Elsevier

Signal Processing

Volume 106, January 2015, Pages 294-300
Signal Processing

States based iterative parameter estimation for a state space model with multi-state delays using decomposition

https://doi.org/10.1016/j.sigpro.2014.08.011Get rights and content

Highlights

  • Consider estimation problems of a class of time-delay state space systems.

  • Use the hierarchical identification principle and the iterative technique.

  • Present a gradient based and a least squares based iterative estimation algorithms.

Abstract

This paper is concerned with the parameter estimation of a class of time-delay systems in the state space form. By using the hierarchical identification principle, a gradient based and a least squares based iterative identification algorithms are developed to compute these parameter estimates. The basic idea is to decompose a state space system into two subsystems, one containing a parameter vector and the other containing a parameter matrix. The simulation example demonstrates the efficacy of the proposed theory.

Introduction

It is well known that time delays are frequently encountered in various practical systems such as chemical processes, communication networks and digital signal analysis [1], [2]. Such a delay is usually a source of instability and performance deterioration. The analysis and synthesis of time-delay systems have received growing interests in recent years [3], [4]. Several approaches have been proposed [5], [6], in particular using the linear matrix inequality approach which has been extensively used for design of uncertain systems [7], [8].

Parameter estimation is the eternal theme of system identification [9] and many identification methods were developed for linear systems [10], [11], [12] and errors-in-variables systems [13], [14], e.g., the iterative parameter estimation algorithm for multivariable systems [15], the hierarchical gradient based identification algorithm [16] and the hierarchical least squares identification method [17], [18] for multivariable systems, and the data filtering based recursive least squares algorithm for Hammerstein systems [19]. Recently, an auxiliary model based parameter estimation algorithm was presented for dual-rate output error systems with colored noises [20], a modified subspace identification algorithm was presented for periodically non-uniformly sampled systems by using the lifting technique [21], an efficient hierarchical identification method was proposed for general dual-rate sampled-data systems [22], and an adaptive identification problem was discussed for linear-in-the-parameters nonlinear filters using the periodic input sequences [23].

Iterative methods can be used for system identification with unknown variables in the information vector [24]. For example, Ding et al. proposed a hierarchical gradient based iterative algorithm and a hierarchical least squares based iterative algorithm for CARARMA systems [25]; Li studied the parameter estimation for Hammerstein CARARMA systems based on the Newton iteration [26]; Ding et al. presented the decomposition based Newton iterative identification method for a Hammerstein nonlinear FIR system with ARMA noise [27].

The significance and difficulty of estimating state space systems are widely admitted [28], [29]. Consequently, there is extremely large and active research effort directed towards this problem. A primary aspect is that the states in the process should be identified first or simultaneously with the system parameters since the states are usually unknown [30], [31]. Rapisarda and Trentelman studied identification and data-driven model reduction of state-space representations of lossless and dissipative systems from noise-free data [32]; Mercère and Bako presented parameterization and identification of multivariable state space systems using a canonical approach [33].

This paper develops a new system identification algorithm using the iterative technique and the hierarchical identification principle. We frame our study in the identification of state space systems with multi-state delays. The basic idea is to use the iterative technique to deal with the identification problem and to present a hierarchical gradient based iterative algorithm and a hierarchical least squares based iterative algorithm for a state space model with multi-state delays.

This paper is organized as follows. Section 2 gives the identification model of state delay systems. Section 3 introduces the hierarchical gradient based iterative algorithm. Section 4 presents the least squares based iterative estimation algorithm. Section 5 provides an illustrative example for the results in this paper. Finally, we offer some concluding remarks in Section 6.

Section snippets

System description and identification model

Let us introduce the necessary notation. “AX” or “XA” stands for “A is defined as X”; the symbol I(In) stands for an identity matrix of appropriate size (n×n); z represents a unit forward shift operator: zx(t)=x(t+1) and z1x(t)=x(t1); the superscript T denotes the matrix/vector transpose; the norm of a matrix X is defined as X2tr[XXT]; λmax[X] represents the maximum eigenvalue of the non-negative definite matrix X; ϑ^(t) denotes the estimate of ϑat time t; 1m×n represents an m×n matrix

The hierarchical gradient based iterative algorithm

For two models in (7), (8), we derive a hierarchical identification algorithm to estimate the parameter matrix θ in S1 and the parameter vector ϑ in S2.

Suppose that the data length Ln2+rn2+n and define the stack information matrices Φ1 and Φ2, the stacked state matrices Ξ and Ω and vector X asΦ1[φ1(1),φ1(2),,φ1(L)]R(nr+1)×L,Φ2[φ2(1),φ2(2),,φ2(L)]R(n+nr+1)×L,Ξ[ξ(2),ξ(3),,ξ(L+1)]R(n1)×L,Ω[ω(1),ω(2),,ω(L)]R(n1)×L,X[xn(2),xn(3),,xn(L+1)]TRL.From (7), (8), we haveS1:ΞΩ=θTΦ1,S2:X=Φ2

The hierarchical least squares based iterative algorithm

The hierarchical gradient based iterative algorithm converges slowly. To improve the convergence rate, here derives a hierarchical least squares based iterative algorithm.

Referring to the method in [16], [17] and minimizing (14), (15), we can obtain the least squares based iterative solutions θ^k and ϑ^k of θ and ϑ:θ^k=θ^k1+μk[Φ1Φ1T]1Φ1[ΞΩθ^k1TΦ1]T,ϑ^k=ϑ^k1+μk[Φ2Φ2T]1Φ2[XΦ2Tϑ^k1],where μk0 is the time-varying step-size or time-varying convergence factor.

Thus, we can obtain the

Example

Consider the following state space system with 2-step state-delay:x(t+1)=[01a2a1]x(t)+[b11b12b13b14]x(t1)+[b21b22b23b24]x(t2)+[g1g2]u(t)=[010.450.80]x(t)+[0.200.150.150.20]x(t1)+[0.200.180.180.05]x(t2)+[11]u(t),θ=[b11,b12,b21,b22,g1]T=[0.20,0.15,0.20,0.18,1]T,ϑ=[a2,a1,b13,b14,b23,b24,g2]T=[0.45,0.80,0.15,0.20,0.18,0.05,1]T.In simulation, the input {u(t)} is taken as an uncorrelated persistent excitation signal vector sequence with zero mean and unit variance. Taking the data

Conclusion

This paper studies parameter estimation problems for a class of linear dynamic systems. Compared with the stochastic gradient algorithm and the recursive least squares algorithms, the proposed iterative algorithms can generate more accurate parameter estimates. The convergence properties of the proposed algorithm can be analyzed by means of the martingale convergence theorem. The methods in this paper can be extended to study identification problems of other linear systems [34] or nonlinear

References (43)

Cited by (57)

  • Moving horizon estimation for multirate systems with time-varying time-delays

    2019, Journal of the Franklin Institute
    Citation Excerpt :

    Parameter and state estimation has wide applications in system identification [1–3], controller design and other fields [4–7]. State estimation is one of the most important elements in development and implementation of advanced process control techniques [8–10]. The performance of state estimation has a significant effect on the performance of control.

  • A multi-innovation state and parameter estimation algorithm for a state space system with d-step state-delay

    2017, Signal Processing
    Citation Excerpt :

    On the basis of the work in [26,39], this paper uses the multi-innovation identification theory, extends the SG algorithm and derives a joint multi-innovation SG (MISG) parameter estimation and state estimation algorithm, where the parameter estimation uses the MISG algorithm and the state estimation uses the parameter estimates based observer. The methods in this paper can simultaneously estimate the parameters and states of the systems and differs from the previous methods which were based on the hierarchical identification principle [40]. This paper is organized as follows.

  • Identification methods for time-delay systems based on the redundant rules

    2017, Signal Processing
    Citation Excerpt :

    The general identification methods, such as the recursive least squares (RLS) algorithms [8–10], the stochastic gradient algorithms and the iterative algorithms [11–13], can be easily extended to systems with known time-delay. For example, Gu et al. presented an auxiliary model based least squares identification method for a state space model with a unit time-delay [14] and two iterative methods for a state space model with multi-state delays [15]. However, all the above works assumed that the time-delays are known in prior.

View all citing articles on Scopus

This work was supported by the National Natural Science Foundation of China (No. 61273194), the Natural Science Foundation of Jiangsu Province (China, BK2012549) and the PAPD of Jiangsu Higher Education Institutions.

View full text