Modelling and Identification

Prof. Dr. Ping Zhang

Institute for Automatic Control

WS 2023/2024
 Organisational issues

 Lecture: Prof. Dr. Ping Zhang

Email: [email protected]
Office: 12/474
Appointment through Ms. Stroh
Email: [email protected]
Office: 12/476

 Exercise: M. Sc. Mengsen Jia

Email: [email protected]
Office: 12/470

 Language of the course: English

 Scope: 4 hours per week (5 ECTS-Credits)

 Script: available on OLAT (password: CDC23ATPLUS)

 Exam: written examination

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 2
 Literature (1)

 Wernstedt. Experimentelle Prozessanalyse. VEB Verlag, 1989.

 R. Isermann, M. Münchhof. Identification of Dynamic Systems.
Springer, 2011.
 T. Söderstrom, P. Stoica. System identification. Prentica Hall,
1989.
 M. Verhaegen, V. Verdult. Filtering and system identification.
Cambridge University Press, 2007.
 S. Bittanti. Model identification and data analysis. Wiley, 2019.

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 3

 Literature (2)

 J. Mikles, M. Fikar. Process modelling, identification and control.

Springer, 2007.
 C.M. Close. Modeling and analysis of dynamic systems. Wiley,
2002.
 H.E. Scherf. Modellbildung und Simulation dynamischer
Systeme: Eine Sammlung von Simulink-Beispielen, Oldenbourg
Verlag, 2009.

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 4

 Introduction

 A number of control methods have been introduced in previous

control courses.
 Many control methods (for instance, state feedback controller, MPC,
observer design, usw.) are model-based and have gained wide
acceptance in the industry.

How to get a model of the system to be controlled?

„Modelling and Identification of dynamic

systems“

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 5

 Systems

 Technical systems:
Technical systems have the purpose of transformation, transportation
and/or storage of matter, energy and/or information

 Non-technical systems (e.g. biological systems, economic systems)

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 6

 Systems

 Viewpoint of control engineers:

inputs outputs
System

 Static systems:
The output signals depends only on the present input signal values.

 Dynamic systems:
The output signals depends not only on the present input signal
values but also on the past values.

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 7

 Modelling and identification

 Objective:
Find a mathematical model in form of mathematical equations to
describe the system

 Requirement on a model:
The model should describe certain aspects of the real system
behaviour with sufficient accuracy.
- validity / accuracy
- complexity

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 8

 Classification of models

 Transfer function
 Differential equation
 Difference equation
 State space model
 Automaton
 Petri nets
 Boolean networks

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 9

 Classification of systems

System Characteristics Model

Continuous-time systems Signals depend Ordinary differential

with concentrated only on time . equations (ODE)
parameters

Continuous-time systems Signals Partial differential

with distributed depend not only on time equations (PDE)
parameters , but also on location
represented by space
coordinates.

Discrete-time systems Signals Difference equations

are functions of
discrete-time

Discrete event systems The state changes only Petri nets, automata
at discrete instances,
when certain events
occur.

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 10

 Classification of systems

System Characteristics Model example

Time-invariant Model parameters doesn‘t
system change with time

Time-varying Model parameters change with

system time

Two special classes of time-varying systems that are much discussed:

System Characteristics Model example

Periodic system Model parameters
change with time
periodically

Parameter-varying Model parameters

system depends on some
parameter vector

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 11

 Classification of systems

System Characteristics Model example

Linear systems The principle of superposition is

satisfied
u (t )  u1 (t )  u2 (t )  y (t )  y1 (t )  y2 (t )
u (t )  ku1 (t )  y (t )  ky1 (t )

Nonlinear The principle of superposition is

systems not satisfied.

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 12

 Models of systems

The models that are used to describe a dynamic system depend on

the modelling objectives!

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 13

 Models of systems

Example: Vehicle distance control

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 14

 Models of systems

Example: Lateral Dynamic Control of Cars

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 15

 Modelling strategies

Depending on which information of the real system is available for

modelling, there are different modelling strategies.

Information Modelling strategy

Theoretical working principles of Theoretical modelling

the system (called also first-principle modelling)

Data collected during experiments Experimental modelling

or during system normal operation (called also identification)

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 16

 Modelling strategies

Strategy 1: Theoretical modelling

Pump
Physical principles

Inflow
dh
Mass balance: A  Qinflow  Qoutflow
dt

h
Outflow Torricelli‘ law: Qoutflow  aA0 2 gh

𝐴: Cross-section area of the tank

𝐴 : Cross-section area of the outflow pipe
𝑎, 𝑔: Konstante dh
ℎ: water level A  Qinflow  aA0 2 gh
𝑄 ,𝑄 : volume flow rate dt

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 17

 Modelling strategies

Strategy 2: Experimental modelling

Do experiments !
Pump
Measurements
FI Flow rate
Inflow

0 t
Water level
LI Outflow
h

0 t
Step response
 first order system

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 18

 Modelling strategies

Example: One-tank system

Theoretical modelling Experimental modelling

dh Flow rate Water level

Mass balance: A  Qinflow  Qoutflow
dt

Torricelli‘ law: Qoutflow  aA0 2 gh 0 t 0 t

dh K
A  Qinflow  aA0 2 gh H ( s)  Qin flow ( s )
dt Ts  1

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 19

 Modelling strategies

Experimental
Theoretical
modelling
modelling
(identification)
Information Physical and chemical Experiments on the system
Basis principles (e.g. balance and measurements of system
equations) inputs and outputs
Results Physical parameteric Input-output models
models (often state (often differential equations)
equations)

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 20

 Modelling strategies

Theoretical modelling
 Advantages:
• Good reconstruction of system structure
• Model parameters are related to system components
• Applicable over a wide range of operating conditions
 Disadvantages:
• Typically the model includes some model parameters that
are not readily available
• Not applicable, if physical principles are not clear
• For complex systems, it is often rather time-consuming and
the resulting model may be too complex.

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 21

 Modelling strategies

Experimental modelling (identification)

 Advantages:
• It is easier to develop
 Disadvantages:
• It doesn‘t describe the internal structure of the system
• Should be used with caution for operating conditions that
were not included in the experimental data
• Sometimes, experiments on the real system or some special
test input signals are not allowed.
• Needs much knowledge of control theory

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 22

 Basic procedure of modelling

 Specify the goal of modeling

 Understand the working mechanism of the system
 In case of theoretical modelling, set up the system equations.
In case of experimental modelling, do experiments, collect the
input and output data and then apply the identification
approach
 Validate the model

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 23

 Application areas of models

 Gain knowledge about the plant behaviour

 Model-based controller design (e.g. IMC, MPC, Robust control, etc)
 Fine-tuning of controller parameters
 Model-based process supervision and fault diagnosis
 Model-based prediction of signals
 Training simulator (train the operators, support the operators, judge
the effect of intervention)
 Real-time optimization (find optimal operating point by means of
mathematical optimization techniques)

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 24

 Scientific communities

 IFAC (International federation of Automatic Control) Technical

Committee 1.1 „Modelling, Identification and Signal Processing”
 GMA Technical Committee 1.30 „Modellbildung, Identifizierung
und Simulation in der Automatisierungstechnik“
 An important topic in the field of control theory and control
engineering
 Numerous publications in scientifc journals and conferences
 Get-Together: IFAC Symposium on System Identification (triennial
event)

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 25

 Organisation of this course

Chapter 1: Introduction
Chapter 2: Theoretical Modelling
Chapter 3: Experimental modelling
Chapter 4: Least-Squares methods
Chapter 5: Prediction error methods
Chapter 6: Instrumental variable methods
Chapter 7: Subspace identification methods (SS model!)
Chapter 8: Some practical aspects

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 26

 Organisation of this course

Chapter 1
Introduction

Chapter 2 Chapter 3
Theoretical Modelling Experimental Modelling

Chapter 4
Least-Squares methods

Chapter 5 Chapter 6
Prediction error methods Instrumental variable methods

Chapter 7
Subspace identification

Chapter 8
Some practical aspects

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 27

 Learning goals

After the course, you should

 Be familiar with the basic principles of important modelling and

identification approaches
 Be able to derive first-principle model of systems with moderate
complexity
 Be able to apply a suitable approach to identify the required model of
a given process with the help of Matlab toolboxes
 Be able to organize an identification project (data collection, pre-
processing of data, etc)

Lehrstuhl für Automatisierungstechnik Ping Zhang, [email protected] 28