NECESSARY AND SUFFICIENT CONDITIONS FOR STATE-SPACE NETWORK REALIZATION. Philip E. Paré Masters Thesis Defense

Similar documents
Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science : MULTIVARIABLE CONTROL SYSTEMS by A.

ECEN 420 LINEAR CONTROL SYSTEMS. Lecture 6 Mathematical Representation of Physical Systems II 1/67

Problem Set 4 Solution 1

4. Matrix inverses. left and right inverse. linear independence. nonsingular matrices. matrices with linearly independent columns

1 Controllability and Observability

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

New concepts: rank-nullity theorem Inverse matrix Gauss-Jordan algorithm to nd inverse

Optimization problems on the rank and inertia of the Hermitian matrix expression A BX (BX) with applications

Matrices: 2.1 Operations with Matrices

This lecture: basis and dimension 4.4. Linear Independence: Suppose that V is a vector space and. r 1 x 1 + r 2 x r k x k = 0

Linear Equations in Linear Algebra

Linear System Theory

Linear Matrix Inequality (LMI)

Control Systems

A STRATEGY FOR IDENTIFICATION OF BUILDING STRUCTURES UNDER BASE EXCITATIONS

EE/ACM Applications of Convex Optimization in Signal Processing and Communications Lecture 4

Denis ARZELIER arzelier

Matrix Arithmetic. a 11 a. A + B = + a m1 a mn. + b. a 11 + b 11 a 1n + b 1n = a m1. b m1 b mn. and scalar multiplication for matrices via.

Lec 6: State Feedback, Controllability, Integral Action

BALANCING-RELATED MODEL REDUCTION FOR DATA-SPARSE SYSTEMS

Linear Systems. Linear systems?!? (Roughly) Systems which obey properties of superposition Input u(t) output

Linear Algebra. Linear Algebra. Chih-Wei Yi. Dept. of Computer Science National Chiao Tung University. November 12, 2008

Kalman Decomposition B 2. z = T 1 x, where C = ( C. z + u (7) T 1, and. where B = T, and

8 A First Glimpse on Design with LMIs

Stability of Parameter Adaptation Algorithms. Big picture

Control Systems. Frequency domain analysis. L. Lanari

ENGI Parametric Vector Functions Page 5-01

Fast Convolution; Strassen s Method

Intro. Computer Control Systems: F9

MATH 323 Linear Algebra Lecture 12: Basis of a vector space (continued). Rank and nullity of a matrix.

MAE 143B - Homework 8 Solutions

1.6 and 5.3. Curve Fitting One of the broadest applications of linear algebra is to curve fitting, especially in determining unknown coefficients in

The Four Fundamental Subspaces

Lifted approach to ILC/Repetitive Control

Lecture 4 and 5 Controllability and Observability: Kalman decompositions

Raktim Bhattacharya. . AERO 632: Design of Advance Flight Control System. Norms for Signals and Systems

Intro. Computer Control Systems: F8

Assessment Exemplars: Polynomials, Radical and Rational Functions & Equations

LECTURE 6: VECTOR SPACES II (CHAPTER 3 IN THE BOOK)

Math 31CH - Spring Final Exam

NPTEL >> Mechanical Engineering >> Modeling and Control of Dynamic electro-mechanical System Module 4- Lecture 31. Observer Design

On some interpolation problems

Control Systems. Laplace domain analysis

Advanced Control Theory

Analysis of Systems with State-Dependent Delay

Lecture 1. Finite difference and finite element methods. Partial differential equations (PDEs) Solving the heat equation numerically

ẋ n = f n (x 1,...,x n,u 1,...,u m ) (5) y 1 = g 1 (x 1,...,x n,u 1,...,u m ) (6) y p = g p (x 1,...,x n,u 1,...,u m ) (7)

Efficient robust optimization for robust control with constraints Paul Goulart, Eric Kerrigan and Danny Ralph

Parametric Model Order Reduction for Linear Control Systems

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

Chapter 1: Systems of Linear Equations

7 : APPENDIX. Vectors and Matrices

Introduction to Nonlinear Control Lecture # 4 Passivity

DSC HW 4: Assigned 7/9/11, Due 7/18/12 Page 1

Modern Optimal Control

An LMI Approach to the Control of a Compact Disc Player. Marco Dettori SC Solutions Inc. Santa Clara, California

Affine transformations

y= sin3 x+sin6x x 1 1 cos(2x + 4 ) = cos x + 2 = C(x) (M2) Therefore, C(x) is periodic with period 2.

EEE582 Homework Problems

State Feedback and State Estimators Linear System Theory and Design, Chapter 8.

Transfer function and linearization

Nonlinear Control Systems

[2] (a) Develop and describe the piecewise linear Galerkin finite element approximation of,

j=1 x j p, if 1 p <, x i ξ : x i < ξ} 0 as p.

Math 240, 4.3 Linear Independence; Bases A. DeCelles. 1. definitions of linear independence, linear dependence, dependence relation, basis

KRONECKER PRODUCT AND LINEAR MATRIX EQUATIONS

CDS Solutions to the Midterm Exam

EE 380. Linear Control Systems. Lecture 10

CSL361 Problem set 4: Basic linear algebra

Week 9-10: Recurrence Relations and Generating Functions

LECTURE 16: LIE GROUPS AND THEIR LIE ALGEBRAS. 1. Lie groups

Representing Structure in Linear Interconnected Dynamical Systems

Categorical techniques for NC geometry and gravity

Zeros and zero dynamics

Introduction to Modern Control MT 2016

ENGG5781 Matrix Analysis and Computations Lecture 9: Kronecker Product

7.4: Integration of rational functions

Subrings and Ideals 2.1 INTRODUCTION 2.2 SUBRING

On Input Design for System Identification

EE263: Introduction to Linear Dynamical Systems Review Session 2

5. Orthogonal matrices

Topic # Feedback Control

YORK UNIVERSITY. Faculty of Science Department of Mathematics and Statistics MATH M Test #1. July 11, 2013 Solutions

Signal Structure for a Class of Nonlinear Dynamic Systems

INVERSE MODEL APPROACH TO DISTURBANCE REJECTION AND DECOUPLING CONTROLLER DESIGN. Leonid Lyubchyk

EE221A Linear System Theory Final Exam

YORK UNIVERSITY. Faculty of Science Department of Mathematics and Statistics MATH M Test #2 Solutions

Math 515 Fall, 2008 Homework 2, due Friday, September 26.

Weak Monotonicity of Interval Matrices

A = u + V. u + (0) = u

Linear Algebra 1 Exam 2 Solutions 7/14/3

1 Continuous-time Systems

Mathematics 136 Calculus 2 Everything You Need Or Want To Know About Partial Fractions (and maybe more!) October 19 and 21, 2016

Nonlinear Control. Nonlinear Control Lecture # 6 Passivity and Input-Output Stability

MEM-BUS ABSOLUTE ENCODER

ECE504: Lecture 9. D. Richard Brown III. Worcester Polytechnic Institute. 04-Nov-2008

A Benders Algorithm for Two-Stage Stochastic Optimization Problems With Mixed Integer Recourse

Matrix operations Linear Algebra with Computer Science Application

Methods in Computer Vision: Introduction to Matrix Lie Groups

Model Order Reduction for Parameter-Varying Systems

Transcription:

NECESSARY AND SUFFICIENT CONDITIONS FOR STATE-SPACE NETWORK REALIZATION Philip E. Paré Masters Thesis Defense

INTRODUCTION

MATHEMATICAL MODELING

DYNAMIC MODELS A dynamic model has memory

DYNAMIC MODELS A dynamic model has memory There are different ways to model dynamic systems Black Box Representation u u x y y u 2 u 2 x 2 y 2 y 2 u 3 u 3 x 3 y 3 y 3

DYNAMIC MODELS A dynamic model has memory There are different ways to model dynamic systems State Machine Representation u x y x 4 u 2 x 2 y 2 x 5 x 6 x 7 u 3 x 3 y 3 x 8

STATE SPACE MODELS OF LINEAR TIME INVARIANT (LTI) SYSTEMS u x y x 4 u 2 x 2 y 2 x 5 x 6 x 7 u 3 x 3 y 3 x 8 x Ax Bu y Cx Du

TRANSFER FUNCTION REPRESENTATION OF LTI SYSTEMS u u x y y u 2 u 2 x 2 y 2 y 2 u 3 u 3 x 3 y 3 y 3 G(s) C(sI A) B D

MANY TO ONE RELATIONSHIP D D CX C XB B XAX A ˆ ˆ ˆ ˆ D B A si C D B A si C s G ˆ ˆ ˆ) ˆ( ) ( ) (

NETWORK REALIZATION Assume input-output data is produced by the system (A, B, C, D) then what must be known about this system to recover the whole thing from input-output data

NETWORK REALIZATION Assume input-output data is produced by the system (A, B, C, D) then what must be known about this system to recover the whole thing from input-output data

PROBLEM FORMULATION

PARAMETERIZATION The parametrization of the system matrices, is a continuously differentiable function P α : Ω R q R N where q is the number of unknown parameters and N = n n + m + p + mp, the number of parameters in the system matrices.

PARAMETERIZATION The parametrization of the system matrices, is a continuously differentiable function P α : Ω R q R N where q is the number of unknown parameters and N = n n + m + p + mp, the number of parameters in the system matrices.

PARAMETERIZATION The parametrization of the system matrices, is a continuously differentiable function P α : Ω R q R N where q is the number of unknown parameters and N = n n + m + p + mp, the number of parameters in the system matrices. For a β = P α R P R N N mp N mp N nm n nm n n n n D C B A 2 ) (, ) (, ) (, ) ( 2 2 2 2

IDENTITY PARAMETERIZATION The identity parameterization I θ : R N R N maps θ to itself and is a vectorization of the system matrices

IDENTITY PARAMETERIZATION The identity parameterization I θ : R N R N maps θ to itself and is a vectorization of the system matrices x y u x x

IDENTITY PARAMETERIZATION The identity parameterization I θ : R N R N maps θ to itself and is a vectorization of the system matrices x y u x x

IDENTITY PARAMETERIZATION The identity parameterization I θ : R N R N maps θ to itself and is a vectorization of the system matrices x y u x x I( )

IDENTITY PARAMETERIZATION The identity parameterization I θ : R N R N maps θ to itself and is a vectorization of the system matrices x y u x x I( )

RESTRICTION OF THE IDENTITY Assuming elements of (A, B, C, D) are known can restrict the identity parameterization

RESTRICTION OF THE IDENTITY Assuming elements of (A, B, C, D) are known can restrict the identity parameterization Let θ R R N encode the known information and the indicator function Θ R be a binary vector indicating the known elements and let the affine restriction of the identity parameterization be represented by I θr

CONSISTENCY Consider a system (A, B, C, D) and let θ be the vectorization of the system matrices (A, B, C, D). An affine restriction of the identity parametrization I θr is consistent with θ if θ R(I θr )

CONSISTENCY Consider a system (A, B, C, D) and let θ be the vectorization of the system matrices (A, B, C, D). An affine restriction of the identity parametrization I θr is consistent with θ if θ R(I θr )

GLOBAL IDENTIFIABILITY Let P α : Ω R q R N be a parameterization of system matrices (A, B, C, D). The parameterization P(α) is globally identifiable from the transfer function G(s) if for all α, α 2 Ω if G s = C α (si A α ) B α = C α 2 (si A α 2 ) B α 2 then α = α 2

NETWORK REALIZATION PROBLEM Consider a system (A, B, C, D), with A, B controllable and A, C observable and suppose G s = C(sI A) B + D is given. Find an affine restriction of the identity parameterization, consistent with A, B, C, D, that is globally identifiable from G(s)

SOLUTION TO NETWORK REALIZATION PROBLEM

SUFFICIENT CONDITION AX ˆ XB CX ˆ XA Bˆ C

SUFFICIENT CONDITION C CX B XB XA AX ˆ ˆ ˆ ) ( ˆ) ( ˆ ˆ C vec B vec x C I I B A A n n T T x b A

ANOTHER SUFFICIENT CONDITION X Aˆ AX BX ˆ B CX Cˆ

ANOTHER SUFFICIENT CONDITION C CX B BX AX A X ˆ ˆ ˆ ˆ) ( ) ( ˆ ˆ C vec B vec x C I I B A A n n T T x b A

COMBINING SUFFICIENT CONDITIONS Let x = A b, x = A b, dim N A dim N A = l. = k, and

COMBINING SUFFICIENT CONDITIONS Let x = A b, x = A b, dim N A dim N A = l. A = k, and N(A) N(A T ) x R(A T ) b R(A)

COMBINING SUFFICIENT CONDITIONS Let x = A b, x = A b, dim N A dim N A = l. A = k, and N(A) N(A T ) R(A T ) R(A) x b

COMBINING SUFFICIENT CONDITIONS

COMBINING SUFFICIENT CONDITIONS Let x = A b, x = A b, dim N A dim N A = l. = k, and Also let N A = span{x n,, x nk } and N A = span{x n,, x n l }, where x ni = vec(x ni ).

MAIN RESULT Given a proper G(s), with two minimal realizations, S = (A, B, C, D) and S = (A, B, C, D) Suppose part of S is known, specified by the entries in θ R R N and the indicator function Θ R, characterizing an affine restriction of the identity parameterization I θr Suppose all of S is known Note that I θr and S are compatible giving A, b, A, b, X, X ni, X, and X nj for i =,, k and j =,, l Then I θr is globally identifiable from G s if and only if X + w X n + + w k X nk = (X + w X n + + w l X nl ) has a unique solution (w,, w k, w,, w l ).

MAIN RESULT Given a proper G(s), with two minimal realizations, S = (A, B, C, D) and S = (A, B, C, D) Suppose part of S is known, specified by the entries in θ R R N and the indicator function Θ R, characterizing an affine restriction of the identity parameterization I θr Suppose all of S is known Note that I θr and S are compatible giving A, b, A, b, X, X ni, X, and X nj for i =,, k and j =,, l Then I θr is globally identifiable from G s if and only if X + w X n + + w k X nk = (X + w X n + + w l X nl ) has a unique solution (w,, w k, w,, w l ).

MAIN RESULT Given a proper G(s), with two minimal realizations, S = (A, B, C, D) and S = (A, B, C, D) Suppose part of S is known, specified by the entries in θ R R N and the indicator function Θ R, characterizing an affine restriction of the identity parameterization I θr Suppose all of S is known Note that I θr and S are compatible giving A, b, A, b, X, X ni, X, and X nj for i =,, k and j =,, l Then I θr is globally identifiable from G s if and only if X + w X n + + w k X nk = (X + w X n + + w l X nl ) has a unique solution (w,, w k, w,, w l ).

MAIN RESULT Given a proper G(s), with two minimal realizations, S = (A, B, C, D) and S = (A, B, C, D) Suppose part of S is known, specified by the entries in θ R R N and the indicator function Θ R, characterizing an affine restriction of the identity parameterization I θr Suppose all of S is known Note that I θr and S are compatible giving A, b, A, b, X, X ni, X, and X nj for i =,, k and j =,, l Then I θr is globally identifiable from G s if and only if X + w X n + + w k X nk = (X + w X n + + w l X nl ) has a unique solution (w,, w k, w,, w l ).

MAIN RESULT Given a proper G(s), with two minimal realizations, S = (A, B, C, D) and S = (A, B, C, D) Suppose part of S is known, specified by the entries in θ R R N and the indicator function Θ R, characterizing an affine restriction of the identity parameterization I θr Suppose all of S is known Note that I θr and S are compatible giving A, b, A, b, X, X ni, X, and X nj for i =,, k and j =,, l Then I θr is globally identifiable from G s if and only if X + w X n + + w k X nk = (X + w X n + + w l X nl ) has a unique solution (w,, w k, w,, w l ).

PHARMACOKINETICS EXAMPLE

PHARMACOKINETICS EXAMPLE

PHARMACOKINETICS EXAMPLE x y k k k 2 x k k 2 x b u

PHARMACOKINETICS EXAMPLE x y u b x k k k k k x 2 2 x y u x x.25.25.227.382 S

PHARMACOKINETICS EXAMPLE x y u x x.25.25.749.3387 x y u x x.25.25.227.382 S S

PHARMACOKINETICS EXAMPLE Assume C = [ I ] is known So for I C=[ I ] we see that k = l = 2

PHARMACOKINETICS EXAMPLE Assume C = [ I ] is known So for I C=[ I ] we see that k = l = 2 2 2.5.5.5.5.5.5.5.5.5.5 w w w w

PHARMACOKINETICS EXAMPLE Assume C = [ I ] and A :, = [.382.25] T are known So for I C,A :, we see that k = and l = 2

PHARMACOKINETICS EXAMPLE Assume C = [ I ] and A :, = [.382.25] T are known So for I C,A :, we see that k = and l = 2 X

PHARMACOKINETICS EXAMPLE Assume C = [ I ] and A :, = [.382.25] T are known So for I C,A :, we see that k = and l = 2 X x y.382.25 x.227 x.25 u

THANK YOU

THANK YOU Questions?