Beruflich Dokumente
Kultur Dokumente
System Identification
J. McLellan
Fall 2005
Module 1:
Introduction and Motivation
What do dynamic data look like?
0 var1
#1 # 301 # 601 # 901 # 1201 # 1501 # 1801 # 2101 var2
# 151 # 451 # 751 # 1051 # 1351 # 1651 # 1951
Overlay Plot
avg_usa_gas
4/15/92 8/15/97
12/15/86 12/15/02
date
• dynamic models
» role
» sources of variability and need for disturbance models
» components in data
• statistical properties of a model - sensor analogy
• making a link to linear regression
» problem formulation and assumptions
» parameter estimates + properties
» role of experimental design
• motivating example - 1st order process
» “lagged regression” problem
» impact of non-ideality in disturbances - serial correlation
• process characterization
» analysis of process dynamics
» analysis of disturbance dynamics
• process control
» loop configuration and tuning (e.g., PID control)
» model-based control algorithms
• cf. Internal Model Control (IMC) formalism
• minimum variance control (MVC)
• model predictive control (MPC) - e.g., Dynamic Matrix Control
• process monitoring
» controller performance assessment, process tracking
• models implemented as
» step response models - primarily for MPC
» impulse response models - MPC
» transfer functions - MVC, IMC, occasionally MPC
• range of disturbance estimation approaches
» “step” disturbances - compare predicted vs. current
measured value - e.g., in MPC
» time series approaches
• MPC reference - survey of Badgwell and Qin
• process disturbances
– “internal”
• flow fluctuations
• deterioration - e.g., channeling in a reactor
– “external”
• from upstream units
• ambient conditions - e.g., air temperature
• operator interventions
• no time variation
• does this mean the time trace is a straight line?
– no - we have noise !
– focus on whether variability patterns (distributions) are
changing
– is mean value constant?
– is variance constant?
• deterministic
– non-random relationships
– physical relationships
» e.g., energy/material balance relationships in column
influence composition in overhead
• stochastic
– random fluctuations - variability pattern
– frequently disturbances
– can evolve in time
time series
model
process disturbance
transfer
function
model
?
chee825 - Fall 2005 J. McLellan 16
Dynamic vs. Steady State Disturbance Cases
output
output
0 0
-2
1 -4 -5
−1 0 100 200 300 0 100 200 300
1 −dz AR parameter of 0.9 Non-stationary
• referred to as an 10 20
output
0 0
» “d” is the AR parameter
-5 -10
-10 -20
0 100 200 300 0 100 200 300
Note the vertical scales of the plots time time
• Deterministic component
» Step response
• dynamic models
» role
» sources of variability and need for disturbance models
» components in data
• statistical properties of a model - sensor analogy
• making a link to linear regression
» problem formulation and assumptions
» parameter estimates + properties
» role of experimental design
• motivating example - 1st order process
» “lagged regression” problem
» impact of non-ideality in disturbances - serial correlation
number of
samples
value
sensor avg true value
bias
chee825 - Fall 2005 J. McLellan 22
Sensor Reproducibility
poor consistency
value
= “+”
parameter 1 parameter 2
variability associated with parameter estimates is
influenced by data collection, model structure
form of summation dictated by model type
• dynamic models
» role
» sources of variability and need for disturbance models
» components in data
• statistical properties of a model - sensor analogy
• making a link to linear regression
» problem formulation and assumptions
» parameter estimates + properties
» role of experimental design
• motivating example - 1st order process
» “lagged regression” problem
» impact of non-ideality in disturbances - serial correlation
Y = β 0 + β1x + ε
response
regressor “noise”
“input” (disturbance)
Given -
• N sets of (x,y) data
• estimate parameters
o
responses
o o o
o
o
x
N N 2 2
min ∑ ( yi − y
$i) = ∑ ei
β0,β1i=1 i=1
chee825 - Fall 2005 J. McLellan 29
Linear Regression Assumptions
Model
Y = Xβ + Ε
Least squares estimates
$ T −1 T
β = (X X ) X Y
operating points
for msmts. observations
“design matrix”
y$
y$
model surface
chee825 - Fall 2005 J. McLellan 34
Properties of LS Parameter Estimates
Consistent
» behaviour as number of data points tends to infinity
» with probability 1,
lim β$ = β
N→ ∞
» distribution narrows as N becomes large
Efficient
» variance of least squares estimates is less than that of
other types of parameter estimates
Covariance Structure
» summarized by variance-covariance matrix
… in matrix form -
β1
marginal confidence limits
chee825 - Fall 2005 J. McLellan 39
Role of Experimental Design
• dynamic models
» role
» sources of variability and need for disturbance models
» components in data
• statistical properties of a model - sensor analogy
• making a link to linear regression
» problem formulation and assumptions
» parameter estimates + properties
» role of experimental design
• motivating example - 1st order process
» “lagged regression” problem
» impact of non-ideality in disturbances - serial correlation
TI
TIC
Input Sequence
Step
OR
PRBS FI
⎡ y5 ⎤ ⎡ u4 u3 u2 u1 ⎤⎡h1 ⎤ ⎡ e5 ⎤
⎢y ⎥ ⎢ u u4 u3 u2 ⎥⎢h2 ⎥ ⎢ e6 ⎥
6
⎢ ⎥= ⎢ 5 ⎥⎢ ⎥ + ⎢ ⎥
⎢ M⎥ ⎢ M ⎥⎢h3 ⎥ ⎢ M⎥
⎢y ⎥ ⎢u u N −4 ⎥⎢h ⎥ ⎢e ⎥
⎣ N ⎦ ⎣ N −1 u N −2 u N −3 ⎦⎣ 4 ⎦ ⎣ N ⎦
Y = UH +E
matrix of uncorrelated
T −1 T
deterministic H$ = (U U ) U Y (white)
elements noise
impulse weights as
chee825 - Fall 2005 J. McLellan regression parameters 44
Properties of Impulse Estimates - Case #1
» unbiased
» consistent
» best linear unbiased estimator
» essentially, possess standard properties of least squares
linear regression estimates
» if input is white noise, the analysis is more complex
• prove properties by examining dependence of estimates on
stochastic inputs
yt = a1yt −1 + b1ut −1 + et
» AutoRegressive with eXogenous input (ARX) model
» dependence of current temperature on temperature at
previous sample time
» dependence of current disturbance component on value
at previous time - disturbance is no longer simply white
noise
» inertia in disturbance exactly matches that in process
output
output
1 0 0
-2
−1
1 −dz -4
0 100 200 300
-5
0 100 200 300
• referred to as an AR parameter of 0.9 Non-stationary
“autoregressive” (AR) time 10 20
series model 5 10
» “d” is the AR parameter
output
output
0 0
-5 -10
-10 -20
0 100 200 300 0 100 200 300
Note the vertical scales of the plots time time
(1 −a1q−1 ) yt = b1q−1ut + et
OR
b1q −1 1
yt = ut + et
(1−a1q −1) (1−a1q −1)
{
chee825 - Fall 2005
process
{
J. McLellan
disturbance
dependence of process and
disturbance on previous
values is the same 48
Formulation as Linear Regression Problem
⎡ y2 ⎤ ⎡ y1 u1 ⎤ ⎡ e2 ⎤
⎢y ⎥ ⎢ y u ⎥a ⎢e ⎥
⎢ 3 ⎥= ⎢ 2 2 ⎥⎡ 1 ⎤ ⎢ 3 ⎥
⎢ ⎥ +
⎢ M⎥ ⎢ M M ⎥⎣b1 ⎦ ⎢ M⎥
⎢y ⎥ ⎢y u ⎥ ⎢e ⎥
⎣ N ⎦ ⎣ N −1 N −1 ⎦ ⎣ N⎦
» consistent
» asymptotically unbiased
» properties proved by examining dependence of the
estimates on the lagged outputs - “stochastic regressor
elements”
» key point is the manner in which disturbance appears
relative to process dynamics - same autoregressive
dynamics (denominator term)
nt
In difference equation form:
nt
In transfer function form:
⎡ y2 ⎤ ⎡ y1 u1 ⎤ ⎡ n2 ⎤
⎢y ⎥ ⎢ y u ⎥a ⎢n ⎥
⎢ 3 ⎥= ⎢ 2 2 ⎥⎡ 1 ⎤ ⎢ 3 ⎥
⎢ ⎥ +
⎢ M⎥ ⎢ M M ⎥⎣b1 ⎦ ⎢ M⎥
⎢y ⎥ ⎢y u ⎥ ⎢n ⎥
⎣ N ⎦ ⎣ N −1 N −1 ⎦ ⎣ N⎦
» biased
» not consistent
» likely to have poor precision - variance