This action might not be possible to undo. Are you sure you want to continue?

BooksAudiobooksComicsSheet Music### Categories

### Categories

### Categories

Editors' Picks Books

Hand-picked favorites from

our editors

our editors

Editors' Picks Audiobooks

Hand-picked favorites from

our editors

our editors

Editors' Picks Comics

Hand-picked favorites from

our editors

our editors

Editors' Picks Sheet Music

Hand-picked favorites from

our editors

our editors

Top Books

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Audiobooks

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Comics

What's trending, bestsellers,

award-winners & more

award-winners & more

Top Sheet Music

What's trending, bestsellers,

award-winners & more

award-winners & more

Welcome to Scribd! Start your free trial and access books, documents and more.Find out more

Subspace Identification of Hammerstein and Wiener Models

**Speaker Juan C. Gómez
**

Laboratory for System Dynamics and Signal Processing FCEIA, Universidad Nacional de Rosario ARGENTINA

jcgomez@fceia.unr.edu.ar

LSD - September 2005 J.C. Gómez 1

Outline Outline

Introduction: Motivation, New results A (very) brief review on Subspace State-Space System IDentifivery cation Methods Block-oriented Nonlinear Models Subspace Identification of Hammerstein Models Subspace Identification of Wiener Models Simulation Examples Conclusions

LSD - September 2005

J.C. Gómez

2

Introduction Introduction

Motivation for Nonlinear (Subspace) Identification (Subspace)

• Most physical processes have a nonlinear behaviour, except in a limited range where they can be considered linear. • The performance of controllers designed from a linear approximation is strongly influenced by a change in the operating point of the system. • Nonlinear models are able to describe more accurately the global behaviour of the system, independently of the operating point. • Many dynamical systems can be represented by the interconnection of static nonlinearities and LTI systems. These models are called block-oriented nonlinear models.

LSD - September 2005 J.C. Gómez 3

• Subspace Methods have been very successful for the identification of LTI models in many practical applications. • Although there is a well developed theory for Subspace Identification methods for LTI systems, this is not the case for nonlinear systems. Some recent contributions in this area are: (Verhaegen & Westwick, 1996) in Subspace Identification of Hammersterin and Wiener models, and (Chen & Maciejowski, 2000) and (Favoreel et al., 1999) in Subspace Identification of bilinear systems.

LSD - September 2005

J.C. Gómez

4

Step 2: a 2-norm minimization problem which is solved via an SVD. J. A. Gómez. 11. European Journal of Control. No. No. 2005. IEE Proceedings on Control Theory and Applications. J. Vol.C.. Subspace Identification of Multivariable Hammerstein and Wiener Models.September 2005 J. and Baeyens. E. 151. • The proposed algorithms consist basically of two steps: Step 1: a standard (linear) subspace algorithm applied to an equivalent linear system whose inputs (outputs) are filtered (by the basis functions describing the static nonlinearities) versions of the original inputs (outputs). Vol.C. May 2004. Gómez 5 References 1..September 2005 J. 3. 2005) • New subspace algorithms for the simultaneous identification of the linear and nonlinear parts of multivariable Hammerstein and Wiener models are presented. 2. 2. pp. LSD . E. • Provided the conditions for the consistency of the linear subspace algorithm used in Step 1 are satisfied.The new results (Gomez & Baeyens. Gómez 6 . LSD . and Baeyens.C. Gómez. 329-338. consistency of the estimates can be guaranteed. Jutan.C. Wiener Model Identification and Predictive Control of a pH Neutralization Process..

C. Gómez no problems 7 They don’t require a particular (canonical) state-space realization numerical conditioning improves. LSD . 1966). LSD . All subspace methods compute at some stage the subspace spanned by the columns of the extended observability matrix. convergence and initialization.September 2005 J. Matlab Main computational tools are QR and SVD.September 2005 J. The algorithms can be (they have been) efficiently implemented in software like Matlab. They have their origin in Realization Theory as developed in the 60/70s (Ho & Kalman.C. They don’t require iterative optimization procedures with local minima. N4SID.Subspace State-Space System IDentification Subspace State-Space System IDentification 4SID Methods Properties They combine tools of System Theory. MOESP. They provide reliable state-space models of multivariable LTI systems directly from input-output data. They require a modest computational load in comparison to traditional identification methods like PEM. Gómez 8 .. Numerical Linear Algebra Theory and Geometry (projections). The various algorithms (e. CVA) differ in the way the extended observability matrix is estimated and also in the way it is used to compute the system matrices.g.

Gómez 10 . C.September 2005 9 Realization-based 4SID Methods For a LTI system.The system model xk +1 = Axk + Buk + Kek y k = Cxk + Duk + ek State-space model in innovation form The identification problem To estimate the system matrices (A. C. Gómez LSD . from an (N+α-1)-point data set of input and output measurements {uk . l > 0 LSD . B. B. D) and K . D) completely defines the input-output properties of the system through yk = ∑ hluk −l l =0 ∞ convolution sum are related to the system where the impulse response coefficients hl matrices by . and the model order n. l=0 ⎧ D hl = ⎨ l−1 ⎩CA B . yk }kN=+1α −1 J.C.C.September 2005 J. a minimal state-space realization (A.

and Σ1 will contain the n non-zero singular values → model order is computed. LSD .September 2005 J.September 2005 J. Hij will be a rank n matrix. This factorization is provided by the SVD of matrix Hij. via correlation analysis) the impulse response of the system → not good LSD . In the presence of noise.C. Gómez 12 . Problems: it is necessary to measure or to estimate (for example.⎡ h1 ⎢h 2 H ij = ⎢ ⎢M ⎢ ⎢ hi ⎣ h2 h3 M hi +1 L L hj ⎤ h j +1 ⎥ ⎥ O M ⎥ ⎥ L hi + j −1 ⎥ ⎦ H ij = Γi C j Impulse Response Hankel Matrix Extended Observability Matrix (i>n) Extended Controlability Matrix (j>n) An estimate of the extended observability matrix can be computed by a full rank factorization of the impulse response Hankel matrix. Gómez 11 H ij = [U1 ⎡Σ1 0 ⎤ ⎡V1T ⎤ T ⎛U Σ 12 ⎞⎛ Σ 12V T ⎞ U 2 ]⎢ ⎢ ⎥ ≈ U1Σ1V1 = ⎜ 1 1 ⎟⎜ 1 1 ⎟ ⎝4 4⎠⎝ 4 4⎠ 0 Σ 2 ⎥ ⎣V2T ⎦ ⎣ ⎦ 1 2 31 2 3 ˆ Γi ˆ Cj rank reduction In the absence of noise. Hij will have full rank and a rank reduction stage will be required for the model order determination.C.

Direct 4SID Methods Yα = Γα X + H α Uα + Nα ⎡ y1 ⎢ y Yα = ⎢ 2 ⎢ M ⎢ ⎣ yα −1 y2 y3 M yα L L yN ⎤ y N +1 ⎥ ⎥ M M ⎥ ⎥ L y N +α −1 ⎦ fundamental equation (1) Output block Hankel matrix (In a similar way are defined the Input block Hankel matrix Uα and the Noise block Hankel matrix Nα.September 2005 J. Gómez X = [x1 . by LSD . x2 . L .September 2005 J.e. x N ] State Sequence Matrix 13 0 0 ⎡ D ⎢ CB D 0 ⎢ H α = ⎢ CAB CB D ⎢ M M ⎢ M α −2 α −3 α −4 ⎢CA B CA B CA B ⎣ 0⎤ 0⎥ ⎥ L 0⎥ ⎥ O M⎥ L D⎥ ⎦ L L Lower triangular block Toeplitz matrix of impulse responses (unknown). i. Gómez 14 . (2) by the orthogonal projection onto the null space of Uα. (1) becomes Yα = Γα X + H α Uα (2) and the part of the output which does not emanate from the state can be removed by multiplying (from the right) both sides of eq.) ⎡ C ⎤ ⎢ CA ⎥ ⎥ Extended (α > n) Γα = ⎢ ⎢ M ⎥ Observability Matrix ⎢ α −1 ⎥ ⎣CA ⎦ LSD .C. The main idea of Direct 4SID methods In the absence of noise (Nα = 0). eq.C.

Gómez 15 The factorization is provided by the SVD of the matrix on the left side Yα Uα = [U1 ⊥ ⎡Σ1 0 ⎤ ⎡V1T ⎤ T ⎛U Σ 12 ⎞⎛ Σ 12V T ⎞ (4) U 2 ]⎢ ⎢ ⎥ ≈ U1Σ1V1 = ⎜ 1 1 ⎟⎜ 1 1 ⎟ ⎝4 4⎠⎝ ⎠ 0 Σ 2 ⎥ ⎣V2T ⎦ ⎣ ⎦ 123 ˆ Γα rank reduction (model order estimation) (In the absence of noise Σ2 = 0) Weighting Matrices Row and column weighting matrices can be introduced in (4) before performing the SVD of the matrix in the left hand side. Estimates of the corresponding system matrices can be obtained by resorting to the shift invariance property of the extended observability matrix. Any choice of positive-definite weighting matrices Wr and Wc will result in consistent estimates of the extended observability matrix. and by solving a system of linear equations in the least squares sense.Π ⊥ T Uα = I − Uα Uα Uα T ∆ ( T −1 ) ⊥ Uα = Uα ∆ orthogonal projection such that ⊥ Uα Uα = I This yields ⊥ ⊥ Yα Uα = Γα XUα (3) Note that the matrix on the left depends exclusively on the input-output data. LSD . Then.C.C.September 2005 J. a full rank factorization of this matrix will provide an ˆ estimate Γα of the extended observability matrix. LSD .September 2005 J. Gómez 16 .

Wc = ⎜ ΦΠ U T Φ T ⎟ Φ α ⎝N ⎠ LSD . 1994): ⎛1 ⎞ ⊥ Wr = I . Gómez 17 −1 Computation of the system matrices ˆ Given an estimate Γα of the extended observability matrix.Wr Yα UαWc = [U1 ⊥ 1 1 ⎡Σ1 0 ⎤ ⎡V1T ⎤ U 2 ]⎢ ≈ U1Σ1V1T = ⎛U1Σ1 2 ⎞⎛ Σ1 2V1T ⎞ ⎜ ⎟⎜ ⎟ ⎥⎢ T ⎥ ⎝4 4⎠⎝ ⎠ ⎣ 0 Σ 2 ⎦ ⎣V2 ⎦ 123 ˆ Γα change of coordinates in state-space Existing algorithms employ the following choices for matrices Wr and Wc . Wc = ⎜ ΦΠ U T Φ ⎟ α α ⎝N ⎠ ⎝N ⎠ −1 • N4SID (Van Overschee and de Moor.September 2005 J. 1994): Wr = I .C. 1 ⊥ ⊥ • MOESP (Verhaegen. 1990): Wr = ⎜ Yα ΠU T Yα ⎟ . Gómez 18 . Wc = ⎛ ΦΠ U T Φ T ⎞ ΦΠ U T ⎜ ⎟ α α ⎝N ⎠ −1 −1 2 2 ⎛1 ⎛1 T ⎞ T ⎞ ⊥ ⊥ • CVA (Larimore.C.September 2005 J. estimates of the system matrices can be computed as: ˆ • C : first row block of ˆ • A : solving in the least squares sense ˆ Γα ˆ Γα = Γα A shift-invariance property ˆ ˆ • B and D : solving a system of linear equations LSD .

Presence of noise In the presence of noise Yα = Γα X + H α Uα + Nα and ⊥ ⊥ Yα Uα = Γα X + Nα Uα noise term needs to be removed The noise term can be removed by correlating it away with a suitable matrix.C. 1: Hammerstein Model (NL) Fig.September 2005 J. 3: Hammerstein-Wiener Model (NLN) Fig. Gómez 20 . This can be interpreted as an oblique projection. Gómez 19 Block-oriented Nonlinear Models Block-oriented Nonlinear Models Static Nonlinearity uk LTI System yk uk LTI System Static Nonlinearity yk Fig. 4: Hammerstein-Wiener Model (LNL) LSD .C. LSD .September 2005 J. 2: Wiener Model (LN) Static Nonlinearity uk LTI System Static Nonlinearity yk uk LTI System Static Nonlinearity LTI System yk Fig.

B. and the model order n.September 2005 α i ∈ ℜ p× p J. (i = 1. and the model order n. Gómez Identifiability problem 22 . respectively. respectively. B. r ) known basis functions ωk ∈ ℜn . from an N-point data set of observed input-output measurements. and D characterizing the nonlinear and the linear N parts. (2) ⇒ ∑ ∑ Normalization LSD . vk ∈ ℜ p v k = N (u k ) = g i (• ) : ℜ p p ∑ α g (u i i i =1 r k ) (3) → ℜ . L . and A.L. r ) . and D characterizing the nonlinear and the linear . C.C.Hammerstein Model Identification Hammerstein Model Identification Problem Formulation Problem Formulation ωk uk vk N(.L . yk }k =1 of parts. Subspace Identification Algorithm Subspace Identification Algorithm r ⎧ Bα i g i (u k ) + ω k ⎪ xk +1 = Axk + ⎪ i =1 ⎨ r ⎪ Dα i g i (u k ) + υ k ⎪ y k = Cxk + i =1 ⎩ (3) → (1).) A C B D υk yk Fig. from an N-point data set {uk . (i = 1.C. Gómez (i = 1. xk ∈ ℜ n . 5: Hammerstein model LTI subsystem Nonlinear subsystem (1) (2) ⎧ xk +1 = Axk + Bvk + ω k ⎨ ⎩ y k = Cxk + Dvk + υ k yk ∈ ℜ m . observed input-output measurements. r ) unknown matrix parameters 21 Identification problem: to estimate the unknown parameter matrices Identification problem: to estimate the unknown parameter matrices α i ∈ ℜ p× p . and A. υk ∈ ℜm LSD . C.September 2005 αi 2 =1 J.

D α r ] .CVA) Fig. and D = Dα T . α r ]T .. Gómez 23 Defining ~ ~ α = [α1 . α = argmin ⎨ Θ BD − ⎢ ⎥α T B . D = [D α 1 . D . and α ~ ~ from the estimate of the matrices B .September 2005 ⎧ ⎡B⎤ ⎪ ˆ ˆ ˆ ˆ B. in the 2-norm sense.e. MOESP. so that ~ ⎡B ⎤ ⎡B⎤ T Θ BD = ⎢ ~ ⎥ = ⎢ ⎥α ⎣D⎦ ⎣D⎦ ∆ The problem then is how to compute estimates of matrices B. 6: Equivalent LTI system with input Uk Estimates ˆ ˆ ˆ ~ ˆ ~ A. D. model order n LSD . D. L . J. L . U k = g 1T (u k ).Defining ∆ T ~∆ ~∆ B = [B α 1 .C. and α ˆ are such that ( LSD . estimates B. D. L . L . D. g r T (u k ) [ ] ωk ~ ⎧ xk +1 = Axk + B U k + ω k ⎪ ⎨ ~ ⎪ y k = Cx k + DU k + υ k ⎩ υk yk Uk A C ~ B ~ D Linear Subspace Algorithms (N4SID. C . from an estimate of Θ BD ) ˆ ˆ It is clear that the closest.September 2005 J. then B = Bα T .α ⎪ ⎣D⎦ ⎩ ) ⎫ ⎪ ⎬ 2⎪ ⎭ 2 ˆ The solution to this optimization problem is provided by the SVD of Θ BD . Gómez 24 .C. and D (i. B α r ] . B .

Normalization in α provided by the SVD 25 2 and the approximation error is given by 2 = σ p +1. V1 ). D . L . and Σ 1 = diag (σ 1 .September 2005 . Then T s ˆ = UΣV T = ∑ σ u vT = [U U ]⎡Σ1 0 ⎤ ⎡V1 ⎤ Θ BD 1 2 ⎢ i i i ⎥⎢ T ⎥ i =1 ⎣ 0 Σ 2 ⎦ ⎣V2 ⎦ ˆ ⎛⎡B⎤ ⎞ B ˆ ⎜ ⎢ ⎥ . C . such as N4SID. and α as ˆ ⎡B⎤ ⎢ ˆ ⎥ = U1Σ1 ⎣ D⎦ ˆ α = V1 J. and let its economy size SVD be (4) with U 1 ∈ ℜ ( n + m )× p . CVA.Result 1 Result 1 ˆ Let Θ BD ∈ ℜ ( n + m )×rp partitioned as have rank s>p. Gómez 26 ~ ˆ ~ ˆ respectively. and the model order n. α ⎟ = argmin Θ BD − ⎡ ⎤α T ˆ ⎢D⎥ ⎜ D ⎟ ˆ B . V1 ∈ ℜ rp × p . ~ ~ Step 1: Compute estimates of the system matrices A.September 2005 J. ˆ Step 3: Compute the SVD of Θ BD and its partition as in (4).C. LSD . σ 2 . Step 4: Compute the estimates of the parameter matrices B.C. Gómez Identification Algorithm Identification Algorithm The subspace algorithm can be summarized as follows. B . D . ( ) ˆ Step 2: Based on the estimates B and D compute an estimate Θ BD of matrix Θ BD . D. using any available (linear) subspace algorithm.α ⎣ ⎦ ⎝⎣ ⎦ ⎠ ˆ ⎡B⎤ ˆ Θ BD − ⎢ ⎥α T ˆ ⎣D⎦ 2 2 = (U 1Σ 1 . 2 LSD . MOESP. σ p ) .

D ⎞ are consistent in the sense that they converge ˆ ~ ˆ ~⎟ the estimates ⎜ ⎝ ⎠ to the true values when the number of data points N →∞ .C.Result 2: Consistency Analysis Result 2: Consistency Analysis Under some assumptions on persistency of excitation of the inputs.September 2005 J. B .C.) yk Fig. ~ ~ ˆ ˆ The consistency of B and D . r ) . D. Gómez 27 Wiener Model Identification Wiener Model Identification Problem Formulation Problem Formulation ωk uk A C B D υk vk N(. and α . C . r ) known basis functions unknown matrix parameters 28 ωk ∈ ℜn . vk ∈ ℜ p n vk = N −1 ( yk ) = ∑ α i g i ( yk ) i =1 r (7) g i (• ) : ℜ m → ℜ m . LSD .L. υk ∈ ℜm LSD . implies that of B.September 2005 α i ∈ ℜ m× m J. which depend on the particular subspace method used in Step 1 of the algorithm. ˆ ˆ ⎛ A. 7: Wiener model LTI subsystem Nonlinear subsystem (5) (6) m ⎧ xk +1 = Axk + Bu k + ω k ⎨ ⎩vk = Cxk + Duk + υ k u k ∈ ℜ . (i = 1. Gómez (i = 1. xk ∈ ℜ . L .

B. Yk = [g1T ( yk ). yk }k =1 linear parts. B. (i = 1.α r ] .C. and D characterizing the nonlinear and the Identification problem: to estimate the unknown parameter matrices Identification problem: to estimate the unknown parameter matrices linear parts. C.and A. L . respectively. N Subspace Identification Algorithm Subspace Identification Algorithm ⎧ xk +1 = Axk + Buk + ω k (7) → (6) ⇒ ⎪ ∆ r ⎨ ⎪αYk = ∑ α i g i ( yk ) = Cxk + Duk + υ k i =1 ⎩ α = [α1 . D. respectively.L. model order n LSD . g rT ( yk )] + Normalization α T ⎧ xk +1 = Axk + Buk + ω k ⇒ ⎨ ~ ~ ~ ⎩Yk = Cxk + Duk + υ k ~∆ + ~∆ + C =α C .CVA) ˆ ˆ ˆ ˆ ~ ~ Estimates A.September 2005 ωk ⎧ xk +1 = Axk + Buk + ω k ⎨ ~ ~ ~ Yk = Cxk + Duk + υ k ⎩ uk υk Yk A ~ C B ~ D Fig. from an N-point data set {uk . 8: Equivalent LTI model with output Yk Linear Subspace Algorithms (N4SID. B. .L . Gómez 30 . D =α D Identifiability problem 29 2 =1 J. C. C .September 2005 J.C. and the model order n. of observed input-output measurements. and the model order n.α i ∈ ℜ m×m . from an N-point data set of observed input-output measurements. MOESP. and D characterizing the nonlinear and the and A. r ). Gómez LSD .

and α + are such that ˆ ˆ ˆ (C. 2 2 ( ) and the approximation error is given by ~ ˆ ⎡C ⎢ ⎣ LSD .α + ⎩ ⎣ ~ ˆ D ⎤ − α + [C ⎥ ⎦ ⎫ D] ⎬ 2⎭ 2 The solution to this optimization problem is provided by the SVD of ~ ~ ˆ ˆ the matrix ⎡C D ⎤ ⎢ ⎥ ⎣ ⎦ LSD . estimates C .September 2005 ~ ˆ ˆ ˆ D⎤ − α + C ⎥ ⎦ J.α + ]) ~ ˆ D ⎤ − α + [C ⎥ ⎦ D ] = U 1 . D. and Σ 1 = diag (σ 1 .The problem is how to compute estimates of matrices C. L . σ 2 . V1 ∈ ℜ ( n + p )× m . σ m ) . Normalization in α+ provided by the SVD 32 2 . Gómez 31 Result 3 Result 3 ˆ ˆ mr ×( n + p ) ⎡ ~ ~⎤ Let ⎢C D ⎥ ∈ ℜ have rank s>m. and α+ ~ ~ from the estimates of the matrices C .αˆ ) + ⎧ ~ ˆ = argmin ⎨ ⎡C ⎢ C . D. D . and let its economy size SVD ⎣ ⎦ be partitioned as ~ ˆ ⎡C ⎢ ⎣ s ⎡Σ1 0 ⎤ ⎡V1T ⎤ ~⎤ ˆ T T D = UΣV = ∑ σ i ui vi = [U1 U 2 ]⎢ ⎥⎢ T ⎥ ⎥ ⎦ i =1 ⎣ 0 Σ 2 ⎦ ⎣V2 ⎦ (8) with U 1 ∈ ℜ mr × m . Σ 1V1T . D.C. [Cˆ + ~ ˆ ˆ D = argmin ⎡C ⎢ ⎣ C . in the ˆ ˆ 2-norm sense. Gómez [ ˆ D ] 2 2 = σ m +1.C.September 2005 J. D . and D Similarly to what was done for the Hammerstein model the closest. Then (αˆ .

using any available (linear) subspace algorithm. and α+ as ˆ ˆ [C D] = Σ V ˆ α = U1+ T 1 1 ⎣ ⎦ respectively. ~ ~ ˆ ˆ Step 2: Compute the SVD of ⎡C D ⎤ and its partition as in (8).C. such as N4SID.15 z + 0.0263 u k nonlinear subsystem The input and noise uk = sin (0.C.7 z − 1 .0025πk ) + 0. LSD . Gómez 34 . B.September 2005 J. Gómez 33 Simulation Examples Simulation Examples Example 1: Hammerstein Model ID ((“academic”) Example 1: Hammerstein Model ID “academic”) The True System G( z) = z 2 + 0 . ~ ~ Step 1: Compute estimates of the system matrices A.September 2005 J. ⎢ ⎥ ( ) Step 3: Compute the estimates of the parameter matrices C.9 z 2 + 0.8589 u k + 0.5 z 3 + 0.5113 u k − 0. MOESP.1 sin (0. C .0015πk ) + + 0.2 − 0.0005πk ) + 0.0035πk ) + γ k input ( γ k white noise with variance 10 −6 ) Spectrum of the zero mean coloured noise 0.3 sin (0.64 × 10−8 Φν (ω ) = 1.0149 u k − 0. CVA. and the model order n.002 linear subsystem 2 3 4 N (u k ) = 0.Identification Algorithm Identification Algorithm The subspace algorithm can be summarized as follows. D . D.4 cos(ω ) LSD .5 sin (0.

0014 J.9986 z 2 + 0.9002 z 2 + 0.4984 z 3 + 0. Gómez Estimated linear subsystem 35 LSD .8589 u k + 0.September 2005 Validation results Fig.6997 z − 1.0142 u k2 − 0.9: True (blue) and Estimated (green) nonlinear characteristic.C. Gómez 36 . The Estimated Linear Subsystem ˆ G (z ) = 0.September 2005 J.1495 z + 0.0260 u k4 Estimated nonlinear subsystem Fig. 10: True (green) and Estimated (blue) Output. LSD .5113 u k − 0.The Estimated Nonlinear Subsystem 3 ˆ N (u k ) = 0.C.

and validation (remaining 1000 points) Input Data.September 2005 J. 11: Schematic representation of the distillation column (Weischedel & McAvoy.September 2005 J. 12: Left Plot: Estimation (first 1000 points). Right Plot: Estimation (first 1000 points) and Validation (remaining 1000 points) Output Data. Gómez 37 Fig. Gómez 38 . LSD .C.C.Example 2: Hammerstein Model ID ((BinaryDistillation Column) Example 2: Hammerstein Model ID Binary Distillation Column) Input: reflux ratio (u) Outputs: overhead flow rate (y1) overhead methanol concentration (y2) bottom flow rate (y3) bottom methanol concentration (y4) Fig. 1980) LSD .

C. 13: True (blue) and Estimated (red) Outputs (validation data) LSD . Gómez 39 The Estimated Linear Subsystem Third order model with eigenvalues at The Estimated Nonlinear Subsystem {0.9557 . 0. 0.C.September 2005 J. 14: Estimated Nonlinear Characteristic LSD .4916.September 2005 J. Gómez 40 .Fig.9726} Third order polynomial Fig.

Example 3: Wiener Model ID ((pHNeutralization Process) Example 3: Wiener Model ID pH Neutralization Process) (u 3 ) • base: NaOH acid: HNO3 buffer: NaHCO3 • Manipulated variable: base flow rate (u1) effluent solution • Disturbances: buffer flow rate (u2) and acid flow rate (u3) • Output: pH of the effluent solution (y) Fig. 97) LSD . (Wb1 − x2 )⎥ V ⎣V ⎦ T 1 ⎡1 ⎤ p ( x) = ⎢ (Wa 2 − x1 ).C. 92.September 2005 J.September 2005 J. Gómez 42 T . Gómez 41 Simulation Model based on first principles (introducing two reaction invariants for each inlet stream) & x = f ( x) + g ( x)u1 + p ( x)u2 h ( x. x2 ] = [Wa . y ) = x1 + 10 − 10 + x2 1 + 10 pK1 − y + 10 y − pK 2 LSD . Wb ] T ∆ T T u ⎡u ⎤ f ( x) = ⎢ 3 (Wa 3 − x1 ). 15: Schematic representation of the pH Neutralization Process (Henson & Seborg. (Wb 2 − x2 )⎥ V ⎣V ⎦ 1 + 2 ×10 y − pK 2 y −14 −y h( x. 94. 3 (Wb 3 − x2 )⎥ V ⎣V ⎦ 1 ⎡1 ⎤ g ( x) = ⎢ (Wa1 − x1 ). y ) = 0 where x =[x1 .C.

0122 z + 0.9466 z 2 + 2. 17: Estimated Nonlinear Characteristic.8940 z − 0. LSD .C.September 2005 J.0319 yk + 0.Estimation and Validation data Fig.C. Gómez 44 .0358 yk + 0. Gómez 43 The Estimated Linear Subsystem Third order model ˆ G (z ) = 0.9474 The Estimated Nonlinear Subsystem Third order polynomial 3 2 ˆ N −1 ( yk ) = 0.9989 yk Fig.006 z 3 − 2. 16: Estimation (first 1000 points) and validation (remaining 600 points) input-output data.September 2005 J. LSD .0062 z 2 − 0.

• The proposed methods generalize all the families of linear subspace methods to this class of nonlinear models. • The proposed methods make use of a standard (linear) subspace method followed by a 2-norm minimization problem which is solved via an SVD.September 2005 J. LSD . Gómez 46 .September 2005 J. LSD .C. 18: True (blue) and estimated (red) Output (Estimation/Validation data).C. • The method provides consistent estimates under the same conditions on persistency of excitation required by the (linear) subspace method used as the first step of the algorithm. • The estimated models are in a format which is suitable for their use in standard (linear) Model Predictive Control schemes. Gómez 45 Conclusions Conclusions • New subspace methods for the simultaneous identification of the linear and nonlinear parts of multivariable Hammerstein and Wiener models have been presented.Validation results Fig.

ar LSD .unr.edu. Universidad Nacional de Rosario ARGENTINA jcgomez@fceia.C.Research Seminar Subspace Identification of Hammerstein and Wiener Models Speaker Juan C. Gómez Laboratory for System Dynamics and Signal Processing FCEIA.September 2005 J. Gómez 47 .

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue listening from where you left off, or restart the preview.

scribd