D.E.

Rivera, Introduction to System Identification, ChE 494/598, January 20, 2004

ChE 494/598 Course Objectives

Introduction to System • Present fundamental background to allow students to
Identification make judicious choices of design variables in system
identification.
Daniel E. Rivera, Associate Professor
• Provide lab exercises that will give students a working feel
Control Systems Engineering Laboratory
Department of Chemical and Materials Engineering for the course topics. MATLAB (particularly the System
Arizona State University Identification Toolbox) will be the program of choice.
Tempe AZ 85287-6006

[email protected] • Provide a glimpse of cutting-edge identification research at

(480) 965-9476 ASU and other academic institutions around the world.
© Copyright, 1998-2004

System Identification Some System Identification Facts

“Identification is the determination, on the basis of input and output, of
a system within a specified class of systems, to which the system under
test is equivalent.”
- L. Zadeh, (1962) • problem not exclusively associated with control design,
Disturbances although it forms a significant part of control implementation

• often times, the system identification task is the most

Inputs Outputs expensive and time consuming part of advanced control
System implementation

• broadly applicable technology with applications in many

diverse fields

System identification focuses on the modeling of dynamical

systems from experimental data

© Copyright 1998-2004 by D.E. Rivera, All Rights Reserved

1
D.E. Rivera, Introduction to System Identification, ChE 494/598, January 20, 2004

Shell Heavy Oil Fractionator Example

PC
Distillation Column Data
T • Manipulate top,
LC side draw and/or OUTPUT #1
Upper Reflux Duty FC
Top Draw 4
bottoms reflux
UPPER REFLUX
duty to maintain 2
A
TOP DRAW top and side 0
Top endpoints at
-2
T
Endpoint setpoint, 0 20 40 60 80 100 120 140 160 180 200
Intermediate Reflux
Side Draw • Reject
Duty INTERMEDIATE REFLUX
INPUT #1
T
SIDE
STRIPPER
disturbances from
LC FC
the upper and
T
intermediate reflux 0
Bottoms A duties.
SIDE DRAW
Reflux Duty Side
BOTTOMS REFLUX
0 20 40 60 80 100 120 140 160 180 200
Endpoint • Keep Bottoms
T
Q(F,T)
CONTROL Reflux
LC
Temperature • response of overhead temperature (top) to changes in reflux
Bottoms
Reflux Temp
F T
above constraints. flowrate (bottom)
BOTTOMS
FEED

Epi Reactor Temperature Control Epi Reactor Identification Data

• keep center, front, side and rear temperatures constant by
adjusting power to the lamp banks 0
solid:master; dashed:side; dotted:front; d-dotted:rear

-2

Power [%]
-4

-6

-8
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Time [seconds]
solid:master; dashed:side; dotted:front; d-dotted:rear
10

Temp. Deviation [C]

0

-10

-20

-30

-40
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Time [seconds]

solid: center; dashed: side; dotted:front; dash-dotted: rear

© Copyright 1998-2004 by D.E. Rivera, All Rights Reserved

2
D.E. Rivera, Introduction to System Identification, ChE 494/598, January 20, 2004

Semiconductor Mfg Supply Chain Managment

Fab/Sort starts
Fab/Test Node Dynamic Response
Controller

Load
LT

ADI A/T starts Forecast

Demand

Outs
LT Real

SFGI Shipments
A/T: Assembly/Test Facility

Starts
ADI: Assembly-Die Inventory LT

SFGI: Semi-Finished Goods Inventory CW Demand

CW: Component Warehouse

D1 t Time
D3 D2

Wing Flutter Example Other Challenging Application Areas

Filtered data used for modeling
10

5
Excitation

0 • Economic/financial systems
-5

-10
• modeling economic indicators such as the Dow Jones,
0 1 2 3 4
Time [s]
5 6 7 8
S&P 500 indices
Filtered data used for modeling
5

• Behavioral/social systems
Response

0
• time-varying adaptive interventions for the prevention of
chronic, relapsing disorders (such as alcoholism, smoking
-5
0 1 2 3 4
Time [s]
5 6 7 8 and drug abuse)
• artificial mechanical vibrations (top) introduced to a wing at certain flight
conditions; responses shown on bottom

© Copyright 1998-2004 by D.E. Rivera, All Rights Reserved

3
D.E. Rivera, Introduction to System Identification, ChE 494/598, January 20, 2004

STAGES OF SYSTEM IDENTIFICATION

Stages of System Identification Start a priori process
information

Experimental Design
and Execution
• Experimental Design and Execution
( Step, Pulse, or PRBS-Generated Data)

"Identification"
• Data Preprocessing • Data Preprocessing
• Model Structure Determination
• Parameter Estimation
• Model Structure Selection
( Linear Plant and Disturbance Models)

• Parameter Estimation Model Validation

(Simulation, Residual auto and

cross- correlation,
• Model Validation step-response)
No
Does the model
meet validation criteria?

Yes
End

Stages of System Identification - II

Keys to Successful System Identification
Prior system knowledge: physics, linguistics, first-hand, etc.

Experiment
design in Practice
Pre-treat
data
Choose
• Understanding the various identification methods and
model associated decision variables in terms of bias-variance
structure
Choose tradeoffs
performance
criterion

Parameter estimation

Validate
• Effective use of a priori knowledge regarding the system
model Not OK revise! to be identified and the intended application (e.g.,
Not OK revise prior?
OK accept model! simulation, prediction, control)
Controller Design & Commissioning

"the classical statistical approach," per Ljung...

• courtesy P. Lindskog, ISY, Linköping University, Sweden

© Copyright 1998-2004 by D.E. Rivera, All Rights Reserved

4
D.E. Rivera, Introduction to System Identification, ChE 494/598, January 20, 2004

Furnace Control Example

System Identification Challenges

• Skill-level issues: many system identification methods

assume the user has extensive background in statistics, (Output)
signal processing, discrete-time systems, and optimization. (Disturbance)

• Large number of design variables. CONTROLLER

• Process operating restrictions make identification one of the

most time consuming tasks in advanced control (Input)
implementation projects.
Objective: Use fuel gas flow to keep outlet temperature under control, in spite
of significant changes in the feed flowrate.

The "Shower Problem" Graphical System Identification Using Step Testing

Consider the problem of adjusting hot water flow to maintain Response of a first-order with deadtime model for a step input of magnitude A
shower temperature despite cold water fluctuations...
τ

KA
Transportation lag -θs
Makes this a difficult p(s) = K e ,
τ s+1
control problem...

Hot Cold Time

θ

Many references for this technique, example: Seborg, Edgar, and Mellichamp,
Process Dynamics and Control, Wiley, 1989, Chapter 7.

© Copyright 1998-2004 by D.E. Rivera, All Rights Reserved

5
D.E. Rivera, Introduction to System Identification, ChE 494/598, January 20, 2004

Furnace Example (Continued) Perils of Step Testing

Measured Output
25 Temperature Response to a Step Increase in Fuel Gas Flow
20 Open-loop disturbance
15 (no control) Compare Step Responses: FOPDT Model[--], PLANT Data[-], TRUE PLANT[-.]
-5.5
10
Temperature
5
(Controlled -6 exp(-5s)
Variable) p(s) (true) = --------------
0 -6.5
0 200 400 600 800 1000
Time[Min]
1200 1400 1600 1800 2000
10 s + 1
Input -7
1
Fuel Gas Flow -7.5
0.5
(Manipulated Variable) -8
0
-8.5 0.748 exp(-6s)
-0.5
p(s) (est) = --------------------
-9
-1 7.095 s + 1
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Time[Min] 2000 minutes -9.5
40 minutes
-10
Consider the application of step testing to a system subject 0 5 10 15 20 25 30 35 40
Time[Min]
to a drifting, nonstationary disturbance.

Principal Sources of Error in System Identification

Design Variable Selection Issues
• Input Signal Selection: Random Binary, ERROR = BIAS + VARIANCE
Pseudo-Random Binary, or Multiple Step/Pulse • BIAS. Systematic errors caused by
Inputs?
- input signal characteristics (i.e., excitation)
• Input Signal Parameters: Example: PRBS - number of
shift registers, switching time, signal magnitude, and - choice of model structure
signal duration
- mode of operation (i.e., closed-loop instead of open-loop)
• Data Preprocessing: Detrending, control-relevant
prefiltering, outlier removal, etc. • VARIANCE. Random errors introduced by the presence of noise
in the data, which do not allow the model to exactly reproduce the
• Model Structure Selection and Parameter Estimation: plant output. It is affected by the following factors:
ARX, ARMAX, Output Error, Box-Jenkins
- number of model parameters
• Model Validation: Simulation, Crossvalidation,
Correlation Analysis; Examination of Step, Impulse, - duration of the identification test
Frequency Responses.
- signal-to-noise ratio

© Copyright 1998-2004 by D.E. Rivera, All Rights Reserved

6
D.E. Rivera, Introduction to System Identification, ChE 494/598, January 20, 2004

From Identification to Controller Implementation Course Outline

OPEN CLOSED
LOOP
IDENTIFICATION
LOOP
• Signals and Systems Overview
DATA
RESPONSE RESPONSE
Measured Output
25

20
• Input Signal Design and Nonparametric Estimation
15
Measured
10
Output
5

0
• Parametric Model Estimation and Validation
-5
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000
Time[Min]

15
Input • Control-Relevant and Closed-Loop Identification
10

Input 5

0 • Multivariable Identification
-5

-10

-15
0 500 1000 1500 2000 2500
Time[Min]
3000 3500 4000 4500 5000 • Issues in nonlinear and semiphysical identification
Furnace example with PRBS input, PID with filter controller

Course Focus Systems Representations

Nonlinear Lumped
• Very broad subject
Parameter System

{
Linearization

}
- Linear or Nonlinear? (Mostly) LINEAR
T
- Continuous or Discrete? DISCRETE State-Space Sampling Discrete-time
Model S-S Model Discrete-
- Parametric or nonparametric? BOTH Step/ time
Impulse Laplace Realization Step/
Response transforms Impulse
- Time or frequency domain? BOTH and Response
T
Frequency and
Response s-domain Sampling z-domain Frequency
Transfer Function Transfer Function Response

Model Model
(difference equation)

© Copyright 1998-2004 by D.E. Rivera, All Rights Reserved

7
D.E. Rivera, Introduction to System Identification, ChE 494/598, January 20, 2004

Discrete Model Representations Pulse Transfer Functions

Nonparametric computer uk u(t) y(t) T yk
Zero-order
Step control
Hold
P(s) computer
algorithm
PLANT Response
{ discrete
ZOH-equivalent Pulse Transfer Function
input
continuous
Impulse uk input
continuous
PLANT Response u(t) output discrete
output
y(t)
yk
time
U(k) Y(k+1) Difference
D time
Equations
{ U(z) Y(z)
time
time
G(z) Z-Transforms
Parametric

Examples
System Identification Structure
time s-domain z-domain
y(t) = p(z)u(t) + H(z)a(t)
Impulse δ (t) 1 1
a Random Signal
1 t ≥ 0 1 z
Step s(t) = 
0 t < 0 s z −1
First-Order ZOH Pulse H(z)
K K(1 - exp(-T/τ))
Lag Transfer Input Signal
Function τs + 1 z - exp(-T/τ) υ Disturbance Signal
u + +
P(z)
Integrating/ ZOH Pulse K KT y
Output Signal
Transfer
Ramp
Function s z −1
P(z) and H(z) are discrete-time (z-domain)
First-Order ZOH Pulse K exp(−θ s) K(1 - exp(-T/τ))z -N transfer functions
with Delay Transfer τ s + 1 z - exp(-T/τ)
Function
θ = NT

© Copyright 1998-2004 by D.E. Rivera, All Rights Reserved

8
D.E. Rivera, Introduction to System Identification, ChE 494/598, January 20, 2004

Signals Overview Example 1: White-noise signal

at xt
1
• Deterministic versus stochastic signals
x t = at at = N(0, σ 2a )
• Stationary versus nonstationary signals
1.5 AUTOCORRELATION OF WHITE NOISE 1.5 POWER SPECTRUM OF WHITE NOISE
γk
ρk =
• White versus autocorrelated signals σ 2a 1
Φ(ω ) σ 2a =1
1

rhok

phi
• Crosscorrelated versus uncorrelated signals
0.5 0.5

Mean, auto and cross-covariance, power and cross-spectra 0 0

will be the measures/tools utilized here
0 2
Lag
4 0 1 2
Frequency [Radians]
ω 3

Example 1: White Noise Signal Analysis

(From Sample Estimators)
Example 3: AR(1) signal
Signal
1.5 Signal
3
1 2

0.5 1

0
0
-1
-0.5
-2
-1
0 200 400 600 800 1000 1200 1400 1600 1800 2000 -3
Sample Number 0 200 400 600 800 1000 1200 1400 1600 1800 2000
Sample Number
Autocorrelation Coefficients 0
Power Spectral Density
1 10 Autocorrelation Coefficients Power Spectral Density
1 10 1

0.5 0
0.5 10
rhok

-1
10

rhok
0 -1
0 10

-2
-0.5 10 -2
-2 -1 0
-20 0 20 40 10 10 10 -0.5 10
-2 -1 0
Lag k Frequency [Radians/Time] -20 0 20 40 10 10 10
Lag k Frequency [Radians/Time]

© Copyright 1998-2004 by D.E. Rivera, All Rights Reserved

9
D.E. Rivera, Introduction to System Identification, ChE 494/598, January 20, 2004

System Identification, Revisited "Plant Friendly" Input Signal Design

a white noise signal
A plant friendly input signal should:
H(z) • be as short as possible
Disturbance Signal
• not take actuators to limits, or exceed move size
Input Signal υ (random, autocorrelated)
u + + restrictions
(Random or P(z)
Deterministic) y • cause minimum disruption to the controlled variables
Output Signal (random, (i.e., low variance, small deviations from setpoint)
autocorrelated)
• u and y are crosscorrelated Note that theoretical requirements may strongly conflict
• a and y are crosscorrelated with "plant-friendly" operation!
• If u and a are statistically independent, then u
and ν will be uncrosscorrelated...

Pseudo-Random Binary Sequence PRBS, continued

One cycle of the PRBS time input signal

The PRBS is a periodic, deterministic input which can be 1

generated using shift registers and Boolean algebra 0.5

0
1 Shift Registers nr
-0.5

1 0 1 1 0 1 1 0 1 -1
0 5 10 15 20 25 30 35 40 45
Time[Min]
Power Spectrum of the PRBS input

0
10

-1
10

AR
Exclusive OR -2
(Modulo 2 Adder) 10

-3
10 0
10
Test Signal Radians/Min

PRBS design for Tsampl = 1, Tsw = 3, n (registers) = 4, and signal

The main design variables are switching time (Tsw), magnitude = +/- 1.0. One cycle duration is 45 minutes long.
number of shift registers (nr), and signal amplitude

© Copyright 1998-2004 by D.E. Rivera, All Rights Reserved

10
D.E. Rivera, Introduction to System Identification, ChE 494/598, January 20, 2004

Inputs to Consider Nonparametric Methods

• Step/Pulse Inputs
• Correlation Analysis:
• Gaussian White Noise
- direct estimation of impulse response coefficients from
• Random Binary Signal (RBS) identification data

• Pseudo-Random Binary Signal (PRBS)

• Spectral Analysis:
• multi-level Pseudo-Random Signals
- direct estimation of frequency response from identification
• Multisine inputs (e.g., Schroeder-phased, minimum data
crest factor)

Correlation Analysis Results, Hairdryer Data Wing Flutter Example, Spectral Analysis
Smoothed SPA model (solid). Raw ETFE (*).
0
Covf for filtered y Covf for prewhitened u
0.15 1.5
-5

Amplitude [dB]
1
0.1 -10
0.5
-15
0.05
0
-20
0 -0.5
-20 -10 0 10 20 -20 -10 0 10 20 -25
4 5 6 7 8 9 10 11
Frequency [Hz]
Smoothed SPA model (solid). Raw ETFE (*).
0.6 Correlation from u to y (prewh) 0.15 Impulse response estimate 150

Phase [degree]
0.4 0.1
100
0.2 0.05

0 0 50

-0.2 -0.05
-20 -10 0 10 20 -20 -10 0 10 20
0
4 5 6 7 8 9 10 11
Frequency [Hz]

© Copyright 1998-2004 by D.E. Rivera, All Rights Reserved

11
D.E. Rivera, Introduction to System Identification, ChE 494/598, January 20, 2004

Prediction-Error Family
Prediction-Error Model of Models
Structures
Smoothing, Filtering, Prediction e
C(z)
D(z)
u B(z) −nk 1 y
z +
F(z) + A(z)
B(z) C(z)
A(z)y(t) = u(t − nk) + e(t)
F (z) D(z)

A(z) = 1 + a1z −1 + . . . + ana z −na

B(z) = b1 + b2z −1 + . . . + bnb z −nb+1
C(z) = 1 + c1z −1 + . . . + cnc z −nc
D(z) = 1 + d1z −1 + . . . + dnd z −nd
• In the prediction problem, current and previous measurements F (z) = 1 + f1z −1 + . . . + fnf z −nf
from the plant are used to obtain estimates k+1 (or beyond)
time steps in the future In transfer function form:
y(t) = p̃(z)u(t) + p̃e(z)e(t)
B(z) C(z)
p̃(z) = z −nk p̃e(z) =
A(z)F (z) A(z)D(z)

Popular PEM Structures ARX Parameter Estimation

Method p̃(z) p̃e(z) The one-step ahead predictor for y

ARX B(z) −nk 1 ŷ(t|t−1) = −a1y(t−1)−. . .−ana y(t−na)+b1u(t−nk)+. . .+bnb u(t−nk−nb+1)

A(z) z A(z)
B(z) −nk C(z) can be expressed as a linear regression problem via
ARMAX A(z) z A(z)
−nk ϕ = [ −y(t − 1) . . . −y(t − na) u(t − nk) . . . u(t − nk − nb + 1) ]T
FIR B(z)z 1
B(z) −nk C(z) and θ, the vector of parameter estimates:
Box-Jenkins F (z) z D(z)
B(z) −nk θ = [ a1 . . . ana b1 . . . bnb ]T
Output Error F (z) z 1
Rewriting the objective (“loss”) function as
1 N  2
min V = min y − ϕT (t)θ
B(z) C(z) θ N i=1θ

A(z)y(t) = u(t − nk) + e(t) leads to the well-established linear least-squares solution
F (z) D(z)  −1
1 N
 1 N

y(t) = p̃(z)u(t) + p̃e(z)e(t) θ̂ =  ϕ(t)ϕT (t) ϕ(t)y(t)
N t=1 N t=1

© Copyright 1998-2004 by D.E. Rivera, All Rights Reserved

12
D.E. Rivera, Introduction to System Identification, ChE 494/598, January 20, 2004

Model Validation Techniques Modeling Requirements for Process

• Simulation (plot the measured output time series versus the predicted Control
output from the model).

"Decomposed" START "Integrated/Synergistic"

• Crossvalidation (simulate on a data set different than the one used for
parameter estimation; for a number of different model structures, plot the
loss function and select the minimum.
Modeling

Modeling/
• Impulse, step, and frequency responses (compare with physical insight Control
regarding process).
Control
• Scatter Plots/correlation analysis on the prediction errors (make sure they
resemble white noise).

• Information criteria (Akaike or Rissanen's Maximum Description Length)

Same result is not obtained from both
approaches!

Control-Relevant Identification Control-Relevant Prefiltering

• Some general ideas behind control-relevant modeling

Overhead Temperature
Solid: Raw Data; Dashed: Prefiltered Data
4
2

• Design variables for control-relevant id 0

-2
– Control-relevant prefiltering 0 20 40 60 80 100 120 140 160 180 200
Time
– Control-relevant input signals 2

Reflux Flow
0
• Brief comments on uncertainty estimation from id data
-2
0 20 40 60 80 100 120 140 160 180 200
Time
• Integrated system id and PID controller design
The purpose of c-r prefiltering is to emphasize
information in the data most important for control
purposes

© Copyright 1998-2004 by D.E. Rivera, All Rights Reserved

13
D.E. Rivera, Introduction to System Identification, ChE 494/598, January 20, 2004

Problems in Closed-Loop Identification Refinery Debutanizer

d

ud CF Pd FC

υ REFLUX FLOW

+ -
r+ C u P + + + + y
+ FEED FLOW FEED TEMP

- F T

P BOTTOMS-TO-FEED DIFFERENTIAL PRESSURE

BOTTOMS TEMP

• crosscorrelation will exist between disturbance (d) FUEL GAS FLOW FC

and input (u) as a result of the control G

FUEL GAS SPECIFIC GRAVITY

MPC loop between

• control action will introduce additional bias by "eating Bottoms T

away" at excitation
REBOIL FEED TEMP

Temperature and
Fuel Gas Flow SP

Debutanizer Closed-Loop Testing Multivariable System Identification

PRBS Signal and Input Series
0.1
• Motivation for multivariable identification
0.05
Fuel Gas Flowrate
0 • Multiple input extensions to:
Setpoint
-0.05
– PRBS, RBS design
-0.1
0 50 100 150 200 250 300
– ARX estimation
4 Output Series – PEM
2
• Brief overviews of Bayard’s, Zhu’s, and subspace methods
Bottoms 0
Temperature -2
• Overview of ASU’s MIMO control-relevant methodology
-4
0 50 100 150
time
200 250 300
– “zippered” multisine signals
Closed-loop data set generated by signal injection at the Fuel Gas Flowrate
Setpoint; dashed line shows external signal (ud); solid lines show u and y,
respectively Illustrations from various applications

© Copyright 1998-2004 by D.E. Rivera, All Rights Reserved

14
D.E. Rivera, Introduction to System Identification, ChE 494/598, January 20, 2004

MIMO PRBS Input Experimental Data (Noise free)

Control-Relevant Identification Methodology
MIMO PRBS Input Experiment Data
0.02
“Plant-Friendly” Nonparametric Control-Relevant

y1[T21]
0
Input Design Estimation Parameter
-0.02

0.02
0 100 200 300 400 500 600 700 800 Estimation
Schroeder-Phased
DFT
y2[T7]

0
SIMO Analysis Frequency-Weighted
-0.02
0x 10 -3 100 200 300 400 500 600 700 800
1
Curvefitting
u1[L]

0
PRBS and
-1
0x 10 -3 100 200 300 400 500 600 700 800
SIMO Controller Design
1
High-Order
MIMO
u2[V]

0
ARX Estimation
-1
0 100 200 300 400
Time[Min]
500 600 700 800 Random Binary
L H max
Sequence
PRBS: Specifying τdom = 5, τdom = 33, αs = 2, βs = 3, Tsettle = 165, SIMO: Single-Input, Multi-Output
and Tsampl = 2 min leads to nr = 7, Tsw = 6 min, and D = 168 min. SIMO MIMO:Multi-Input, Multi-Output
Signal magnitude set at usat = 0.001. MIMO

Model Predictive Control

Semiphysical Modeling
Past Future
Brine-Water Mixing Tank Example
Prediction Horizon

O O O
O
^
y(k) O O
O
O
O
O
O
O
O u(k)
O Move Horizon

k k+1 k+2 k+M k+P

Following prediction, the MPC controller solves the fol-
lowing multiobjective optimization problem,

Keep

controlled variables

at setpoint

p 
 2

min 
ΓY (y(k +  | k) − r(k + ))
[∆u(k),...,∆u(k+m)] =1
Move suppression Consider the dynamics of a tank mixing
  

+
m

Γu∆u(k +  − 1) 2 fresh and brine flow streams
=1

© Copyright 1998-2004 by D.E. Rivera, All Rights Reserved

15
D.E. Rivera, Introduction to System Identification, ChE 494/598, January 20, 2004

Mixing Tank Example, Continued System Identification Toolbox (SITB)

The first-principles model for this system is: Graphical User Interface (GUI)
dc
V = qc cc − (qc + qw ) c
dt
Using a forward-difference approximation on the derivative leads to

c(t + 1) − c(t) qc(t) cc(t) (qc(t) + qw (t)) c(t)

= −
T V V
which solving for c(t + 1) yields

qc(t) cc(t) T (qc(t) + qw (t)) c(t) T

c(t + 1) = c(t) + −
V V
Rearranging and consolidating terms leads to the semiphysical structure

c(t) = θ1c(t−1)+θ2qc(t−1) cc(t−1)+θ3qc(t−1) c(t−1)+θ4qw (t−1) c(t−1)

θ1, θ2, θ3, and θ4 can be estimated via linear regression.

© Copyright 1998-2004 by D.E. Rivera, All Rights Reserved

16