42
www.zacher-automation.de 2016 Copyright S. Zacher 1 Prof. Dr. S. Zacher Automation-Letter Nr. 17 31.03.2016 There are known two approaches to the design of adaptive systems: a) Identification of the unknown plant with the model, which is build like an adaptive filter. Such systems consist from the model and the algorithm, which minimized the error between output of plant and model by the same inputs. The controller is not involved to the identification, so it is a kind of the parametric adaptive control. b) The unknown plant parameters are constants (LTI), the controller itself is also an LTI-process with adjustable gain and lag. The assumption is, that the plant parameters are constant long enough for the steps, measurements and the step responses to achieve the controller. This kind of adaptive control will be called feedback adaptive control. Identification: Estimation of unknown plant-parameters with an error-minimization S. Zacher: Adaptive Control. 2013, Script, Hochschule Darmstadt, FB EIT, MSE

Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

  • Upload
    others

  • View
    3

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 1

Prof. Dr. S. Zacher

Automation-Letter Nr. 17

31.03.2016

There are known two approaches to the design of adaptive systems:

a) Identification of the unknown plant with the model, which is build like an adaptive filter. Such systems consist from the model

and the algorithm, which minimized the error between output of plant and model by the same inputs. The controller is not

involved to the identification, so it is a kind of the parametric adaptive control.

b) The unknown plant parameters are constants (LTI), the controller itself is also an LTI-process with adjustable gain and lag. The

assumption is, that the plant parameters are constant long enough for the steps, measurements and the step responses to

achieve the controller. This kind of adaptive control will be called feedback adaptive control.

Identification: Estimation of unknown plant-parameters with an error-minimization

S. Zacher: Adaptive Control. 2013, Script, Hochschule Darmstadt, FB EIT, MSE

Page 2: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 2

Abstract, Urheberrechts- und Haftungshinweis

This publication has copyright. All rights are reserved by S. Zacher. The further development or use of the

publication without reference of author is not allowed.

For the purposes of this publication, in industry, in the laboratory and in other practical cases as well as for

any damage that may arise over the dynamic systems of incomplete or incorrect information, the author

assumes no liability.

Described is the identification of the unknown plant with algorithms, which minimized the error between output

of plant and model by the same inputs.

Generally is the output of the plant-model is .

The unknown parameters could be calculated from the following equation, if the inverse matrix M-1 exists and if

the output of the model describes the plant without error e (white noise):

If there is a white noise on the output of the plant, the model equation includes the error:

The minimization of the least mean error leads to the expression

The inverse of this matrix equation is known as Pseudoinverse or Moore -Penrose Inverse. For this equaition

is the MATLAB-command pinv to be applied, which includes also the function arx of the System-Identification-

Toolbox. Instead of pseudoinverse it is possible to use so called RLS-algorithm, which is described in this letter

with examples and solutions.

Px

M

xP 1

M

PxE plant

M

XTT

MMMP1)(

Page 3: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 3

I N H A L T :

Estimation of plant-parameters with an error-minimization …………………….… page 4

1. SLE (Solution of Linear Equations) ………………………………………………. page 8

2. LMS (Least Mean Squares) ………………………………………………………... page 24

3 RLS (Recursive Least Squares) ……..…………………………………………….. page 38

4. Exercises ……………………………………………………………………………….page 42

Page 4: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 4

Estimation of plant-parameters with an error-minimization

Generally are known three identifications-methods for a plant separated from the control system:

a) Theoretical description with the experimental adjustment of model parameters

With knowledge about physical nature of the plant you describe the plant with differential equations or systems

differential equations. Then you solve this equations for known inputs (step, pulse, wave) with known methods

and adjust the model parameters to the experimental measurements:

in time domain

in s-domain by Laplace-Transformation

In frequency domain .

b) Experimental Identification upon supposed plant dynamics

Without knowledges about physical nature of the plant you choose the model of the plant like one of the

followings: transfer function, recursive equation, Volterra-model, ANN (artificial neural network) etc. The you do

the same as in previous point, i.e. you apply inputs, measure outputs and adjust the model parameters to the

experimental measurements.

http://www.zacher-international.com/Automation_Letters/03_Hinweise_Identifikation.pdf

c) Block-Box Appoach

Without knowledges about physical nature of the plant and without supposing the model its dynamics you

apply inputs, measure outputs and adjust the vector of model parameters to minimize the difference between

plant and model.

In the following are some methods of last approach (identification with an error-minimization) based on

Pseudoinverse or Moore-Penrose Inverse described:

1. SLE (Solution of lineare equations)

2. LMS (Least Mean Squars)

3. RLS (Recursive Least Squars or recursive LMS)

Page 5: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 5

Estimation of plant-parameters with an error-minimization

How to win the information from the input step and step

response is shown below for k = N = 7.

The system of the equations above could be shown in

the vector and matrix form:

Page 6: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 6

Estimation of model-parameters with an error-minimization

Page 7: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 7

Estimation of model-parameters with an error-minimization

Instead of pseudoinverse it is possible to use so called RLS-algorithm, which will be described in the following

examples.

Page 8: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 8

1 SLE (Solution of Linear Equations)

Page 9: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 9

1 SLE (Solution of Linear Equations)

Page 10: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 10

1 SLE (Solution of Linear Equations)

Page 11: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 11

1 SLE (Solution of Linear Equations)

Page 12: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 12

1 SLE (Solution of Linear Equations)

Page 13: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 13

1 SLE (Solution of Linear Equations)

Page 14: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 14

1 SLE (Solution of Linear Equations)

Page 15: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 15

1 SLE (Solution of Linear Equations)

Page 16: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 16

1 SLE (Solution of Linear Equations)

Page 17: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 17

1 SLE (Solution of Linear Equations)

Page 18: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 18

1 SLE (Solution of Linear Equations)

Page 19: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 19

1 SLE (Solution of Linear Equations)

Page 20: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 20

1 SLE (Solution of Linear Equations)

Page 21: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 21

1 SLE (Solution of Linear Equations)

Page 22: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 22

1 SLE (Solution of Linear Equations)

Page 23: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 23

1 SLE (Solution of Linear Equations)

Page 24: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 24

2 LMS (Least Mean Squares)

Page 25: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 25

2 LMS (Least Mean Squares)

Page 26: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 26

2 LMS (Least Mean Squares)

Page 27: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 27

2 LMS (Least Mean Squares)

Page 28: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 28

2 LMS (Least Mean Squares)

Page 29: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 29

2 LMS (Least Mean Squares)

Page 30: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 30

2 LMS (Least Mean Squares)

Page 31: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 31

2 LMS (Least Mean Squares)

Page 32: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 32

2 LMS (Least Mean Squares)

Page 33: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 33

2 LMS (Least Mean Squares)

Page 34: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 34

2 LMS (Least Mean Squares)

Page 35: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 35

2 LMS (Least Mean Squares)

Page 36: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 36

2 LMS (Least Mean Squares)

Page 37: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 37

2 LMS (Least Mean Squares)

Page 38: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 38

3 RLS (Recursive Least Squares)

Page 39: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 39

3 RLS (Recursive Least Squares)

Page 40: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 40

3 RLS (Recursive Least Squares)

Page 41: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 41

3 RLS (Recursive Least Squares)

Page 42: Identification: Estimation of unknown plant-parameters ... · 2016 Copyright S. Zacher 4 Estimation of plant-parameters with an error-minimization Generally are known three identifications-methods

www.zacher-automation.de 2016 Copyright S. Zacher 42

4 Exercises

Solutions: www. zacher-international.com/Automation_Letters/LMS_Solutions.zip