2

Size: px
Start display at page:

Download "2"

Transcription

1 Topics in Model Based Control with Application to the Thermo Mechanical Pulping Process Dr. ing. thesis David Di Ruscio f Department of Engineering Cybernetics The Norwegian Institute of Technology 1993 June 27, 1994 Report W Department of Engineering Cybernetics The Norwegian Institute of Technology N-7034 Trondheim, Norway

2 2

3 Acknowledgement I would like to gratefully acknowledge the guidance and valuable discussions with my advisor, Professor Jens Glad Balchen at the Department of Engineering Cybernetics, the Norwegian Institute of Technology. I would also acknowledge the discussions with Alfred Holmberg at Norske Skog Teknikk. Additionally, Professor Rolf Henriksen at the Department of Engineering Cybernetics and Dr. Ing. Dag Ljungquist at Hydro Aluminium are acknowledged for their contributions to the work on realization of time series. This research has been sponsored by Norske Skog and the Royal Norwegian Council for Scientic and Industrial Research (NTNF). The nancial support is gratefully acknowledged.

4 ii ACKNOWLEDGEMENT

5 Contents Acknowledgement i Introduction 1 I REALIZATION OF TIME SERIES 3 1 State Space Model Realization from Input-Output Time Series Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : Problem denition and preliminaries : : : : : : : : : : : : : Method 1 : : : : : : : : : : : : : : : : : : : : : : : : : : : : Correlation analysis of the output y : : : : : : : : : Covariance analysis of the output y : : : : : : : : : : Cross-correlation analysis of y and u : : : : : : : : : Cross-covariance analysis of y and u : : : : : : : : : Determination of the noise covariance matrix : : : : Discussion : : : : : : : : : : : : : : : : : : : : : : : : Method 2 : : : : : : : : : : : : : : : : : : : : : : : : : : : : Numerical examples : : : : : : : : : : : : : : : : : : : : : : Concluding remarks : : : : : : : : : : : : : : : : : : : : : : References : : : : : : : : : : : : : : : : : : : : : : : : : : : : Appendix: Tables for the numerical examples : : : : : : : : 25

6 iv CONTENTS 2 A Method for the Identication of State Space Models Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : Problem denition and preliminaries : : : : : : : : : : : : : Main results : : : : : : : : : : : : : : : : : : : : : : : : : : : Result 1 : : : : : : : : : : : : : : : : : : : : : : : : : Result 2 : : : : : : : : : : : : : : : : : : : : : : : : : Discussion : : : : : : : : : : : : : : : : : : : : : : : : Determination of C and : : : : : : : : : : : : : : : : : : : Numerical examples : : : : : : : : : : : : : : : : : : : : : : Concluding remarks : : : : : : : : : : : : : : : : : : : : : : References : : : : : : : : : : : : : : : : : : : : : : : : : : : : Appendix: Tables for the numerical example : : : : : : : : 39 3 A Solution to the Problem of Constructing a State Space Model from Time Series Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : Problem denition and preliminaries : : : : : : : : : : : : : Results I : : : : : : : : : : : : : : : : : : : : : : : : : : : : : Results II : : : : : : : : : : : : : : : : : : : : : : : : : : : : Numerical examples : : : : : : : : : : : : : : : : : : : : : : Conclusions : : : : : : : : : : : : : : : : : : : : : : : : : : : References : : : : : : : : : : : : : : : : : : : : : : : : : : : : Appendix : : : : : : : : : : : : : : : : : : : : : : : : : : : : 54 II ANALYSIS AND DESIGN OF LQ SYSTEMS 55 4 A Note on a Necessary Condition for Optimality Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : Theory : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : Conclusion : : : : : : : : : : : : : : : : : : : : : : : : : : : 60

7 CONTENTS v 4.4 References : : : : : : : : : : : : : : : : : : : : : : : : : : : : 60 5 Maximal Imaginary Eigenvalues in Optimal Systems Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : Problem formulation : : : : : : : : : : : : : : : : : : : : : : Main results : : : : : : : : : : : : : : : : : : : : : : : : : : : Numerical examples : : : : : : : : : : : : : : : : : : : : : : Example 1 : : : : : : : : : : : : : : : : : : : : : : : : Example 2 : : : : : : : : : : : : : : : : : : : : : : : : Example 3 : : : : : : : : : : : : : : : : : : : : : : : : Concluding remarks : : : : : : : : : : : : : : : : : : : : : : References : : : : : : : : : : : : : : : : : : : : : : : : : : : : Appendix : : : : : : : : : : : : : : : : : : : : : : : : : : : : 75 6 On the Location of LQ-Optimal Closed-Loop Poles Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : Problem formulation : : : : : : : : : : : : : : : : : : : : : : Main results : : : : : : : : : : : : : : : : : : : : : : : : : : : Imaginary parts of the closed loop eigenvalues : : : : Magnitude of the closed loop eigenvalues : : : : : : : Real parts of the closed loop eigenvalues : : : : : : : Concluding remarks : : : : : : : : : : : : : : : : : : : : : : References : : : : : : : : : : : : : : : : : : : : : : : : : : : : Appendix : : : : : : : : : : : : : : : : : : : : : : : : : : : : 88 7 A Schur Method for Designing LQ-Optimal Systems with Prescribed Eigenvalues Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : Linear algebra review : : : : : : : : : : : : : : : : : : : : : Basic denition : : : : : : : : : : : : : : : : : : : : : : : : : Solution by block triangulization : : : : : : : : : : : : : : : 97

8 vi CONTENTS 7.5 The second order problem : : : : : : : : : : : : : : : : : : : The n-dimensional problem : : : : : : : : : : : : : : : : : : Numerical examples : : : : : : : : : : : : : : : : : : : : : : Example 1 : : : : : : : : : : : : : : : : : : : : : : : : Example 2 : : : : : : : : : : : : : : : : : : : : : : : : Example 3 : : : : : : : : : : : : : : : : : : : : : : : : Example 4 : : : : : : : : : : : : : : : : : : : : : : : : Concluding remarks : : : : : : : : : : : : : : : : : : : : : : References : : : : : : : : : : : : : : : : : : : : : : : : : : : : Appendix : : : : : : : : : : : : : : : : : : : : : : : : : : : : An Algorithm for Design of Decentralized Suboptimal Controllers with a Specied Structure Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : Problem denition : : : : : : : : : : : : : : : : : : : : : : : Solution algorithm : : : : : : : : : : : : : : : : : : : : : : : The output feedback design and the stabilization problem : Design of robust controllers : : : : : : : : : : : : : : : : : : Numerical examples : : : : : : : : : : : : : : : : : : : : : : Example 1, distillation column. : : : : : : : : : : : : Example 2, unstable plant : : : : : : : : : : : : : : : Example 3, robust controller : : : : : : : : : : : : : Concluding remarks : : : : : : : : : : : : : : : : : : : : : : References : : : : : : : : : : : : : : : : : : : : : : : : : : : : Appendix : : : : : : : : : : : : : : : : : : : : : : : : : : : : A Method for the Stabilization of Linear Feedback Systems Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : Problem denition and preliminaries : : : : : : : : : : : : : Main results : : : : : : : : : : : : : : : : : : : : : : : : : : : Concluding remarks : : : : : : : : : : : : : : : : : : : : : : 142

9 CONTENTS vii 9.5 References : : : : : : : : : : : : : : : : : : : : : : : : : : : : Measures for Stability Robustness in Linear Quadratic Systems Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : Problem formulation and preliminaries : : : : : : : : : : : : Robustness measures : : : : : : : : : : : : : : : : : : : : : : Results from time domain conditions : : : : : : : : : Results from frequency domain conditions : : : : : : Numerical examples : : : : : : : : : : : : : : : : : : : : : : Concluding remarks : : : : : : : : : : : : : : : : : : : : : : References : : : : : : : : : : : : : : : : : : : : : : : : : : : : Appendix : : : : : : : : : : : : : : : : : : : : : : : : : : : : 156 III MODEL BASED CONTROL OF TMP PROCESS The Thermo Mechanical Pulping Process Description of the TMP process : : : : : : : : : : : : : : : : Main process input variables : : : : : : : : : : : : : Main process measurements : : : : : : : : : : : : : : Problem denition : : : : : : : : : : : : : : : : : : : : : : : Control system design : : : : : : : : : : : : : : : : : Modeling the TMP process : : : : : : : : : : : : : : References : : : : : : : : : : : : : : : : : : : : : : : : : : : : Basic Rener Modeling Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : Modeling the screw feeding system : : : : : : : : : : : : : : Modeling the conical screw : : : : : : : : : : : : : : Modeling the cylindrical screw : : : : : : : : : : : : Summary of the chips feeding model : : : : : : : : : 180

10 viii CONTENTS 12.3 The basic conservation equations : : : : : : : : : : : : : : : Total conservation of mass : : : : : : : : : : : : : : : The consistency : : : : : : : : : : : : : : : : : : : : : Total conservation of energy : : : : : : : : : : : : : : Steady state analysis : : : : : : : : : : : : : : : : : : Summary of the basic rener model : : : : : : : : : : : : : Concluding remarks : : : : : : : : : : : : : : : : : : : : : : Appendix: Least squares equations for saturated steam : : A State Space Model for the TMP Process Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : Modeling of ber breakage : : : : : : : : : : : : : : : : : : : Two dimensional particle geometry : : : : : : : : : : One dimensional particle geometry : : : : : : : : : : Radial ber size distribution : : : : : : : : : : : : : : : : : : Model development : : : : : : : : : : : : : : : : : : : Boundary and initial conditions : : : : : : : : : : : : Simplied ber distribution models : : : : : : : : : : : : : : Static model without diusion : : : : : : : : : : : : Dynamic model without diusion : : : : : : : : : : : Determination of breakage rates from experiments : : : : : Examples : : : : : : : : : : : : : : : : : : : : : : : : : : : : Example 1 : : : : : : : : : : : : : : : : : : : : : : : : Example 2 : : : : : : : : : : : : : : : : : : : : : : : : Example 3 : : : : : : : : : : : : : : : : : : : : : : : : Concluding remarks : : : : : : : : : : : : : : : : : : : : : : References : : : : : : : : : : : : : : : : : : : : : : : : : : : : Appendix : : : : : : : : : : : : : : : : : : : : : : : : : : : : Power Consumption in the TMP Process 223

11 CONTENTS ix 14.1 Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : Modeling the power : : : : : : : : : : : : : : : : : : : : : : Basic equations : : : : : : : : : : : : : : : : : : : : : Shear stress equation : : : : : : : : : : : : : : : : : : Proposed relation for : : : : : : : : : : : : : : : : The rate of debration : : : : : : : : : : : : : : : : : Summary of the power model : : : : : : : : : : : : : : : : : The Power Model : : : : : : : : : : : : : : : : : : : : Simplied power models : : : : : : : : : : : : : : : : Concluding remarks : : : : : : : : : : : : : : : : : : : : : : References : : : : : : : : : : : : : : : : : : : : : : : : : : : : An Estimator for the TMP Process The process model and the measurements : : : : : : : : : : The measurements : : : : : : : : : : : : : : : : : : : The state and disturbance model : : : : : : : : : : : The estimator : : : : : : : : : : : : : : : : : : : : : : : : : : Concluding remarks : : : : : : : : : : : : : : : : : : : : : : A Model Based Control Strategy for the TMP Process Introduction : : : : : : : : : : : : : : : : : : : : : : : : : : : A system theoretic description of the model : : : : : : : : : Process model for control : : : : : : : : : : : : : : : Denition of model variables : : : : : : : : : : : : : Linearized model : : : : : : : : : : : : : : : : : : : : Controllability analysis : : : : : : : : : : : : : : : : A system theoretic description of the control : : : : : : : : Description of the control strategy : : : : : : : : : : Linear control system design : : : : : : : : : : : : : Discussion : : : : : : : : : : : : : : : : : : : : : : : : 261

12 x CONTENTS 16.4 References : : : : : : : : : : : : : : : : : : : : : : : : : : : : Appendix : : : : : : : : : : : : : : : : : : : : : : : : : : : : 264 References 267

13 Introduction This thesis is basically a collection of papers. These papers are listed in the reference list on page 267. The thesis is divided into the following three parts: Part I Part II Realization of time series. Analysis and design of LQ systems. Part III Model based control of the TMP process.

14 2 INTRODUCTION

15 Part I REALIZATION OF TIME SERIES

16

17 Chapter 1 State Space Model Realization from Input-Output Time Series 1 Abstract. Methods are presented that are based on known system input and output time series for discrete realization of linear, time invariant state space models on innovations form. New methods are compared with an existing method and found to give improved results. The methods work for systems which have nonminimum phase behavior from the input to the output time series. 1.1 Introduction Aoki (1987), (1990) has presented a method for the realization of state space linear discrete stochastic models on innovations form. This method has many interesting numerical properties. It is based on factorization of the Hankel matrix, constructed from covariance matrices of the output time series, by singular value decomposition. The states are presented relative to an "internally balanced" coordinate system, which has some interesting properties when dealing with model reduction. The denition of the "internally balanced" coordinate system is due to Moore (1981). See also Silverman and Bettayeb (1980), and Laub (1980). The above mentioned method has given very good results in the state space modeling of economic time series. However, it cannot handle the combined 1 This chapter is based on the paper Di Ruscio and Ljungquist (1992).

18 6 State Space Model Realization deterministic/stochastic problem, i.e. the problem of model realization of stochastic systems which have manipulable input variables. A modication of the above method to take input variables into account is presented in Aoki (1991), and Ostermark and Aoki (1992). Although this last method is reported to give promising results, a simulation study has pointed out that the method in some cases fails to give satisfactory results. The contribution of this paper is to present some modied methods which is shown by numerical simulations of known systems to give improved results. The rest of the paper is organized as follows. Section 1.2 presents the problem denitions. The main results are presented in Sections 1.3, 1.4 and 1.5. A method based on system analysis by correlation and covariance estimation is presented in Section 1.3. Algorithms based on instrumental variables are presented in Section 1.4. The methods are compared with numerical examples in Section 1.5, and some concluding remarks follow in Section Problem denition and preliminaries We assume that a system can be illustrated as shown in Figure 1.2. y 2 < m u? y - System - e Figure 1.1: Dynamic system with inputs and outputs is the system output which is measured, u 2 < r is the system input which can be manipulated, e 2 < m is an unknown innovations process of white noise, assumed to be serially uncorrelated, covariance stationary, with zero mean and E(e i e T i ) =. We also assume that the system can be described by the discrete, time invariant, linear innovations state space model x i+1 = Ax i + Bu i + Ce i (1.1) y i = Dx i + Eu i + e i (1.2)

19 1.3 Method 1 7 where i 0 is discrete time, ie. an integer, x 2 < n is the state vector with initial value x 0. A, B, C, D and E are constant matrices of appropriate dimensions, where (D; A) is an observable pair. Moreover, the transfer function from e to y, h s (z), is assumed to be invertible and h?1 s (z) is assumed to be stable. This means that h s (z) has all its zeros inside the unit circle. The problem investigated in this paper is to estimate the state space model matrices A, B, C, D, E and from known input output time series. The following notation is used throughout this paper: (A) The eigenvalues of a square matrix A. h d (z) The transfer function from u to y; D(Iz? A)?1 B + E. h d (1) The steady state gain of h d (z); D(I? A)?1 B + E. h d (0) The gain at time zero of h d (z);?da?1 B + E. p d The zeros of h d (z). ~h d (z) The transfer function from u to y when E 6= 0 is left out; D(Iz? A)?1 B. ~h d (1) The steady state gain of h ~ d (1); D(I? A)?1 B. ~p d The zeros of h ~ d (z). h s (z) The transfer function from e to y; D(Iz? A)?1 C + I m. I m m m identity matrix. h s (1) The steady state gain of h s (z); D(I? A)?1 C + I m. h s (0) The gain at time zero of h s (z);?da?1 C + I m. p s The zeros of h s (z). 1.3 Method 1 The algorithms presented in this section are based on system analysis by correlation and covariance estimation. The methods have some restrictions for the input sequence. However, the presentation is also valuable for the understanding of the general problem, ie. when there are no restrictions on the input type. For simplicity, E is assumed to be zero throughout this section. The algorithms presented are extensions of the method by Aoki (1987) Ch. 9 to take input variables into account. Simulation experiments show that the system matrices A and D should be estimated from correlation matrices E(y i+k yi T ) of the output time series, rather than from cross-correlation matrices E(y i+k u T i ) between the output and the input time series, provided the dynamic system is satised, ie. the

20 8 State Space Model Realization correlation matrices must satisfy Equations (1.6) and (1.7) presented in Section The justication for this statement is also explained as follows: Assume that process noise is present in the system. In this case the correlation matrices of the outputs reect the dynamics better than cross-correlation between outputs and inputs, because the excitations from the noise in addition to the input excitations are reected in the correlation matrices. This is also easily seen in the case when the input sequences are zero, and the system dynamics is only excited by process noise. In this case the cross-correlation matrices are zero, but the correlation matrices contain information about the system dynamics. Another fact is that the correlation and the cross-correlation matrices give the same information of system dynamics in the deterministic case, ie. when no process noise is present, provided the dynamic system given by Equations (1.6) and (1.7) is satised. However, the cross-correlation matrices between the input and output time series give important information about the input matrix B. This section is divided into six subsections. Sections and show how A, D and h d (1) can be computed. In addition some useful relations to the noise covariance matrix are presented. Determination of B is discussed in Sections and 1.3.4, while Section shows how C and can be computed. Finally, a discussion of the general case, ie. when there are no restrictions on the type of input, follows in Section Correlation analysis of the output y Dene the correlation and cross-correlation matrices k = E(y i+k y T i ) 2 < mm (1.3) S k = E(u i+k y T i ) 2 < rm (1.4) Z k = E(x i+k y T i ) 2 < nm (1.5) where k 0. These matrices are assumed to satisfy the following dynamic system, which is obtained by shifting time index with i =: i + k, post multiply with yi T in Equations (1.1) and (1.2) and take expectations where Z k+1 = AZ k + BS k 8 k 1 (1.6) k = DZ k (1.7) Z 1 = AZ 0 + BS 0 + C (1.8)

21 1.3 Method = DZ 0 + (1.9) Assume that the input is a step change. In this case the correlation S k+1 = S ns 8 k ns, where ns depends on the time instant the step change is applied to the system and on the number of inputs. For example, if r = 1 and the step is applied at time instant i, then S 0 = S 1 =, and we can choose ns = 0. Dene the matrices which satisfy the relation d p = k+p? k+p?1 (1.10) dz k+1 = Z k+1? Z k (1.11) d p = DA p?1 dz k+1 8 p = 1; 2; ; 2K k = ( 1 if ns 1 ns if ns > 1 (1.12) when S k+1 = S ns 8 k ns (1.13) The correlation dierence matrices d p 8 p = 1; ; 2K are sucient for the construction of the block Hankel matrix, K 2 < KmKm, and the shifted block Hankel matrix, A 2 < KmKm. The structure of these matrices are given by. 2 3 d 1 d K 6 K = = OC (1.14) d K d 2K?1 A = d 2 d K = OAC (1.15) d K+1 d 2K where O 2 < Kmn is the observability matrix for the pair (D; A) and C 2 < nkm is the controllability matrix for the pair (A; dz k+1 ). The order of the dynamic system, n, and the system matrices D and dz k+1 are estimated by factorization of the block Hankel matrix, K, into the product OC by singular value decomposition. The system matrix A is determined from the shifted block Hankel matrix by solving A = OAC. See Silverman and Bettayeb (1980), Moore (1981) and Aoki (1987) ch. 9.

22 10 State Space Model Realization We must ensure that K n to obtain a proper estimate. K = n is the theoretical minimum in the deterministic case, ie. when e i = 0 in Equations (1.1) and (1.2). More system information can be deduced at this stage. We will rst nd an equation for h d (1), and then an equation for the gain of the stochastic part of the system at time zero, h s (0). Finally the problem of determining h s (1) is illustrated. These results are presented in the following three remarks. Remark 1 The formula, relating h d (1) to the correlation matrices, is determined as follows dz k+1 =?(I? A)Z k + BS k (1.16) Pre-multiplication with D(I? A)?1 and substitution for k give k + D(I? A)?1 dz k+1 = D(I? A)?1 BS k (1.17) Equation (1.17) shows that h d (1) is identiable if S k is invertible, because k, A, D and dz k+1 are known. Note that the condition for S k always is satised for single input single output systems, provided the input is a step at time instant i. However, this condition can be removed as shown in Section 1.3.3, Equations (1.43) and (1.44). See also Sections and for the computation of h d (0). This section has not presented any information which can be directly used to estimate B. The B matrix should be estimated from cross-correlation or cross-covariance analysis of y and u. This will be shown in Sections and B can therefore be considered known in Remarks 2 and 3. Remark 2 In order to simplify the presentation we will assume that A, D and dz 2 are determined. The nesting formula from the known matrix dz k+1 to dz 2 is given by X k?1 dz 2 = A?(k?1) dz k+1? A?j B(S j? S j?1 ) (1.18) j=1 An expression for h s (0) is derived as follows. We have dz 2 = AZ 1 + BS 1? (AZ 0 + BS 0 + C) (1.19) Pre-multiplication by DA?1 and substituting for 1 and 0 give 0? 1 + DA?1 dz 2? DA?1 B(S 1? S 0 ) = (?DA?1 C + I) (1.20)

23 1.3 Method 1 11 Equation (1.20) shows that h s (0) times the noise correlation matrix is i- dentiable. Remark 3 At this stage we have not presented any information that can be used to derive h s (1). The following equations will illustrate the problem. For k = 0 we have (I? A)?1 dz 1 =?Z 0 + (I? A)?1 BS 0 + (I? A)?1 C (1.21) Pre-multiplication by D and substitution for 0 give 0 + D(I? A)?1 Z 1 = D(I? A)?1 Z 0 + D(I? A)?1 BS 0 +D(I? A)?1 C + (1.22) where Z 1 is related to known information by Z 1 =?(I? A)?1 dz 2 + (I? A)?1 BS 1 (1.23) Equation (1.22) shows that the steady state gain for the stochastic part only can be determined if the initial statistics Z 0 is known. Note that Z 0 = X 0 D T where X 0 = E(x i x T i ). X 0 and a relation for the steady state gain must be developed from X k = E(x i+k x T i+k ), see Section Covariance analysis of the output y Dene the covariance and cross-covariance matrices k = E[(y i+k? y)(y i? y) T ] 2 < mm (1.24) S k = E[(u i+k? u)(y i? y) T ] 2 < rm (1.25) Z k = E[(x i+k? x)(y i? y) T ] 2 < nm (1.26) where k 0, y, u and x are the mean values of the time series y i, u i and x i, respectively. By assumption, the innovations process e i has zero mean, and hence, =. Assume that the covariance matrices of the input satisfy S k = 0 8 k ns (1.27) where ns 1. The covariance matrices are assumed to satisfy the dynamic system, ie. Z 1 = A Z0 + B S0 + C (1.28) 0 = D Z0 + (1.29) Z k = A Zk?1 + B Sk?1 2 ns k (1.30) p+k?1 = DA p?1 Zk p = 1; :::; 2K k = ns (1.31)

24 12 State Space Model Realization The matrices A, D and Zk can be estimated from the block Hankel matrix and the shifted block Hankel matrix constructed from the sequence given by Equation (1.31), in the same way as described in Section 1.3.1, Equation (1.15). The unused covariance matrices, k 8 k = 0; 1; :::; ns? 1, can be used to determine B in the same way as shown in Section if Sk is invertible. The method is found to be very robust in the detection of A and D and the gain at time zero. The method can detect nonminimum phase behavior in the deterministic part of the system when the signal to noise ratio is large, but will have problems when the data are noisy, ie. when the signal to noise ratio is small. The covariance matrices will probably not satisfy the dynamic system in this case. Here we will discuss the case when ns = 1. Note that dz 1 6= Z 1, where dz 1 is dened in Section We have the following relation?da?1 Z DA?1 B S0 = (?DA?1 C + I) (1.32) The last equation is important because it gives the h s (0) times the noise covariance matrix. When the noise is unknown, only the product is identi- able Cross-correlation analysis of y and u Dene the correlation and cross-correlation matrices s k = E(y i+k u T i ) 2 < mr (1.33) U k = E(u i+k u T i ) 2 < rr (1.34) z k = E(x i+k u T i ) 2 < nr (1.35) where k 0. The matrices, Equations (1.33)-(1.35), are assumed to satisfy the dynamic system which is obtained by shifting time index with i =: i+k, post-multiply with u T i in Equations (1.1) and (1.2) and take expectations z k+1 = Az k + BU k 8 k 0 (1.36) s k = Dz k (1.37) Assume that the input is a step change. In this case the correlation U k+1 = U ns 8 k ns, where ns depends on the time instant the step change is applied to the system and on the number of inputs. For example, if r = 1 and the step is applied at time instant i, then, U 0 = U 1 =, and ns = 0.

25 1.3 Method 1 13 Dene the matrices which satisfy the dynamic system when ds p = s k+p? s k+p?1 (1.38) dz k+1 = z k+1? z k (1.39) ds p = DA p?1 dz k+1 8 p = 1; 2; ; 2K k = ns (1.40) U k+1 = U ns 8 k ns (1.41) The matrices A, D and dz k+1 can be determined from the block Hankel matrix and the shifted block Hankel matrix constructed from Equation (1.40). See Section However, when process noise is present, A and D should be determined as described in Section Assume that A and D are known. dz k+1 is then determined by dz k+1 = (O T O)?1 O T ds 1 ds 2. ds n where O is the observability matrix for the system (D; A). A formula, relating the steady state gain to the data is (1.42) s k + D(I? A)?1 dz k+1 = D(I? A)?1 BU k (1.43) which gives the steady state gain if U k is invertible. Nesting back to k = 0 gives s 0 + D(I? A)?1 dz 1 = D(I? A)?1 BU 0 (1.44) A fundamental condition for h d (1) to be identiable is that U 0 is invertible. Note that the nesting formula from the known matrix dz k+1 to dz 1 is given by dz 1 = A?k dz k+1? kx j=1 A?j B(U j? U j?1 ) (1.45) An expression which can be used to compute h d (1) and a solution for B will be presented in the next section.

26 14 State Space Model Realization Cross-covariance analysis of y and u Dene the covariance and cross-covariance matrices s k = E[(y i+k? y)(u i? u) T ] 2 < mr (1.46) U k = E[(u i+k? u)(u i? u) T ] 2 < rr (1.47) z k = E[(x i+k? x)(u i? u) T ] 2 < nr (1.48) Assume that the covariance matrices of the input satisfy U k = 0 8 k ns (1.49) where ns 1. The matrices, Equations (1.46)-(1.48), are assumed to satisfy and z k = Az k?1 + B Uk?1 1 ns k (1.50) s k?1 = Dz k?1 (1.51) s p+k?1 = DA p?1 z k p = 1; :::; 2K k = ns (1.52) As usual, A, D and z k can be estimated from Hankel matrix factorization of the sequence given by Equation (1.52). See Section Assume that A and D are known, then we must determine z k in the same coordinate system. We have from Equation (1.52) z k = (O T O)?1 O T s k s k+1. s n+k? (1.53) where O is the observability matrix for the system (D; A). z k can now be substituted into the unused cross-covariance matrices. We have s i = DA?(k?i) z k? kx j=1 DA?(k?i?j) B Uj?1 0 i ns? 1 (1.54) To illustrate the value of this formula, assume ns = 3, then ~s 0 ~s 1 5 =? ~s D DA DA A?3 B U2 (1.55)

27 1.3 Method 1 15 which can be solved for B as 2 3 B =?A 3 (O T O)?1 O T 6 4 ~s 0 7 ~s 1 5 U?1 2 (1.56) ~s 2 where O is the observability matrix for the pair (D; A), and ~s 2 = s 2? DA?1 z 3 (1.57) ~s 1 = s 1? DA?2 z 3? ~s 2U?1 2 U 1 (1.58) ~s 0 = s 0? DA?3 z 3? ~s 1 U?1 2 U 1? ~s 2U?1 2 U 0 (1.59) This is an identication method which can be used if an "arbitrary" input signal is applied to the process from time zero to time instant i, and then switch to a signal satisfying Equation (1.49), say a step change. It is also possible to do two experiments on the process. The rst experiment is used to determine A and D, the second to determine B. Note that the cross-correlation dierence dz k dened in Section is related to the covariance matrix z k in the single input case by z k = dz k (1.60) Assume that ns = 1, in this case z 1 is known. An equation for h d (0) is then given by (s 0? DA?1 z 1 ) U?1 0 =?DA?1 B (1.61) Some alternatives to Equation (1.56), for the determination of B are discussed below. Case 1 Combining Equations (1.44) and (1.61) gives the following equation # B = ( O ~ T O) ~?1 O ~ T (1.62) " (s0 + D(I? A)?1 dz 1 )U?1 0 (s 0? DA?1 z 1 ) U?1 0 which denes B for the special case when O ~ T O ~ is invertible, where O ~ is an equivalent observability matrix for the control input matrix B, given by ~O = " # D(I? A)?1?DA?1 (1.63) This means that B is determined to satisfy the transfer function in steady state and at time zero. Moreover, h d (0) is needed to detect non-minimum phase behavior.

28 16 State Space Model Realization We have shown that the steady state gain and the gain at time zero are identiable without any assumptions about initial statistics, z 0 and z 0. Case 2 It can be shown that z 1 = z 0 when N is large and x 0 =0. In this case, an equation for B is simply B = (I? A)dz 1 U?1 0 (1.64) which denes B if U0 is non-singular. This method is not recommended when noise is present. Case 3 The solution for B in the case when D is non-singular is simply z 0 = D?1 s 0 (1.65) B = (z 1? AD?1 s 0 ) U?1 0 (1.66) Case 4 The matrix B can be constructed from the estimate of h d (1): x i+1 = Ax i + Bu i (1.67) y i = Dx i (1.68) B = (I? A)D T (DD T )?1 h d (1) (1.69) This model can be viewed as an approximately dynamic model with an exact steady state gain. The model has the same spectrum of eigenvalues as the underlying process, but may have a dierent phase behavior, ie. the model is approximately dynamic. This is probably a good approach when the process is known to be minimum phase and the data are noisy, because in this case the estimation of A, D and h d (1) is robust. Case 5 Assume that the input u i is white noise. In this case z 0 = 0, and A, B and D satisfy the normalized sequence s 0U?1 0 = E (1.70) s ku?1 0 = DA k?1 B 8 k 1 (1.71) The matrices A, B and D can be determined by Hankel matrix factorization or determined from a possibly known observability matrix for the pair (D; A).

29 1.3 Method Determination of the noise covariance matrix The model, Equations (1.1) and (1.2), can be separated as and x s i+1 = Ax s i + Ce i x s 0 = x 0 (1.72) y s i = Dx s i + e i (1.73) x d i+1 = Ax d i + Bu i x d 0 = 0 (1.74) y d i = Dx d i (1.75) y i = y d i + y s i (1.76) Once the the matrices A, B, and D are determined, the deterministic part of the model, Equations (1.74) and (1.75), can be simulated for yi d and ys i computed from (1.76). The method by Aoki (1987) can now be applied to determine and C. From weak stationarity, X s 0 = E(x s i xst i ) = E(x s i+1 xst i+1 ), the following Riccati type equation can be solved for X0. s X s 0 = AX s 0 AT + (Z s 1? AXs 0 DT )( s 0? DXs 0 DT )?1 (Z s 1? AXs 0 DT ) T (1.77) This equation can be solved iteratively. X s 0 = 0 is a good choice as initial value for the iterations. The Schur method by Laub (1979) is also recommended. For very ill-conditioned systems, the solution from the Schur method should be used as the initial value for the iterative method, to rene the solution further. C and are determined from X s 0 and Z s 1 as = s 0? DX s 0D T (1.78) C = (Z s 1? AXs 0 DT )?1 (1.79) Note that Z s 1 satises the sequence s k = DAk?1 Z s 1 k 1, where s k = E(y s i+k (ys i )T ). When A and D are known, Z s 1 should be determined as shown in Equation (1.53) Discussion We have discussed four methods in Sections to These can be used to estimate A and D by factorization of the Hankel matrix determined from the sequence given by Equations (1.12), (1.31), (1.40) or (1.52). These methods work well if the correlation or covariance matrices for the input

30 18 State Space Model Realization satisfy, or approximately satisfy, the conditions given by Equations (1.13), (1.27), (1.41) or (1.49). However, the following combination of the above mentioned equations is found to give satisfactory results in the general case, ie. when there are no restrictions to the input type: Dene dh k = s k U?1 0? k S?1 0 8 k = 1; :::; 2K (1.80) where s k, U 0, k and S 0 are given by Equations (1.33), (1.34), (1.3) and (1.4). We have assumed that m = r, and that U 0 and S 0 are invertible. The matrices dh k 8 k = 1; :::; 2K are sucient for the construction of the block Hankel matrix and the shifted block Hankel matrix, in the same way as illustrated in Section 1.3.1, Equation (1.15). The order of the dynamic system, n, and the system matrices A and D are estimated from the block Hankel matrix and the shifted Hankel matrix, see Section No major dierences are found if covariance matrices are used in place of correlation matrices in Equation (1.80), ie. by using d Hk = s k U?1 0? k S?1 0 8 k = 1; :::; 2K (1.81) for the construction of the block Hankel matrix and the shifted block Hankel matrix. s k, U0, k and S0 are given by Equations (1.46), (1.47), (1.24) and (1.25). The matrix B can be estimated as illustrated in Sections and However, a more interesting approach seems to be to combine this method for the determination of A and D with the instrumental variable method, presented in Section 1.4. The idea is to use the instruments to estimate the initial value z 0, Equation (1.47), and then, with A and D known, solve Equations (1.50) and (1.51) for B. This procedure is under investigation. 1.4 Method 2 A modication of Aoki's original algorithm (Aoki, 1987) to take input (exogenous) variables into account, is outlined in Aoki, This algorithm, which will be referred to as Algorithm A1, can be used to generate a statespace representation for input-output data from an underlying process. The basic idea of this algorithm is to construct a set of Hankel matrices from auto-covariance matrices and cross-covariance matrices for the input-output data and utilize singular value decomposition. The input variables are incorporated using them as a part of the instruments as will be claried below.

31 1.4 Method 2 19 Although Algorithm A1 is reported to give promising results ( Ostermark and Aoki, 1992), a simulation study has pointed out that the algorithm fails to give satisfactory results when the input variables are changing rapidly. However, the algorithm can easily be modied to give improved results. Two alternative algorithms are briey outlined below. (1991), we introduce the notation y? i?1 = " yi?1 y? i?2 # Following Aoki (1.82) which denes y? i?1 recursively. Denote the covariance matrix of this vector by R? = cov(y? i?1 ; y? ) and dene the matrix = i?1 cov(x i; y? ). i?1 These matrices are constant if weak stationarity of the data is assumed. The state vector of the lter is then given by z i = E(x i jy? i?1 ) (1.83) z i = R?1? y? i?1 (1.84) The state vector z cannot be computed from Equation (1.84) since the time series of x i is unknown. However, by using z i = 2 R??1 y? i?1 as instruments, for some 2, Algorithm A1 can be developed as described in Aoki (1991) Section 8 and Ostermark and Aoki (1992) Section 2.1. Remark: The reason why the notation cov( ; ) is used instead of the expectation operator E( ) is twofold. First, using the cov operator will simplify the notation. Second, the algorithms presented in this section will work if either of the denitions cov(y i ; u i ) = E[y i u T i ] or cov(y i; u i ) = E[(y i? y i )(u i? u i ) T ] is used. Strictly speaking the second denition is the correct one and it results in a better estimate of the model order, although errors may be introduced in the estimated zeros and steady-state gain. A modied version of A1 is obtained by using the instruments which corresponds to the lter z i = 2 R?1? y? i (1.85) z i = E(x i jy? i ) (1.86) The resulting algorithm, denoted A2, is summarized in Table 1.1 where the following notation is used:

32 20 State Space Model Realization y + i y? i = [y 1;i y 2;i : : : y m;ijy 1;i+1 y 2;i+1 : : : y m;i+1j : : : jy 1;i+K y 2;i+K : : : y m;i+k] T = [y 1;i y 2;i : : : y m;ijy 1;i?1 y 2;i?1 : : : y m;i?1j : : : jy 1;i?K+2 y 2;i?K+2 : : : y m;i?k+2] T y? i?1 u? i = [y1;i?1 y2;i?1 : : : ym;i?1jy1;i?2 y2;i?2 : : : ym;i?2j : : : jy1;i?k+1 y2;i?k+1 : : : ym;i?k+1]t = [u 1;i u 2;i : : : u r;iju 1;i?1 u 2;i?1 : : : u r;i?1j : : : ju 1;i?L+2 u 2;i?L+2 : : : u r;i?l+2] T The integers K and L are cuto factors for the output and input time series respectively. Moreover, y k;i denotes element number k of y at the time instant i. Table 1.1: The Algorithm A2. Step 1. 1 = cov(u i ; y? ) i?1 2 <rkm H = cov(y i + ; y? ) i?1 2 <KmKm H u = cov(y i + ; u? i ) 2 <KmLr R? = cov(y? i ; y? i ) 2 <KmKm Y = cov(y? i ; u? i ) 2 <KmLr U = cov(u i ; u? i ) 2 <rlr U + = U T (UU T )?1 2 < Lrr Step 2. O 2 2 = (H? H u U + 1 )(I Km? R?1? Y U + 1 )?1 2 < KmKm Step 3. Compute the singular value decomposition of O 2 2 = UV T and assign O 2 = U 1=2 2 < Kmn, 2 = 1=2 V T 2 < nkm where n is the chosen model order. Step 4. O 1 = (H u? O 2 2 R?1? Y )U + 2 < Kmr Step 5. ^E = [Im 0 : : : 0]O 1 2 < mr ^D = [I m 0 : : : 0]O 2 2 < mn Step 6. ^z i = 2 R?1? y? i ^e i = y i? ^D^zi? ^Eui [ ^A ^B] = [cov(^zi+1 ; ^z i ) cov(^z i+1 ; u i )] ^C = cov(^z i+1? ^A^zi? ^Bui ; ^e i )cov(^e i ; ^e i )?1 " #?1 cov(^zi ; ^z i ) cov(^z i ; u i ) cov(u i ; ^z i ) cov(u i ; u i ) In Table 1.1, 0 denotes a matrix of zeros with appropriate dimensions. = cov(e i ; e i ) can be estimated in Step 6 of Algorithm A2. Simulations show that the dierence between the responses from the underlying process and the generated model is signicantly reduced by extending Step 6 by

33 1.5 Numerical examples 21 the following iteration: where superscript stable. ^z (2) = (2) i+1 ^A^z i + ^Bui + ^C^ei ^e (2) (2) i = y i? ^D^z i? ^Eui (1.87) ^ = cov(^e (2) i ; ^e (2) i ) ^C (2) = cov(^z (2) i+1? (2) ^A^z i? ^Bui ; ^e (2) i ) ^?1 (2) denotes iterated quatities and ^A is assumed to be In many dynamical systems the E-matrix in Equation (1.2) will be zero. If E = 0 is known, A2 can be simplied to an algorithm denoted A3. The simplications which result in Algorithm A3 are summarized in Table 1.2. Table 1.2: Algorithm A3 described by simplications of A2. In Step 1 only H and R? need to be computed. Step 2 is replaced by O 2 2 = H. Step 4 is omitted. Step 5 is replaced by ^E = 0 2 < mr, ^D = [Im 0 : : : 0]O 2 2 < mn. Steps 3 and 6 are kept unchanged. 1.5 Numerical examples Results from a simulation study of two examples are presented and discussed in this section. Both examples are single-input single-output (SISO). SISO systems are chosen to simplify the presentation of the results, and the discussed algorithms can be directly applied to multiple-input multipleoutput systems. The rst example is of the second order, and the transfer function from e to y, h s (z) has both zeros outside the unit circle. This nonminimum phase phenomenon cannot be estimated and is the reason for the assumptions made on h s (z) in the problem denition in Section 1.2. The second example is of the third order, and the transfer function from u to y has one zero outside the unit circle. The system is chosen to illustrate that this nonminimum phase phenomenon can be estimated. The following input sequences were tried out in the simulations.

34 22 State Space Model Realization u 1 A step change from 0 to 2 u 2 u i = 0:2( sin(i=25) + sin(i=10) + sin(i=5) + sin(i) ) u 3 A white noise sequence, E(u i u T k ) = 1:0 ik Example Ex1 The model, Equations (1.1) and (1.2), is simulated with dierent values of the noise covariance matrix. The following model matrices are specied. " # " # " # 1:5 1:0 2:0 0:5 A = B = C = (1.88)?0:7 0?1:3 1:5 h i D = 1 0 E = 0 The simulation results obtained by Method 1 with 4000 samples are shown in Table 1.3. For the input type u 1, applied at time instant i = 1, A and D were estimated from Equation (1.12), the steady state gain for the deterministic part, ^h d (1), was estimated from Equation (1.17) and B was estimated from Equation (1.62). For the input type u 3, A, B and D were estimated from Equation (1.71). Finally, C and were determined as shown in Section The simulation results when Algorithms A1, A2 and A3 (Method 2) were applied to 201 samples with the mean value subtracted are summarized in Tables 1.5, 1.6 and 1.7, where the step change for u 1 was applied at time instant i = 20. In Algorithms A2 and A3 Step 6 was extended by the iteration in Equation (1.87). Only 201 samples were used to demonstrate that satisfactory results can be obtained with such a low number of datapoints. It should be noted that the results were not signicantly improved if the number of samples was increased. Some important properties of the underlying system are summarized in the last row of the tables. Example Ex2 The model, Equations (1.1) and (1.2), is simulated with dierent values on the noise covariance matrix. The following model matrices are specied :8 0:3 0:1 6 A = 4 7?0:3 0:8 0:1 5 6 B = C = 4 0:1 7 0:6 5 (1.89) 0 0 0: h i D = 3 0?0:6 E = 0 The simulation results obtained by Method 1 with 4000 samples are shown

35 1.5 Numerical examples 23 in Table 1.4. A and D were estimated from Equation (1.52) and B from Equation (1.56). The step change was applied at time instant i = 3. Table 1.8 summarizes some simulation results from Algorithms A1, A2 and A3 based on 201 samples with the mean value subtracted. Only the results obtained for the input sequence which gave the best results (u 1 ) for all the algorithms are given. In Table 1.8 \No" denotes the algorithm which was used in the simulation corresponding to that row. Discussion The simulation results show that nonminimum phase phenomena in h s (z) cannot be estimated. For h d (z), however, zeros outside the unit circle can be estimated. This is in agreement with theoretical considerations. Tables 1.3 and 1.4 show that Method 1 works very well when the input is either a step change or approximately a white noise sequence. Moreover, a simulation study (not presented here) has shown that Method 1 gives satisfactory results even when the number of datapoints is low (200 samples for SISO systems) and that Method 1 gives results that are very close to those obtained by Method 2 for a general input sequence when A and D are generated according to Equation (1.80). The following conclusions can be made regarding Method 2 based on Tables 1.5 to 1.8: The realizations obtained by Algorithm A1 have eigenvalues that are satisfactory close to that of the true system for all the simulations considered. However, a large deviation in ^E will frequently result in poor estimates of the zeros and the steady-state gain of the transfer function from the inputs to the measurements. This is especially so when the inputs are changing rapidly, see the results obtained when the input is approximately white noise (measured) in Table 1.5. The problem connected to a bad ^E estimate is signicantly reduced when Algorithm A2 is applied. More specically A2 gives satisfactory results also when the input is white noise. All the algorithms give a rather large deviation in the steady-state gain when the input is white noise. This is due to the well-known fact that the slow dynamics are poorly identiable for such an input, so the result is not specic for the algorithms under consideration. The deterministic part of the realized models is satisfactory even under very small signal to noise ratios (0.043 when u 2 is applied to Ex1

36 24 State Space Model Realization with = 1:0). However, further research is needed to modify the algorithms to give improved estimates of ^C and ^. This can probably be done by using a strategy that is more similar to that used by Aoki (1987) for systems without input (exogenous) variables. Algorithm A3 does not seem to be able to estimate the nonminimum phase in Example Ex2. It should be noted that the nonminimum phase in the realizations obtained by A1 and A2 is due to the estimate ^E 6= 0, compare the ~p d and p d columns in Table 1.8. Algorithm A2 seems to have better overall performance for the two examples than A1 and A Concluding remarks The method presented in Section 1.3 is found to be very robust in the determination of A, D, and the steady state gain for the deterministic part of the system. It gives an exact result in the deterministic case, provided the correlation or covariance matrices are exactly determined, and it works for systems which have nonminimum phase behavior from the input to the output. An important feature is that the method is constructed to give exact results when certain restrictions on the input sequence are satised. Moreover, the concept can be modied to obtain an algorithm which is general with respect to the input sequence. The method presented in Section 1.4 is also general with respect to the input sequence. These algorithms do not have the property of giving an exact result in the deterministic case, and further research is needed to improve the estimated stochastic part of the model. Both methods, presented in Sections 1.2 and 1.4, are found to give satisfactory results even when the number of datapoints is low (200 samples for SISO systems). Method 1 is found to give better results compared to method 2 in the case that the input is a step change. Therefore, Method 1 should be chosen in the case that the input is a step change and Method 2 should be chosen in the case of other input signals, irrespective of the number of datapoints. Although satisfactory realizations were obtained by the methods presented in this paper, it is believed that the presented results can be combined and developed further to obtain a general algorithm with even better overall performance.

37 1.7 References References Aoki M. (1987). State Space Modeling of Time Series. Springer-Verlag Berlin, Heidelberg. Aoki M. (1990). State Space Modeling of Time Series. Second, Revised and Enlarged Edition. Springer-Verlag Berlin, Heidelberg. Aoki M. (1991). Two Complimentary Representations of Multiple Time Series in State Space Innovation Forms. Invited paper presented at Seminar on Recent Advances on Time Series Analysis and their Impact on Economic Forecasting, Madrid, Spain, December Laub, A. J. (1979). A Schur method for solving algebraic Riccati equations. IEEE Trans. on Automatic Control, Vol. AC-24, pp Laub, A. J. (1980). Computation of "balancing" transformations. Proc. JACC, San Francisco, FA8-E. Moore, B. C. (1981). Principal Component Analysis in Linear Systems: Controllability, Observability, and Model Reduction. IEEE Trans. on Automatic Control, Vol. AC-26, pp Silverman L. M. and M. Bettayeb (1980). Optimal approximation of linear systems. Proc. JACC, San Francisco, FA8-A. Ostermark, R. and M. Aoki (1992) Time Series Evidence of Impacts of The U.S. Economy on The Scandinavian Economy (by State Space Modeling). IFAC workshop on economic time series analysis and system identication, Vienna July Appendix: Tables for the numerical examples

38 26 State Space Model Realization Table 1.3: Results obtained by Method 1 on Ex1. System Algorithm Model Realization u K ^h d (1) ( ^A) ~p d hd ~ (1) p s h s(1) ^ u i u i 0.644i u i 0.652i u i 0.662i u i u i 0.595i u i 0.421i True System i 1.396i Table 1.4: Results obtained by Method 1 on Ex2. System Algorithm Model Realization u K ^h d (1) ( ^A) ~p d ~ hd (1) p s h s(1) ^ u , i True System 0.850, i

In: Proc. BENELEARN-98, 8th Belgian-Dutch Conference on Machine Learning, pp 9-46, 998 Linear Quadratic Regulation using Reinforcement Learning Stephan ten Hagen? and Ben Krose Department of Mathematics,

More information

SUBSPACE SYSTEM IDENTIFICATION Theory and applications

SUBSPACE SYSTEM IDENTIFICATION Theory and applications SUBSPACE SYSTEM IDENTIFICATION Theory and applications Lecture notes Dr. ing. David Di Ruscio Telemark Institute of Technology Email: david.di.ruscio@hit.no Porsgrunn, Norway January 1995 6th edition December

More information

Chapter 7 Interconnected Systems and Feedback: Well-Posedness, Stability, and Performance 7. Introduction Feedback control is a powerful approach to o

Chapter 7 Interconnected Systems and Feedback: Well-Posedness, Stability, and Performance 7. Introduction Feedback control is a powerful approach to o Lectures on Dynamic Systems and Control Mohammed Dahleh Munther A. Dahleh George Verghese Department of Electrical Engineering and Computer Science Massachuasetts Institute of Technology c Chapter 7 Interconnected

More information

Closed and Open Loop Subspace System Identification of the Kalman Filter

Closed and Open Loop Subspace System Identification of the Kalman Filter Modeling, Identification and Control, Vol 30, No 2, 2009, pp 71 86, ISSN 1890 1328 Closed and Open Loop Subspace System Identification of the Kalman Filter David Di Ruscio Telemark University College,

More information

only nite eigenvalues. This is an extension of earlier results from [2]. Then we concentrate on the Riccati equation appearing in H 2 and linear quadr

only nite eigenvalues. This is an extension of earlier results from [2]. Then we concentrate on the Riccati equation appearing in H 2 and linear quadr The discrete algebraic Riccati equation and linear matrix inequality nton. Stoorvogel y Department of Mathematics and Computing Science Eindhoven Univ. of Technology P.O. ox 53, 56 M Eindhoven The Netherlands

More information

Notes on Time Series Modeling

Notes on Time Series Modeling Notes on Time Series Modeling Garey Ramey University of California, San Diego January 17 1 Stationary processes De nition A stochastic process is any set of random variables y t indexed by t T : fy t g

More information

An LQ R weight selection approach to the discrete generalized H 2 control problem

An LQ R weight selection approach to the discrete generalized H 2 control problem INT. J. CONTROL, 1998, VOL. 71, NO. 1, 93± 11 An LQ R weight selection approach to the discrete generalized H 2 control problem D. A. WILSON², M. A. NEKOUI² and G. D. HALIKIAS² It is known that a generalized

More information

MODEL PREDICTIVE CONTROL and optimization

MODEL PREDICTIVE CONTROL and optimization MODEL PREDICTIVE CONTROL and optimization Lecture notes Model Predictive Control PhD., Associate professor David Di Ruscio System and Control Engineering Department of Technology Telemark University College

More information

and the nite horizon cost index with the nite terminal weighting matrix F > : N?1 X J(z r ; u; w) = [z(n)? z r (N)] T F [z(n)? z r (N)] + t= [kz? z r

and the nite horizon cost index with the nite terminal weighting matrix F > : N?1 X J(z r ; u; w) = [z(n)? z r (N)] T F [z(n)? z r (N)] + t= [kz? z r Intervalwise Receding Horizon H 1 -Tracking Control for Discrete Linear Periodic Systems Ki Baek Kim, Jae-Won Lee, Young Il. Lee, and Wook Hyun Kwon School of Electrical Engineering Seoul National University,

More information

1 Introduction 198; Dugard et al, 198; Dugard et al, 198) A delay matrix in such a lower triangular form is called an interactor matrix, and almost co

1 Introduction 198; Dugard et al, 198; Dugard et al, 198) A delay matrix in such a lower triangular form is called an interactor matrix, and almost co Multivariable Receding-Horizon Predictive Control for Adaptive Applications Tae-Woong Yoon and C M Chow y Department of Electrical Engineering, Korea University 1, -a, Anam-dong, Sungbu-u, Seoul 1-1, Korea

More information

RECURSIVE SUBSPACE IDENTIFICATION IN THE LEAST SQUARES FRAMEWORK

RECURSIVE SUBSPACE IDENTIFICATION IN THE LEAST SQUARES FRAMEWORK RECURSIVE SUBSPACE IDENTIFICATION IN THE LEAST SQUARES FRAMEWORK TRNKA PAVEL AND HAVLENA VLADIMÍR Dept of Control Engineering, Czech Technical University, Technická 2, 166 27 Praha, Czech Republic mail:

More information

Problem Description The problem we consider is stabilization of a single-input multiple-state system with simultaneous magnitude and rate saturations,

Problem Description The problem we consider is stabilization of a single-input multiple-state system with simultaneous magnitude and rate saturations, SEMI-GLOBAL RESULTS ON STABILIZATION OF LINEAR SYSTEMS WITH INPUT RATE AND MAGNITUDE SATURATIONS Trygve Lauvdal and Thor I. Fossen y Norwegian University of Science and Technology, N-7 Trondheim, NORWAY.

More information

Unbiased minimum variance estimation for systems with unknown exogenous inputs

Unbiased minimum variance estimation for systems with unknown exogenous inputs Unbiased minimum variance estimation for systems with unknown exogenous inputs Mohamed Darouach, Michel Zasadzinski To cite this version: Mohamed Darouach, Michel Zasadzinski. Unbiased minimum variance

More information

BUMPLESS SWITCHING CONTROLLERS. William A. Wolovich and Alan B. Arehart 1. December 27, Abstract

BUMPLESS SWITCHING CONTROLLERS. William A. Wolovich and Alan B. Arehart 1. December 27, Abstract BUMPLESS SWITCHING CONTROLLERS William A. Wolovich and Alan B. Arehart 1 December 7, 1995 Abstract This paper outlines the design of bumpless switching controllers that can be used to stabilize MIMO plants

More information

A MULTIGRID ALGORITHM FOR. Richard E. Ewing and Jian Shen. Institute for Scientic Computation. Texas A&M University. College Station, Texas SUMMARY

A MULTIGRID ALGORITHM FOR. Richard E. Ewing and Jian Shen. Institute for Scientic Computation. Texas A&M University. College Station, Texas SUMMARY A MULTIGRID ALGORITHM FOR THE CELL-CENTERED FINITE DIFFERENCE SCHEME Richard E. Ewing and Jian Shen Institute for Scientic Computation Texas A&M University College Station, Texas SUMMARY In this article,

More information

EL2520 Control Theory and Practice

EL2520 Control Theory and Practice EL2520 Control Theory and Practice Lecture 8: Linear quadratic control Mikael Johansson School of Electrical Engineering KTH, Stockholm, Sweden Linear quadratic control Allows to compute the controller

More information

Adaptive Dual Control

Adaptive Dual Control Adaptive Dual Control Björn Wittenmark Department of Automatic Control, Lund Institute of Technology Box 118, S-221 00 Lund, Sweden email: bjorn@control.lth.se Keywords: Dual control, stochastic control,

More information

Experimental evidence showing that stochastic subspace identication methods may fail 1

Experimental evidence showing that stochastic subspace identication methods may fail 1 Systems & Control Letters 34 (1998) 303 312 Experimental evidence showing that stochastic subspace identication methods may fail 1 Anders Dahlen, Anders Lindquist, Jorge Mari Division of Optimization and

More information

Chapter Stability Robustness Introduction Last chapter showed how the Nyquist stability criterion provides conditions for the stability robustness of

Chapter Stability Robustness Introduction Last chapter showed how the Nyquist stability criterion provides conditions for the stability robustness of Lectures on Dynamic Systems and Control Mohammed Dahleh Munther A Dahleh George Verghese Department of Electrical Engineering and Computer Science Massachuasetts Institute of Technology c Chapter Stability

More information

Adaptive linear quadratic control using policy. iteration. Steven J. Bradtke. University of Massachusetts.

Adaptive linear quadratic control using policy. iteration. Steven J. Bradtke. University of Massachusetts. Adaptive linear quadratic control using policy iteration Steven J. Bradtke Computer Science Department University of Massachusetts Amherst, MA 01003 bradtke@cs.umass.edu B. Erik Ydstie Department of Chemical

More information

Departement Elektrotechniek ESAT-SISTA/TR About the choice of State Space Basis in Combined. Deterministic-Stochastic Subspace Identication 1

Departement Elektrotechniek ESAT-SISTA/TR About the choice of State Space Basis in Combined. Deterministic-Stochastic Subspace Identication 1 Katholieke Universiteit Leuven Departement Elektrotechniek ESAT-SISTA/TR 994-24 About the choice of State Space asis in Combined Deterministic-Stochastic Subspace Identication Peter Van Overschee and art

More information

SELECTION OF VARIABLES FOR REGULATORY CONTROL USING POLE VECTORS. Kjetil Havre 1 Sigurd Skogestad 2

SELECTION OF VARIABLES FOR REGULATORY CONTROL USING POLE VECTORS. Kjetil Havre 1 Sigurd Skogestad 2 SELECTION OF VARIABLES FOR REGULATORY CONTROL USING POLE VECTORS Kjetil Havre 1 Sigurd Skogestad 2 Chemical Engineering, Norwegian University of Science and Technology N-734 Trondheim, Norway. Abstract:

More information

Vector autoregressive Moving Average Process. Presented by Muhammad Iqbal, Amjad Naveed and Muhammad Nadeem

Vector autoregressive Moving Average Process. Presented by Muhammad Iqbal, Amjad Naveed and Muhammad Nadeem Vector autoregressive Moving Average Process Presented by Muhammad Iqbal, Amjad Naveed and Muhammad Nadeem Road Map 1. Introduction 2. Properties of MA Finite Process 3. Stationarity of MA Process 4. VARMA

More information

Linear-Quadratic Optimal Control: Full-State Feedback

Linear-Quadratic Optimal Control: Full-State Feedback Chapter 4 Linear-Quadratic Optimal Control: Full-State Feedback 1 Linear quadratic optimization is a basic method for designing controllers for linear (and often nonlinear) dynamical systems and is actually

More information

MA 575 Linear Models: Cedric E. Ginestet, Boston University Revision: Probability and Linear Algebra Week 1, Lecture 2

MA 575 Linear Models: Cedric E. Ginestet, Boston University Revision: Probability and Linear Algebra Week 1, Lecture 2 MA 575 Linear Models: Cedric E Ginestet, Boston University Revision: Probability and Linear Algebra Week 1, Lecture 2 1 Revision: Probability Theory 11 Random Variables A real-valued random variable is

More information

DESIGNING A KALMAN FILTER WHEN NO NOISE COVARIANCE INFORMATION IS AVAILABLE. Robert Bos,1 Xavier Bombois Paul M. J. Van den Hof

DESIGNING A KALMAN FILTER WHEN NO NOISE COVARIANCE INFORMATION IS AVAILABLE. Robert Bos,1 Xavier Bombois Paul M. J. Van den Hof DESIGNING A KALMAN FILTER WHEN NO NOISE COVARIANCE INFORMATION IS AVAILABLE Robert Bos,1 Xavier Bombois Paul M. J. Van den Hof Delft Center for Systems and Control, Delft University of Technology, Mekelweg

More information

Centro de Processamento de Dados, Universidade Federal do Rio Grande do Sul,

Centro de Processamento de Dados, Universidade Federal do Rio Grande do Sul, A COMPARISON OF ACCELERATION TECHNIQUES APPLIED TO THE METHOD RUDNEI DIAS DA CUNHA Computing Laboratory, University of Kent at Canterbury, U.K. Centro de Processamento de Dados, Universidade Federal do

More information

Time Series Models and Inference. James L. Powell Department of Economics University of California, Berkeley

Time Series Models and Inference. James L. Powell Department of Economics University of California, Berkeley Time Series Models and Inference James L. Powell Department of Economics University of California, Berkeley Overview In contrast to the classical linear regression model, in which the components of the

More information

Decentralized control with input saturation

Decentralized control with input saturation Decentralized control with input saturation Ciprian Deliu Faculty of Mathematics and Computer Science Technical University Eindhoven Eindhoven, The Netherlands November 2006 Decentralized control with

More information

Expressions for the covariance matrix of covariance data

Expressions for the covariance matrix of covariance data Expressions for the covariance matrix of covariance data Torsten Söderström Division of Systems and Control, Department of Information Technology, Uppsala University, P O Box 337, SE-7505 Uppsala, Sweden

More information

Linear Riccati Dynamics, Constant Feedback, and Controllability in Linear Quadratic Control Problems

Linear Riccati Dynamics, Constant Feedback, and Controllability in Linear Quadratic Control Problems Linear Riccati Dynamics, Constant Feedback, and Controllability in Linear Quadratic Control Problems July 2001 Revised: December 2005 Ronald J. Balvers Douglas W. Mitchell Department of Economics Department

More information

Linear Regression and Its Applications

Linear Regression and Its Applications Linear Regression and Its Applications Predrag Radivojac October 13, 2014 Given a data set D = {(x i, y i )} n the objective is to learn the relationship between features and the target. We usually start

More information

Kalman Filter. Predict: Update: x k k 1 = F k x k 1 k 1 + B k u k P k k 1 = F k P k 1 k 1 F T k + Q

Kalman Filter. Predict: Update: x k k 1 = F k x k 1 k 1 + B k u k P k k 1 = F k P k 1 k 1 F T k + Q Kalman Filter Kalman Filter Predict: x k k 1 = F k x k 1 k 1 + B k u k P k k 1 = F k P k 1 k 1 F T k + Q Update: K = P k k 1 Hk T (H k P k k 1 Hk T + R) 1 x k k = x k k 1 + K(z k H k x k k 1 ) P k k =(I

More information

Performance assessment of MIMO systems under partial information

Performance assessment of MIMO systems under partial information Performance assessment of MIMO systems under partial information H Xia P Majecki A Ordys M Grimble Abstract Minimum variance (MV) can characterize the most fundamental performance limitation of a system,

More information

Standardization and Singular Value Decomposition in Canonical Correlation Analysis

Standardization and Singular Value Decomposition in Canonical Correlation Analysis Standardization and Singular Value Decomposition in Canonical Correlation Analysis Melinda Borello Johanna Hardin, Advisor David Bachman, Reader Submitted to Pitzer College in Partial Fulfillment of the

More information

Balancing of Lossless and Passive Systems

Balancing of Lossless and Passive Systems Balancing of Lossless and Passive Systems Arjan van der Schaft Abstract Different balancing techniques are applied to lossless nonlinear systems, with open-loop balancing applied to their scattering representation.

More information

4 Derivations of the Discrete-Time Kalman Filter

4 Derivations of the Discrete-Time Kalman Filter Technion Israel Institute of Technology, Department of Electrical Engineering Estimation and Identification in Dynamical Systems (048825) Lecture Notes, Fall 2009, Prof N Shimkin 4 Derivations of the Discrete-Time

More information

Chapter 9 Observers, Model-based Controllers 9. Introduction In here we deal with the general case where only a subset of the states, or linear combin

Chapter 9 Observers, Model-based Controllers 9. Introduction In here we deal with the general case where only a subset of the states, or linear combin Lectures on Dynamic Systems and Control Mohammed Dahleh Munther A. Dahleh George Verghese Department of Electrical Engineering and Computer Science Massachuasetts Institute of Technology c Chapter 9 Observers,

More information

Robust control for a multi-stage evaporation plant in the presence of uncertainties

Robust control for a multi-stage evaporation plant in the presence of uncertainties Preprint 11th IFAC Symposium on Dynamics and Control of Process Systems including Biosystems June 6-8 16. NTNU Trondheim Norway Robust control for a multi-stage evaporation plant in the presence of uncertainties

More information

RICE UNIVERSITY. System Identication for Robust Control. Huipin Zhang. A Thesis Submitted. in Partial Fulfillment of the. Requirements for the Degree

RICE UNIVERSITY. System Identication for Robust Control. Huipin Zhang. A Thesis Submitted. in Partial Fulfillment of the. Requirements for the Degree RICE UNIVERSITY System Identication for Robust Control by Huipin Zhang A Thesis Submitted in Partial Fulfillment of the Requirements for the Degree Master of Science Approved, Thesis Committee: Athanasios

More information

[A + 1 ] + (1 ) v: : (b) Show: the derivative of T at v = v 0 < 0 is: = (v 0 ) (1 ) ; [A + 1 ]

[A + 1 ] + (1 ) v: : (b) Show: the derivative of T at v = v 0 < 0 is: = (v 0 ) (1 ) ; [A + 1 ] Homework #2 Economics 4- Due Wednesday, October 5 Christiano. This question is designed to illustrate Blackwell's Theorem, Theorem 3.3 on page 54 of S-L. That theorem represents a set of conditions that

More information

October 7, :8 WSPC/WS-IJWMIP paper. Polynomial functions are renable

October 7, :8 WSPC/WS-IJWMIP paper. Polynomial functions are renable International Journal of Wavelets, Multiresolution and Information Processing c World Scientic Publishing Company Polynomial functions are renable Henning Thielemann Institut für Informatik Martin-Luther-Universität

More information

THE PROBLEMS OF ROBUST LPC PARAMETRIZATION FOR. Petr Pollak & Pavel Sovka. Czech Technical University of Prague

THE PROBLEMS OF ROBUST LPC PARAMETRIZATION FOR. Petr Pollak & Pavel Sovka. Czech Technical University of Prague THE PROBLEMS OF ROBUST LPC PARAMETRIZATION FOR SPEECH CODING Petr Polla & Pavel Sova Czech Technical University of Prague CVUT FEL K, 66 7 Praha 6, Czech Republic E-mail: polla@noel.feld.cvut.cz Abstract

More information

Multirate MVC Design and Control Performance Assessment: a Data Driven Subspace Approach*

Multirate MVC Design and Control Performance Assessment: a Data Driven Subspace Approach* Multirate MVC Design and Control Performance Assessment: a Data Driven Subspace Approach* Xiaorui Wang Department of Electrical and Computer Engineering University of Alberta Edmonton, AB, Canada T6G 2V4

More information

Lecture 2: Discrete-time Linear Quadratic Optimal Control

Lecture 2: Discrete-time Linear Quadratic Optimal Control ME 33, U Berkeley, Spring 04 Xu hen Lecture : Discrete-time Linear Quadratic Optimal ontrol Big picture Example onvergence of finite-time LQ solutions Big picture previously: dynamic programming and finite-horizon

More information

R. Balan. Splaiul Independentei 313, Bucharest, ROMANIA D. Aur

R. Balan. Splaiul Independentei 313, Bucharest, ROMANIA D. Aur An On-line Robust Stabilizer R. Balan University "Politehnica" of Bucharest, Department of Automatic Control and Computers, Splaiul Independentei 313, 77206 Bucharest, ROMANIA radu@karla.indinf.pub.ro

More information

Chapter Robust Performance and Introduction to the Structured Singular Value Function Introduction As discussed in Lecture 0, a process is better desc

Chapter Robust Performance and Introduction to the Structured Singular Value Function Introduction As discussed in Lecture 0, a process is better desc Lectures on Dynamic Systems and Control Mohammed Dahleh Munther A Dahleh George Verghese Department of Electrical Engineering and Computer Science Massachuasetts Institute of Technology c Chapter Robust

More information

Linear State Feedback Controller Design

Linear State Feedback Controller Design Assignment For EE5101 - Linear Systems Sem I AY2010/2011 Linear State Feedback Controller Design Phang Swee King A0033585A Email: king@nus.edu.sg NGS/ECE Dept. Faculty of Engineering National University

More information

Optimal Distributed Lainiotis Filter

Optimal Distributed Lainiotis Filter Int. Journal of Math. Analysis, Vol. 3, 2009, no. 22, 1061-1080 Optimal Distributed Lainiotis Filter Nicholas Assimakis Department of Electronics Technological Educational Institute (T.E.I.) of Lamia 35100

More information

A Simple Derivation of Right Interactor for Tall Transfer Function Matrices and its Application to Inner-Outer Factorization Continuous-Time Case

A Simple Derivation of Right Interactor for Tall Transfer Function Matrices and its Application to Inner-Outer Factorization Continuous-Time Case A Simple Derivation of Right Interactor for Tall Transfer Function Matrices and its Application to Inner-Outer Factorization Continuous-Time Case ATARU KASE Osaka Institute of Technology Department of

More information

On Controllability and Normality of Discrete Event. Dynamical Systems. Ratnesh Kumar Vijay Garg Steven I. Marcus

On Controllability and Normality of Discrete Event. Dynamical Systems. Ratnesh Kumar Vijay Garg Steven I. Marcus On Controllability and Normality of Discrete Event Dynamical Systems Ratnesh Kumar Vijay Garg Steven I. Marcus Department of Electrical and Computer Engineering, The University of Texas at Austin, Austin,

More information

Elements of Multivariate Time Series Analysis

Elements of Multivariate Time Series Analysis Gregory C. Reinsel Elements of Multivariate Time Series Analysis Second Edition With 14 Figures Springer Contents Preface to the Second Edition Preface to the First Edition vii ix 1. Vector Time Series

More information

Linear Riccati Dynamics, Constant Feedback, and Controllability in Linear Quadratic Control Problems

Linear Riccati Dynamics, Constant Feedback, and Controllability in Linear Quadratic Control Problems Linear Riccati Dynamics, Constant Feedback, and Controllability in Linear Quadratic Control Problems July 2001 Ronald J. Balvers Douglas W. Mitchell Department of Economics Department of Economics P.O.

More information

Singular Value Decomposition Analysis

Singular Value Decomposition Analysis Singular Value Decomposition Analysis Singular Value Decomposition Analysis Introduction Introduce a linear algebra tool: singular values of a matrix Motivation Why do we need singular values in MIMO control

More information

Discrete-Time H Gaussian Filter

Discrete-Time H Gaussian Filter Proceedings of the 17th World Congress The International Federation of Automatic Control Discrete-Time H Gaussian Filter Ali Tahmasebi and Xiang Chen Department of Electrical and Computer Engineering,

More information

STABILITY OF INVARIANT SUBSPACES OF COMMUTING MATRICES We obtain some further results for pairs of commuting matrices. We show that a pair of commutin

STABILITY OF INVARIANT SUBSPACES OF COMMUTING MATRICES We obtain some further results for pairs of commuting matrices. We show that a pair of commutin On the stability of invariant subspaces of commuting matrices Tomaz Kosir and Bor Plestenjak September 18, 001 Abstract We study the stability of (joint) invariant subspaces of a nite set of commuting

More information

Tutorial on Principal Component Analysis

Tutorial on Principal Component Analysis Tutorial on Principal Component Analysis Copyright c 1997, 2003 Javier R. Movellan. This is an open source document. Permission is granted to copy, distribute and/or modify this document under the terms

More information

CANONICAL LOSSLESS STATE-SPACE SYSTEMS: STAIRCASE FORMS AND THE SCHUR ALGORITHM

CANONICAL LOSSLESS STATE-SPACE SYSTEMS: STAIRCASE FORMS AND THE SCHUR ALGORITHM CANONICAL LOSSLESS STATE-SPACE SYSTEMS: STAIRCASE FORMS AND THE SCHUR ALGORITHM Ralf L.M. Peeters Bernard Hanzon Martine Olivi Dept. Mathematics, Universiteit Maastricht, P.O. Box 616, 6200 MD Maastricht,

More information

Zeros and zero dynamics

Zeros and zero dynamics CHAPTER 4 Zeros and zero dynamics 41 Zero dynamics for SISO systems Consider a linear system defined by a strictly proper scalar transfer function that does not have any common zero and pole: g(s) =α p(s)

More information

Optimal Control of Linear Systems with Stochastic Parameters for Variance Suppression: The Finite Time Case

Optimal Control of Linear Systems with Stochastic Parameters for Variance Suppression: The Finite Time Case Optimal Control of Linear Systems with Stochastic Parameters for Variance Suppression: The inite Time Case Kenji ujimoto Soraki Ogawa Yuhei Ota Makishi Nakayama Nagoya University, Department of Mechanical

More information

Limit Cycles in High-Resolution Quantized Feedback Systems

Limit Cycles in High-Resolution Quantized Feedback Systems Limit Cycles in High-Resolution Quantized Feedback Systems Li Hong Idris Lim School of Engineering University of Glasgow Glasgow, United Kingdom LiHonIdris.Lim@glasgow.ac.uk Ai Poh Loh Department of Electrical

More information

Performance Comparison of Two Implementations of the Leaky. LMS Adaptive Filter. Scott C. Douglas. University of Utah. Salt Lake City, Utah 84112

Performance Comparison of Two Implementations of the Leaky. LMS Adaptive Filter. Scott C. Douglas. University of Utah. Salt Lake City, Utah 84112 Performance Comparison of Two Implementations of the Leaky LMS Adaptive Filter Scott C. Douglas Department of Electrical Engineering University of Utah Salt Lake City, Utah 8411 Abstract{ The leaky LMS

More information

6.241 Dynamic Systems and Control

6.241 Dynamic Systems and Control 6.241 Dynamic Systems and Control Lecture 24: H2 Synthesis Emilio Frazzoli Aeronautics and Astronautics Massachusetts Institute of Technology May 4, 2011 E. Frazzoli (MIT) Lecture 24: H 2 Synthesis May

More information

A Cross-Associative Neural Network for SVD of Nonsquared Data Matrix in Signal Processing

A Cross-Associative Neural Network for SVD of Nonsquared Data Matrix in Signal Processing IEEE TRANSACTIONS ON NEURAL NETWORKS, VOL. 12, NO. 5, SEPTEMBER 2001 1215 A Cross-Associative Neural Network for SVD of Nonsquared Data Matrix in Signal Processing Da-Zheng Feng, Zheng Bao, Xian-Da Zhang

More information

PARAMETERIZATION OF STATE FEEDBACK GAINS FOR POLE PLACEMENT

PARAMETERIZATION OF STATE FEEDBACK GAINS FOR POLE PLACEMENT PARAMETERIZATION OF STATE FEEDBACK GAINS FOR POLE PLACEMENT Hans Norlander Systems and Control, Department of Information Technology Uppsala University P O Box 337 SE 75105 UPPSALA, Sweden HansNorlander@ituuse

More information

= 0 otherwise. Eu(n) = 0 and Eu(n)u(m) = δ n m

= 0 otherwise. Eu(n) = 0 and Eu(n)u(m) = δ n m A-AE 567 Final Homework Spring 212 You will need Matlab and Simulink. You work must be neat and easy to read. Clearly, identify your answers in a box. You will loose points for poorly written work. You

More information

DECENTRALIZED CONTROL DESIGN USING LMI MODEL REDUCTION

DECENTRALIZED CONTROL DESIGN USING LMI MODEL REDUCTION Journal of ELECTRICAL ENGINEERING, VOL. 58, NO. 6, 2007, 307 312 DECENTRALIZED CONTROL DESIGN USING LMI MODEL REDUCTION Szabolcs Dorák Danica Rosinová Decentralized control design approach based on partial

More information

Supervisory Control of Petri Nets with. Uncontrollable/Unobservable Transitions. John O. Moody and Panos J. Antsaklis

Supervisory Control of Petri Nets with. Uncontrollable/Unobservable Transitions. John O. Moody and Panos J. Antsaklis Supervisory Control of Petri Nets with Uncontrollable/Unobservable Transitions John O. Moody and Panos J. Antsaklis Department of Electrical Engineering University of Notre Dame, Notre Dame, IN 46556 USA

More information

The Solvability Conditions for the Inverse Eigenvalue Problem of Hermitian and Generalized Skew-Hamiltonian Matrices and Its Approximation

The Solvability Conditions for the Inverse Eigenvalue Problem of Hermitian and Generalized Skew-Hamiltonian Matrices and Its Approximation The Solvability Conditions for the Inverse Eigenvalue Problem of Hermitian and Generalized Skew-Hamiltonian Matrices and Its Approximation Zheng-jian Bai Abstract In this paper, we first consider the inverse

More information

Outline Introduction: Problem Description Diculties Algebraic Structure: Algebraic Varieties Rank Decient Toeplitz Matrices Constructing Lower Rank St

Outline Introduction: Problem Description Diculties Algebraic Structure: Algebraic Varieties Rank Decient Toeplitz Matrices Constructing Lower Rank St Structured Lower Rank Approximation by Moody T. Chu (NCSU) joint with Robert E. Funderlic (NCSU) and Robert J. Plemmons (Wake Forest) March 5, 1998 Outline Introduction: Problem Description Diculties Algebraic

More information

J.TAUSCH AND J.WHITE 1. Capacitance Extraction of 3-D Conductor Systems in. Dielectric Media with high Permittivity Ratios

J.TAUSCH AND J.WHITE 1. Capacitance Extraction of 3-D Conductor Systems in. Dielectric Media with high Permittivity Ratios JTAUSCH AND JWHITE Capacitance Extraction of -D Conductor Systems in Dielectric Media with high Permittivity Ratios Johannes Tausch and Jacob White Abstract The recent development of fast algorithms for

More information

Network Reconstruction from Intrinsic Noise: Non-Minimum-Phase Systems

Network Reconstruction from Intrinsic Noise: Non-Minimum-Phase Systems Preprints of the 19th World Congress he International Federation of Automatic Control Network Reconstruction from Intrinsic Noise: Non-Minimum-Phase Systems David Hayden, Ye Yuan Jorge Goncalves Department

More information

Linear Algebra (part 1) : Vector Spaces (by Evan Dummit, 2017, v. 1.07) 1.1 The Formal Denition of a Vector Space

Linear Algebra (part 1) : Vector Spaces (by Evan Dummit, 2017, v. 1.07) 1.1 The Formal Denition of a Vector Space Linear Algebra (part 1) : Vector Spaces (by Evan Dummit, 2017, v. 1.07) Contents 1 Vector Spaces 1 1.1 The Formal Denition of a Vector Space.................................. 1 1.2 Subspaces...................................................

More information

Information About Ellipses

Information About Ellipses Information About Ellipses David Eberly, Geometric Tools, Redmond WA 9805 https://www.geometrictools.com/ This work is licensed under the Creative Commons Attribution 4.0 International License. To view

More information

David Sagan. The standard analysis for a coupled lattice is based upon the formalism of Edwards

David Sagan. The standard analysis for a coupled lattice is based upon the formalism of Edwards Twiss Analysis With a Mobius Lattice David Sagan CBN 98{14 July 10, 1998 1 Introduction The standard analysis for a coupled lattice is based upon the formalism of Edwards and Teng[1]. This is ne for weak

More information

A Finite Element Method for an Ill-Posed Problem. Martin-Luther-Universitat, Fachbereich Mathematik/Informatik,Postfach 8, D Halle, Abstract

A Finite Element Method for an Ill-Posed Problem. Martin-Luther-Universitat, Fachbereich Mathematik/Informatik,Postfach 8, D Halle, Abstract A Finite Element Method for an Ill-Posed Problem W. Lucht Martin-Luther-Universitat, Fachbereich Mathematik/Informatik,Postfach 8, D-699 Halle, Germany Abstract For an ill-posed problem which has its origin

More information

Control, Stabilization and Numerics for Partial Differential Equations

Control, Stabilization and Numerics for Partial Differential Equations Paris-Sud, Orsay, December 06 Control, Stabilization and Numerics for Partial Differential Equations Enrique Zuazua Universidad Autónoma 28049 Madrid, Spain enrique.zuazua@uam.es http://www.uam.es/enrique.zuazua

More information

Here represents the impulse (or delta) function. is an diagonal matrix of intensities, and is an diagonal matrix of intensities.

Here represents the impulse (or delta) function. is an diagonal matrix of intensities, and is an diagonal matrix of intensities. 19 KALMAN FILTER 19.1 Introduction In the previous section, we derived the linear quadratic regulator as an optimal solution for the fullstate feedback control problem. The inherent assumption was that

More information

Linear Quadratic Zero-Sum Two-Person Differential Games Pierre Bernhard June 15, 2013

Linear Quadratic Zero-Sum Two-Person Differential Games Pierre Bernhard June 15, 2013 Linear Quadratic Zero-Sum Two-Person Differential Games Pierre Bernhard June 15, 2013 Abstract As in optimal control theory, linear quadratic (LQ) differential games (DG) can be solved, even in high dimension,

More information

DESIGN AND IMPLEMENTATION OF SENSORLESS SPEED CONTROL FOR INDUCTION MOTOR DRIVE USING AN OPTIMIZED EXTENDED KALMAN FILTER

DESIGN AND IMPLEMENTATION OF SENSORLESS SPEED CONTROL FOR INDUCTION MOTOR DRIVE USING AN OPTIMIZED EXTENDED KALMAN FILTER INTERNATIONAL JOURNAL OF ELECTRONICS AND COMMUNICATION ENGINEERING & TECHNOLOGY (IJECET) International Journal of Electronics and Communication Engineering & Technology (IJECET), ISSN 0976 ISSN 0976 6464(Print)

More information

Lecture 2: Review of Prerequisites. Table of contents

Lecture 2: Review of Prerequisites. Table of contents Math 348 Fall 217 Lecture 2: Review of Prerequisites Disclaimer. As we have a textbook, this lecture note is for guidance and supplement only. It should not be relied on when preparing for exams. In this

More information

Gaussian process for nonstationary time series prediction

Gaussian process for nonstationary time series prediction Computational Statistics & Data Analysis 47 (2004) 705 712 www.elsevier.com/locate/csda Gaussian process for nonstationary time series prediction Soane Brahim-Belhouari, Amine Bermak EEE Department, Hong

More information

Automatic Control Systems theory overview (discrete time systems)

Automatic Control Systems theory overview (discrete time systems) Automatic Control Systems theory overview (discrete time systems) Prof. Luca Bascetta (luca.bascetta@polimi.it) Politecnico di Milano Dipartimento di Elettronica, Informazione e Bioingegneria Motivations

More information

SUCCESSIVE POLE SHIFTING USING SAMPLED-DATA LQ REGULATORS. Sigeru Omatu

SUCCESSIVE POLE SHIFTING USING SAMPLED-DATA LQ REGULATORS. Sigeru Omatu SUCCESSIVE POLE SHIFING USING SAMPLED-DAA LQ REGULAORS oru Fujinaka Sigeru Omatu Graduate School of Engineering, Osaka Prefecture University, 1-1 Gakuen-cho, Sakai, 599-8531 Japan Abstract: Design of sampled-data

More information

Applied Mathematics 505b January 22, Today Denitions, survey of applications. Denition A PDE is an equation of the form F x 1 ;x 2 ::::;x n ;~u

Applied Mathematics 505b January 22, Today Denitions, survey of applications. Denition A PDE is an equation of the form F x 1 ;x 2 ::::;x n ;~u Applied Mathematics 505b January 22, 1998 1 Applied Mathematics 505b Partial Dierential Equations January 22, 1998 Text: Sobolev, Partial Dierentail Equations of Mathematical Physics available at bookstore

More information

The norms can also be characterized in terms of Riccati inequalities.

The norms can also be characterized in terms of Riccati inequalities. 9 Analysis of stability and H norms Consider the causal, linear, time-invariant system ẋ(t = Ax(t + Bu(t y(t = Cx(t Denote the transfer function G(s := C (si A 1 B. Theorem 85 The following statements

More information

CONTROL SYSTEMS, ROBOTICS AND AUTOMATION CONTENTS VOLUME VII

CONTROL SYSTEMS, ROBOTICS AND AUTOMATION CONTENTS VOLUME VII CONTENTS VOLUME VII Control of Linear Multivariable Systems 1 Katsuhisa Furuta,Tokyo Denki University, School of Science and Engineering, Ishizaka, Hatoyama, Saitama, Japan 1. Linear Multivariable Systems

More information

Enforcing Passivity for Admittance Matrices Approximated by Rational Functions

Enforcing Passivity for Admittance Matrices Approximated by Rational Functions IEEE TRANSACTIONS ON POWER SYSTEMS, VOL. 16, NO. 1, FEBRUARY 2001 97 Enforcing Passivity for Admittance Matrices Approximated by Rational Functions Bjørn Gustavsen, Member, IEEE and Adam Semlyen, Life

More information

7. MULTIVARATE STATIONARY PROCESSES

7. MULTIVARATE STATIONARY PROCESSES 7. MULTIVARATE STATIONARY PROCESSES 1 1 Some Preliminary Definitions and Concepts Random Vector: A vector X = (X 1,..., X n ) whose components are scalar-valued random variables on the same probability

More information

Some Time-Series Models

Some Time-Series Models Some Time-Series Models Outline 1. Stochastic processes and their properties 2. Stationary processes 3. Some properties of the autocorrelation function 4. Some useful models Purely random processes, random

More information

Chapter 6. Maximum Likelihood Analysis of Dynamic Stochastic General Equilibrium (DSGE) Models

Chapter 6. Maximum Likelihood Analysis of Dynamic Stochastic General Equilibrium (DSGE) Models Chapter 6. Maximum Likelihood Analysis of Dynamic Stochastic General Equilibrium (DSGE) Models Fall 22 Contents Introduction 2. An illustrative example........................... 2.2 Discussion...................................

More information

Prediction, filtering and smoothing using LSCR: State estimation algorithms with guaranteed confidence sets

Prediction, filtering and smoothing using LSCR: State estimation algorithms with guaranteed confidence sets 2 5th IEEE Conference on Decision and Control and European Control Conference (CDC-ECC) Orlando, FL, USA, December 2-5, 2 Prediction, filtering and smoothing using LSCR: State estimation algorithms with

More information

The Uniformity Principle: A New Tool for. Probabilistic Robustness Analysis. B. R. Barmish and C. M. Lagoa. further discussion.

The Uniformity Principle: A New Tool for. Probabilistic Robustness Analysis. B. R. Barmish and C. M. Lagoa. further discussion. The Uniformity Principle A New Tool for Probabilistic Robustness Analysis B. R. Barmish and C. M. Lagoa Department of Electrical and Computer Engineering University of Wisconsin-Madison, Madison, WI 53706

More information

FIR Filters for Stationary State Space Signal Models

FIR Filters for Stationary State Space Signal Models Proceedings of the 17th World Congress The International Federation of Automatic Control FIR Filters for Stationary State Space Signal Models Jung Hun Park Wook Hyun Kwon School of Electrical Engineering

More information

O_hp. W_e. d_ohp PWR. U_cex. Q_ex. d_ucex. Primary and. N_cd. Secondary Circuit. K_tb. W_ref. K_cd. turbine control valve. secondary circuit TURBINE

O_hp. W_e. d_ohp PWR. U_cex. Q_ex. d_ucex. Primary and. N_cd. Secondary Circuit. K_tb. W_ref. K_cd. turbine control valve. secondary circuit TURBINE A Comparison Between Model Reduction and Contrler Reduction: Application to a PWR Nuear Plant y Beno^t Codrons,Pascale Bendotti, Clement-Marc Falinower and Michel Gevers CESAME, Universite Cathique de

More information

1 Matrices and Systems of Linear Equations

1 Matrices and Systems of Linear Equations Linear Algebra (part ) : Matrices and Systems of Linear Equations (by Evan Dummit, 207, v 260) Contents Matrices and Systems of Linear Equations Systems of Linear Equations Elimination, Matrix Formulation

More information

1 Outline Part I: Linear Programming (LP) Interior-Point Approach 1. Simplex Approach Comparison Part II: Semidenite Programming (SDP) Concludin

1 Outline Part I: Linear Programming (LP) Interior-Point Approach 1. Simplex Approach Comparison Part II: Semidenite Programming (SDP) Concludin Sensitivity Analysis in LP and SDP Using Interior-Point Methods E. Alper Yldrm School of Operations Research and Industrial Engineering Cornell University Ithaca, NY joint with Michael J. Todd INFORMS

More information

Mathematical Theory of Control Systems Design

Mathematical Theory of Control Systems Design Mathematical Theory of Control Systems Design by V. N. Afarias'ev, V. B. Kolmanovskii and V. R. Nosov Moscow University of Electronics and Mathematics, Moscow, Russia KLUWER ACADEMIC PUBLISHERS DORDRECHT

More information

Linear Algebra (part 1) : Matrices and Systems of Linear Equations (by Evan Dummit, 2016, v. 2.02)

Linear Algebra (part 1) : Matrices and Systems of Linear Equations (by Evan Dummit, 2016, v. 2.02) Linear Algebra (part ) : Matrices and Systems of Linear Equations (by Evan Dummit, 206, v 202) Contents 2 Matrices and Systems of Linear Equations 2 Systems of Linear Equations 2 Elimination, Matrix Formulation

More information

MARKOVIANITY OF A SUBSET OF COMPONENTS OF A MARKOV PROCESS

MARKOVIANITY OF A SUBSET OF COMPONENTS OF A MARKOV PROCESS MARKOVIANITY OF A SUBSET OF COMPONENTS OF A MARKOV PROCESS P. I. Kitsul Department of Mathematics and Statistics Minnesota State University, Mankato, MN, USA R. S. Liptser Department of Electrical Engeneering

More information