Model
Predictive
Control
©
2009
by
A.
Bemporad Controllo
di
Processo
e
dei
Siste
l:
Basic
Concepts
emi
di
Produzione
‐
A.a.
2008/09 1 /94
Model
Predictive
Control
model‐based
optimizer
reference in
r(t) u
measu
A
model
of
the
process
is
used
to
of
the
process
to
optimize
the
co
©
2009
by
A.
Bemporad Controllo
di
Processo
e
dei
Siste
l
(MPC)
process
nput output
u(t) y(t)
urements
o
predict
the
future
evolution
ontrol
signal
emi
di
Produzione
‐
A.a.
2008/09 2 /94
Receding
horizon
philoso
•
At
time t:
solve
an
optimal
control
problem
over
a
finite
future
horizon
of N steps:
•
Only
apply
the
first
optimal
move
•
At
time
t+1:
Get
new
measurements,
rep
Advantage
of
repeated
on
©
2009
by
A.
Bemporad Controllo
di
Processo
e
dei
Siste
ophy
r(t) yt+k
Predicted
outputs
Manipulated ut+k
Inputs
t+N
t t+1
t+1 t+2 t+N+1
peat
the
optimization.
And
so
on
…
n‐line
optimization:
FEEDBACK!
emi
di
Produzione
‐
A.a.
2008/09 3 /94
Receding
Horizon
‐
Exam
•
MPC
is
like
playing
chess
!
•
“Rolling
horizon”
policies
are
also
used
frequently
in
finance
©
2009
by
A.
Bemporad Controllo
di
Processo
e
dei
Siste
mples
emi
di
Produzione
‐
A.a.
2008/09 4 /94
Receding
Horizon
‐
Exam
‐
prediction
model how
vehicle
moves
‐
constraints drive
on
roads,
resp
‐
disturbances mainly
driver’s
inat
‐
set
point desired
location
‐
cost
function minimum
time,
minimum
distance,
‐
receding
horizon
mechanism
event‐based
(optimal
route
re‐planned
when
path
©
2009
by
A.
Bemporad Controllo
di
Processo
e
dei
Siste
mples
s
on
the
map
pect
one‐way
roads,
etc.
ttention
!
x
=
GPS
position
u
=
navigation
commands
,
etc. Fastest route
is
lost) Shortest route
Avoid motorways
Walking route
Bicycle route
Limited speed
emi
di
Produzione
‐
A.a.
2008/09 5 /94
Good
Models
for
(MPC)
C
Note:
computational
complexity
an
stability)
depend
on
chosen
model/
Good
models
for
MPC:
•
Descriptive
enough
to
capture
th
dynamics
of
the
system
TRAD
•
Simple
enough
for
solving
the
op
“Make
everything
as
simple
as
possi
—
Albert
Einstein
©
2009
by
A.
Bemporad Controllo
di
Processo
e
dei
Siste
Control
nd
theoretical
properties
(e.g.
/objective/constraints
he
most
significant
6 /94
DE
OFF
ptimization
problem
ible,
but
not
simpler.”
emi
di
Produzione
‐
A.a.
2008/09
MPC
in
Industry
•
History:
Computer
control
(“Man
Fluid
catalytic
cracking
(courtesy
of
Shell
/
M.
Morari)
©
2009
by
A.
Bemporad Controllo
di
Processo
e
dei
Siste
nual”
MPC)
Fluid
catalytic
cracking
(FCC)
is
the
most
important
conversion
process
used
in
petroleum
refineries.
It
is
widely
used
to
convert
the
high‐boiling
hydrocarbon
fractions
of
petroleum
crude
oils
to
more
valuable
gasoline,
olefinic
gases
and
other
products
(http://en.wikipedia.org/wiki/Catalytic_cracking)
emi
di
Produzione
‐
A.a.
2008/09 7 /94
MPC
in
Industry
•
History:
1979
Dynamic
Matrix
Co
(Motivation:
multivariable,
const
•
Present
Industrial
Practice
•
linear
impulse/step
response
models
•
sum
of
squared
errors
objective
function
•
executed
in
supervisory
mode
•
Particularly
suited
for
problems
w
•
many
inputs
and
outputs
•
constraints
on
inputs,
outputs,
states
•
varying
objectives
and
limits
on
variab
(e.g.
because
of
faults)
©
2009
by
A.
Bemporad Controllo
di
Processo
e
dei
Siste
ontrol
(DMC)
by
Shell
trained)
with 8 /94
bles
emi
di
Produzione
‐
A.a.
2008/09
MPC
in
Industry
Hierarchy
of
control
system
functio
Conventional
©
2009
by
A.
Bemporad Controllo
di
Processo
e
dei
Siste
ons:
MPC
emi
di
Produzione
‐
A.a.
2008/09 (Qin,
Badgewell,
1997)
9 /94
MPC
in
Industry
(snapshot
survey
conducted
in
mid‐1999)
“For
us
multivariable
con
Tariq
Samad,
Honeywell
(IEEE
Control
Syste
©
2009
by
A.
Bemporad Controllo
di
Processo
e
dei
Siste
(Qin,
Badgewell,
2003)
ntrol
is
predictive
control
” 10 /94
em
Society,
President)
(1997)
emi
di
Produzione
‐
A.a.
2008/09
MPC
in
Industry
Results
from
a
recent
survey
(November
7,
2005)
real‐time
optimization
in
a
set
of
US
industries:
Industrial
area
of
respondents
to
the
survey:
Pharmaceuticals 0.7
%
Machinery 0.7
%
0.7
%
Food
&
Beverage 0.7
%
Aerospace
Automotive 1.5
%
Mining
1.5
%
Cement
&
Glass 2.2
%
Metals 2.2
%
Plastics
&
Rubber 2.9
%
Electronics 5.1
%
Power 5.1
%
Pulp
&
Paper 5.1
%
Other
11.0
%
Oil
&
Gas
Chemicals 5 10 15 20 15.4
% 20.6
%
25 24.3
Refining
30 35
0
courtesy: Controllo
di
Processo
e
dei
Siste
©
2009
by
A.
Bemporad
)
about
the
use
of
MPC
techniques
/
Do
you
see
your
use
of
MPC
accelerating,
staying
constant,
or
declining?
Decreasing
Constant 28.8
%
Increasing 69.8
%
0 20 40 60 80 100
Does
your
company
use
MPC
?
No 19.0
%
%
Routinely 35.2
%
Sometimes 45.8
%
0 10 20 30 40 50 60 70 11/94
emi
di
Produzione
‐
A.a.
2008/09
MMoodedl
ePrl
ePdircetidvei
cCtoinvtero
lC
Tooonlbtorxol
•
MPC
Toolbox
3.0
(Bemporad,
Ricker,
Mo
–
Object‐oriented
implementation
(MPC
–
MPC
Simulink
Library
–
MPC
Graphical
User
Interface
–
RTW
extension
(code
generation)
[xPC
Target,
dSpace,
etc.]
–
Linked
to
OPC
Toolbox
v2.0.1
Only
linear
models
are
handled
http://www.mathworks.com/products/
©
2009
by
A.
Bemporad Controllo
di
Processo
e
dei
Siste
l
Toolbox
orari,
1998‐today):
C
object)
/mpc/ 12 /94
emi
di
Produzione
‐
A.a.
2008/09
MPC
Simulink
Library
•
Single
MPC
and
multiple
swicthed
MPC
b
•
Reference/disturbance
preview
and
time
©
2009
by
A.
Bemporad Controllo
di
Processo
e
dei
Siste
blocks
supported 13 /94
e‐varying
limits
supported
emi
di
Produzione
‐
A.a.
2008/09
MPC
Graphical
User
Inter
©
2009
by
A.
Bemporad Controllo
di
Processo
e
dei
Siste
rface
emi
di
Produzione
‐
A.a.
2008/09 14 /94
MPC
Tuning
Advisor
©
2009
by
A.
Bemporad Controllo
di
Processo
e
dei
Siste
emi
di
Produzione
‐
A.a.
2008/09 15 /94
Hybrid
Toolbox
for
MATL
Features:
•
Hybrid
models:
design,
simulation,
verifi
•
Control
design
for
linear
systems
w/
con
and
hybrid
systems
(on‐line
optimization
•
Explicit
MPC
control
(via
multi‐parametr
•
C‐code
generation
•
Simulink
library
http://www.dii.unisi
©
2009
by
A.
Bemporad Controllo
di
Processo
e
dei
Siste
LAB (Bemporad,
2003‐2009)
fication
nstraints
n
via
QP/MILP/MIQP)
ric
programming)
2450+
download
requests
since
October
2004
i.it/hybrid/toolbox 16 /94
emi
di
Produzione
‐
A.a.
2008/09
Basics
of
Constrained
Opt
©
2009
by
A.
Bemporad Controllo
di
Processo
e
dei
Siste
timization
emi
di
Produzione
‐
A.a.
2008/09 17/94
Mathematical
Programm
In
general,
problem
is
difficult
t
use
software
tools
©
2009
by
A.
Bemporad Controllo
di
Processo
e
dei
Siste
ming
to
solve 18 /94
emi
di
Produzione
‐
A.a.
2008/09
Optimization
Software
• Taxonomy
of
most
known
solver
optimization
problems:
http://www-fp.mcs.anl.gov/o
• Network
Enabled
Optimization
S
solving
optimization
problems:
http://neos.mcs.anl.gov/neo
• Comparison
on
benchmark
prob
http://plato.la.asu.edu/ben
• Good
open‐source
optimization
http://www.coin-or.org/
©
2009
by
A.
Bemporad Controllo
di
Processo
e
dei
Siste
rs,
for
different
classes
of
otc/Guide/SoftwareGuide/
Server
(NEOS)
for
remotely
os/solvers/
blems:
nch.html
software
emi
di
Produzione
‐
A.a.
2008/09 19 /94
Convex
sets
Convex
set
©
2009
by
A.
Bemporad Controllo
di
Processo
e
dei
Siste
Nonconvex
set
emi
di
Produzione
‐
A.a.
2008/09 20 /94
Convex
function
©
2009
by
A.
Bemporad Controllo
di
Processo
e
dei
Siste
emi
di
Produzione
‐
A.a.
2008/09 21 /94
Convex
Optimization
Pro
• Very
efficient
numerical
algorithm
• Global
solution
attained
• Extensive
useful
theory
• Often
occurring
in
engineering
pr
• Tractable
in
theory
and
practice
Excellent
reference
textbook:
“Convex
O
and
L.
Vandenberghe http://www.s
©
2009
by
A.
Bemporad Controllo
di
Processo
e
dei
Siste
oblem
f
and
C
convex
ms
exist
roblems
Optimization”
by
S.
Boyd
22 /94
stanford.edu/~boyd/cvxbook/
emi
di
Produzione
‐
A.a.
2008/09
Polyhedra
• A
convex
polyhedron
is
the
inter
set
of
halfspaces
of
Rd
• A
convex
polytope
is
a
bounded
A2
A3x=b3
•
Hyperplane
representation:
©
2009
by
A.
Bemporad Controllo
di
Processo
e
dei
Siste
rsection
of
a
finite
d
convex
polyhedron
A 2x=b 2
A
1
A x=b1
1
A3
emi
di
Produzione
‐
A.a.
2008/09 23 /94
Linear
Program
‐f
f’x
Slack
variables
©
2009
by
A.
Bemporad Controllo
di
Processo
e
dei
Siste
George
Dantzig
(1914
‐
2005)
Standard
form:
emi
di
Produzione
‐
A.a.
2008/09 24 /94
Linear
Program
trasformation
from
max
to
min:
Change
inequality
direction:
It
is
always
possible
to
using
“min”
and
©
2009
by
A.
Bemporad Controllo
di
Processo
e
dei
Siste
o
formulate
LP
problems 25 /94
d
“·”
inequalities
emi
di
Produzione
‐
A.a.
2008/09