Professional Documents
Culture Documents
Ac2 10 MPC PDF
Ac2 10 MPC PDF
Automatic Control 2
University of Trento
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 1 /37
Lecture: Model predictive control Introduction
model-‐based
optimizer process
measurements
Use
a
dynamical
model
of
the
process
to
predict
its
future
evolution
and
optimize
the
control
signal
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 2 /37
Lecture: Model predictive control Introduction
•
At
time
t+1:
Get
new
measurements,
repeat
the
optimization.
And
so
on
…
Advantage
of
repeated
on-‐line
optimization:
FEEDBACK
!
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 3 /37
Lecture: Model predictive control Introduction
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 4 /37
Lecture: Model predictive control Introduction
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 5 /37
Lecture: Model predictive control Introduction
MPC in Industry
•
History:
1979
Dynamic
Matrix
Control
(DMC)
by
Shell
(Motivation:
multivariable,
constrained)
MPC in Industry
Hierarchy
of
control
system
functions:
Conventional MPC
MPC in Industry
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 /37
Lecture: Model predictive control Linear MPC
batch
least
and
hence squares
Alternative approach: use dynamic programming to find U* (Riccati iterations)
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 /37
Lecture: Model predictive control Linear MPC
• Constraints:
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 11 /37
Lecture: Model predictive control Linear MPC
(quadratic)
(linear)
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 12 /37
Lecture: Model predictive control Linear MPC
• Solution:
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 14 /37
Lecture: Model predictive control Toolboxes
Complete solution for linear MPC design based on on-‐line QP
http://www.mathworks.com/products/mpc/
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 15 /37
Lecture: Model predictive control Toolboxes
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 16 /37
Lecture: Model predictive control Toolboxes
measured
disturbances
from
simulation
diagram
(Bemporad, 2003-today)
Hybrid Toolbox for MATLAB
Features:
•
Hybrid
models:
design,
simulation,
verification
•
Control
design
for
linear
systems
w/
constraints
and
hybrid
systems
(on-‐line
optimization
via
QP/MILP/MIQP)
•
Explicit
MPC
control
(via
multi-‐parametric
programming)
•
C-‐code
generation
•
Simulink
library
http://www.dii.unisi.it/hybrid/toolbox
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 18 /37
Lecture: Model predictive control Linear MPC example
•
System:
sampling
+
ZOH
Ts=1
s
• Constraints:
cost:
constraints:
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 /37
Lecture: Model predictive control Linear MPC example
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 20 /37
Lecture: Model predictive control Linear MPC example
cost:
constraints:
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 21 /37
Lecture: Model predictive control Linear MPC example
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 22 /37
Lecture: Model predictive control Tracking
Again
a
linear
system
with
states
x(t),
xu(t)
and
input
Δu(t)
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 23 /37
Lecture: Model predictive control Tracking
optimization
vector:
• Note:
same formulation as before (W=Cholesky factor of weight matrix Q)
•
Optimization
problem:
Convex
Quadratic
Program
(QP)
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 24 /37
Lecture: Model predictive control MPC features
HOWEVER
MPC Features
•
Multivariable
constrained
“non-‐square”
systems
(i.e.
#inputs
and
#outputs
are
different)
• Delay compensation
• New possibility for MPC: explicit piecewise affine solutions (Bemporad et al., 2002)
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 26 /37
Lecture: Model predictive control MPC theory
MPC Theory
•
Areas:
•
Linear
MPC:
linear
model
•
Nonlinear
MPC:
nonlinear
model
•
Robust
MPC:
uncertain
(linear)
model
•
Hybrid
MPC:
model
integrating
logic,
dynamics,
and
constraints
•
Issues:
–
Feasibility
–
Stability
(Convergence)
–
Computations
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 27 /37
Lecture: Model predictive control MPC theory
Convergence Result
Convergence Proof
Lyapunov function
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 29 /37
Lecture: Model predictive control MPC theory
Jacopo
Francesco
Riccati
(1676
-‐
1754)
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 30 /37
Lecture: Model predictive control MPC example
Example: AFTI-16
•
Linearized
model:
Yaw Pitch
Aileron
Rudder
Flaps
Elevators
Roll
Aileron
Longitudinal
axis
Vertical
axis
Lateral
axis
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 31 /37
Lecture: Model predictive control MPC example
Example: AFTI-16
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 32 /37
Lecture: Model predictive control MPC example
Example: AFTI-16
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 33 /37
Lecture: Model predictive control MPC tuning
Tuning Guidelines
• Weights: the larger the ratio Wy/WΔu the more aggressive the controller
• Input
horizon: the larger Nu, the more “optimal” but the more complex the controller
• Output
horizon: the smaller N, the more aggressive the controller
Conclusions on MPC
Main pros of MPC:
Can handle nonlinear/switching/MIMO dynamics with delays
Can enforce constraints on inputs and outputs
Performance is optimized
Systematic design approach, MPC designs are easy to maintain
MATLAB tools exist to assist the design and for code generation
Started in the 80‘s in the process industries, now reaching automotive, avionics
aerospace, power systems, ...
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 35 /37
Lecture: Model predictive control Conclusions of the course
Italian-English Vocabulary
Prof. A. Bemporad (University of Trento) Automatic Control 2 Academic Year 2010/2011 37 /37