements in modeling performance.     

Recursive identification for Hammerstein–Wiener systems with dead-zone input nonlinearity

A new recursive algorithm is proposed for the identification of a special form of Hammerstein–Wiener system with dead-zone nonlinearity input block. The direct motivation of this work is to implement on-line control strategies on this kind of system to produce adaptive control algorithms. With the parameterization model of the Hammerstein–Wiener system, a special form of model estimation error is defined; and then its approximate formula is given for the following derivation. Based on these, a recursive identification algorithm is established that aims at minimizing the sum of the squared parameter estimation errors. The conditions of uniform convergence are obtained from the property analysis of the proposed algorithm and an adaptive setting method for a weighted factor in the algorithm is given, which enhances the convergence of the proposed algorithm. This algorithm can also be used for the identification of the Hammerstein systems with dead-zone nonlinearity input block. Three simulation examples show the validity of this algorithm.     

Recursive identification for Wiener model with discontinuous piece-wise linear function

This paper deals with identification of Wiener systems with nonlinearity being a discontinuous piece-wise linear function. Recursive estimation algorithms are proposed to estimate six unknown parameters contained in the nonlinearity and all unknown coefficients of the linear subsystem by using the iid Gaussian inputs. The estimates are proved to converge to the corresponding true values with probability one. A numerical example is given to justify the obtained theoretical results.     

On identification of Hammerstein and Wiener model with application to virtualised software system

This paper proposes a system identification method for estimating virtualised software system dynamics within the framework of a Hammerstein–Wiener model. Building on the authors’ previous work in identification and control of the software systems, the approach utilises frequency sampling filter structure to describe the linear dynamics and B-spline curve functions for the inverse static output nonlinearity. Furthermore, the issue on parameter selection for B-spline model approximation of scatter data is addressed by using a data clustering method. An experimental test-bed of virtualised software system is established to generate real observational data which are used to confirm the performance of the proposed approach. The identification results have shown that the model efficacy is increased with the proposed approach because the dimension of the nonlinear model can be significantly reduced while maintaining the desired accuracy.     

Identification of time-varying pH processes using sinusoidal signals

This paper presents an approach to the identification of time-varying, nonlinear pH processes based on the Wiener model structure. The algorithm produces an on-line estimate of the titration curve, where the shape of this static nonlinearity changes as a result of changes in the weak-species concentration and/or composition of the process feed stream. The identification method is based on the recursive least-squares algorithm, a frequency sampling filter model of the linear dynamics and a polynomial representation of the inverse static nonlinearity. A sinusoidal signal for the control reagent flow rate is used to generate the input-output data along with a method for automatically adjusting the input mean level to ensure that the titration curve is identified in the pH operating region of interest. Experimental results obtained from a pH process are presented to illustrate the performance of the proposed approach. An application of these results to a pH control problem is outlined.     

Interval fuzzy modeling applied to Wiener models with uncertainties.

This correspondence addresses the problem of interval fuzzy model identification and its use in the case of the robust Wiener model. The method combines a fuzzy identification methodology with some ideas from linear programming theory. On a finite set of measured data, an optimality criterion which minimizes the maximum estimation error between the data and the proposed fuzzy model output is used. The min-max optimization problem can then be seen as a linear programming problem that is solved to estimate the parameters of the fuzzy model in each fuzzy domain. This results in lower and upper fuzzy models that define the confidence interval of the observed data. The model is called the interval fuzzy model and is used to approximate the static nonlinearity in the case of the Wiener model with uncertainties. The resulting model has the potential to be used in the areas of robust control and fault detection.     

Circle criteria in recursive identification

Positive real conditions and differential sector conditions have recently been shown to imply global convergence w.p. 1, for recursive identification schemes based on a class of single-input/single-output linear Wiener models. The models consist of linear dynamics followed by a static output nonlinearity. The model structure is hence closely related to that of the Lure problem in the stability theory of feedback systems. This paper proves that the conditions for convergence can be transformed to graphical circle criteria, depending on the sector conditions and on the Nyquist plot of a transfer function related to the prior knowledge of the poles of the identified system