Professional Documents
Culture Documents
(Download PDF) Model Predictive Control Theory Computation and Design 2Nd Edition Rawlings James B Full Chapter PDF
(Download PDF) Model Predictive Control Theory Computation and Design 2Nd Edition Rawlings James B Full Chapter PDF
https://ebookmass.com/product/process-control-modeling-design-
and-simulation-2nd-edition-b-wayne-bequette/
https://ebookmass.com/product/predictive-control-daniel-lequesne/
https://ebookmass.com/product/model-predictive-control-for-
doubly-fed-induction-generators-and-three-phase-power-converters-
alfeu-j-sguarezi-filho/
https://ebookmass.com/product/surface-area-and-porosity-
determinations-by-physisorption-measurement-classical-theories-
and-quantum-theory-2nd-edition-james-b-condon/
Philosophy and Model Theory Tim Button
https://ebookmass.com/product/philosophy-and-model-theory-tim-
button/
https://ebookmass.com/product/ebook-pdf-data-mining-and-
predictive-analytics-2nd-edition/
https://ebookmass.com/product/engineering-computation-an-
introduction-using-matlab-and-exce-2nd-edition-musto/
https://ebookmass.com/product/tidal-energy-systems-design-
optimization-and-control-1st-edition/
https://ebookmass.com/product/organization-theory-and-
design-13th-edition-daft/
Model Predictive Control:
Theory, Computation, and Design
2nd Edition
ISBN 9780975937730
9 780975 937730
Model Predictive Control:
Theory, Computation, and Design
2nd Edition
James B. Rawlings
Department of Chemical Engineering
University of California
Santa Barbara, California, USA
David Q. Mayne
Department of Electrical and Electronic Engineering
Imperial College London
London, England
Moritz M. Diehl
Department of Microsystems Engineering and
D
Department of Mathematics
University of Freiburg
Freiburg, Germany
b Hi
No ll Publishing
First Edition
First Printing August 2009
Electronic Download November 2013
Electronic Download (2nd) April 2014
Electronic Download (3rd) July 2014
Electronic Download (4th) October 2014
Electronic Download (5th) February 2015
Second Edition
First Printing October 2017
Electronic Download October 2018
Electronic Download (2nd) February 2019
To Cheryl, Josephine, and Stephanie,
In the eight years since the publication of the first edition, the field
of model predictive control (MPC) has seen tremendous progress. First
and foremost, the algorithms and high-level software available for solv-
ing challenging nonlinear optimal control problems have advanced sig-
nificantly. For this reason, we have added a new chapter, Chapter 8,
“Numerical Optimal Control,” and coauthor, Professor Moritz M. Diehl.
This chapter gives an introduction into methods for the numerical so-
lution of the MPC optimization problem. Numerical optimal control
builds on two fields: simulation of differential equations, and numeri-
cal optimization. Simulation is often covered in undergraduate courses
and is therefore only briefly reviewed. Optimization is treated in much
more detail, covering topics such as derivative computations, Hessian
approximations, and handling inequalities. Most importantly, the chap-
ter presents some of the many ways that the specific structure of opti-
mal control problems arising in MPC can be exploited algorithmically.
We have also added a software release with the second edition of
the text. The software enables the solution of all of the examples and
exercises in the text requiring numerical calculation. The software is
based on the freely available CasADi language, and a high-level set of
Octave/MATLAB functions, MPCTools, to serve as an interface to CasADi.
These tools have been tested in several MPC short courses to audiences
composed of researchers and practitioners. The software can be down-
loaded from www.chemengr.ucsb.edu/~jbraw/mpc.
In Chapter 2, we have added sections covering the following topics:
• economic MPC
• MPC with discrete actuators
We also present a more recent form of suboptimal MPC that is prov-
ably robust as well as computationally tractable for online solution of
nonconvex MPC problems.
In Chapter 3, we have added a discussion of stochastic MPC, which
has received considerable recent research attention.
In Chapter 4, we have added a new treatment of state estimation
with persistent, bounded process and measurement disturbances. We
have also removed the discussion of particle filtering. There are two
vi
vii
viii
ix
JBR DQM
Madison, Wisconsin, USA London, England
Acknowledgments
Both authors would like to thank the Department of Chemical and Bio-
logical Engineering of the University of Wisconsin for hosting DQM’s
visits to Madison during the preparation of this monograph. Funding
from the Paul A. Elfers Professorship provided generous financial sup-
port.
JBR would like to acknowledge the graduate students with whom
he has had the privilege to work on model predictive control topics:
Rishi Amrit, Dennis Bonné, John Campbell, John Eaton, Peter Findeisen,
Rolf Findeisen, Eric Haseltine, John Jørgensen, Nabil Laachi, Scott Mead-
ows, Scott Middlebrooks, Steve Miller, Ken Muske, Brian Odelson, Mu-
rali Rajamani, Chris Rao, Brett Stewart, Kaushik Subramanian, Aswin
Venkat, and Jenny Wang. He would also like to thank many colleagues
with whom he has collaborated on this subject: Frank Allgöwer, Tom
Badgwell, Bhavik Bakshi, Don Bartusiak, Larry Biegler, Moritz Diehl,
Jim Downs, Tom Edgar, Brian Froisy, Ravi Gudi, Sten Bay Jørgensen,
Jay Lee, Fernando Lima, Wolfgang Marquardt, Gabriele Pannocchia, Joe
Qin, Harmon Ray, Pierre Scokaert, Sigurd Skogestad, Tyler Soderstrom,
Steve Wright, and Robert Young.
DQM would like to thank his colleagues at Imperial College, espe-
cially Richard Vinter and Martin Clark, for providing a stimulating and
congenial research environment. He is very grateful to Lucien Polak
and Graham Goodwin with whom he has collaborated extensively and
fruitfully over many years; he would also like to thank many other col-
leagues, especially Karl Åström, Roger Brockett, Larry Ho, Petar Koko-
tovic, and Art Krener, from whom he has learned much. He is grateful
to past students who have worked with him on model predictive con-
trol: Ioannis Chrysochoos, Wilbur Langson, Hannah Michalska, Sasa
Raković, and Warren Schroeder; Hannah Michalska and Sasa Raković,
in particular, contributed very substantially. He owes much to these
past students, now colleagues, as well as to Frank Allgöwer, Rolf Find-
eisen Eric Kerrigan, Konstantinos Kouramus, Chris Rao, Pierre Scokaert,
and Maria Seron for their collaborative research in MPC.
Both authors would especially like to thank Tom Badgwell, Bob Bird,
Eric Kerrigan, Ken Muske, Gabriele Pannocchia, and Maria Seron for
their careful and helpful reading of parts of the manuscript. John Eaton
x
xi
again deserves special mention for his invaluable technical support dur-
ing the entire preparation of the manuscript.
Added for the second edition. JBR would like to acknowledge the
most recent generation of graduate students with whom he has had the
privilege to work on model predictive control research topics: Doug Al-
lan, Travis Arnold, Cuyler Bates, Luo Ji, Nishith Patel, Michael Risbeck,
and Megan Zagrobelny.
In preparing the second edition, and, in particular, the software re-
lease, the current group of graduate students far exceeded expectations
to help finish the project. Quite simply, the project could not have been
completed in a timely fashion without their generosity, enthusiasm,
professionalism, and selfless contribution. Michael Risbeck deserves
special mention for creating the MPCTools interface to CasADi, and
updating and revising the tools used to create the website to distribute
the text- and software-supporting materials. He also wrote code to cal-
culate explicit MPC control laws in Chapter 7. Nishith Patel made a
major contribution to the subject index, and Doug Allan contributed
generously to the presentation of moving horizon estimation in Chap-
ter 4.
A research leave for JBR in Fall 2016, again funded by the Paul A.
Elfers Professorship, was instrumental in freeing up time to complete
the revision of the text and further develop computational exercises.
MMD wants to especially thank Jesus Lago Garcia, Jochem De Schut-
ter, Andrea Zanelli, Dimitris Kouzoupis, Joris Gillis, Joel Andersson,
and Robin Verschueren for help with the preparation of exercises and
examples in Chapter 8; and also wants to acknowledge the following
current and former team members that contributed to research and
teaching on optimal and model predictive control at the Universities of
Leuven and Freiburg: Adrian Bürger, Hans Joachim Ferreau, Jörg Fis-
cher, Janick Frasch, Gianluca Frison, Niels Haverbeke, Greg Horn, Boris
Houska, Jonas Koenemann, Attila Kozma, Vyacheslav Kungurtsev, Gio-
vanni Licitra, Rien Quirynen, Carlo Savorgnan, Quoc Tran-Dinh, Milan
Vukov, and Mario Zanon. MMD also wants to thank Frank Allgöwer, Al-
berto Bemporad, Rolf Findeisen, Larry Biegler, Hans Georg Bock, Stephen
Boyd, Sébastien Gros, Lars Grüne, Colin Jones, John Bagterp Jørgensen,
Christian Kirches, Daniel Leineweber, Katja Mombaur, Yurii Nesterov,
Toshiyuki Ohtsuka, Goele Pipeleers, Andreas Potschka, Sebastian Sager,
Johannes P. Schlöder, Volker Schulz, Marc Steinbach, Jan Swevers, Phil-
ippe Toint, Andrea Walther, Stephen Wright, Joos Vandewalle, and Ste-
fan Vandewalle for inspiring discussions on numerical optimal control
xii
xiii
xiv Contents
C Optimization 729
C.1 Dynamic Programming . . . . . . . . . . . . . . . . . . . . . 729
C.1.1 Optimal Control Problem . . . . . . . . . . . . . . . 731
C.1.2 Dynamic Programming . . . . . . . . . . . . . . . . . 733
C.2 Optimality Conditions . . . . . . . . . . . . . . . . . . . . . 737
C.2.1 Tangent and Normal Cones . . . . . . . . . . . . . . 737
C.2.2 Convex Optimization Problems . . . . . . . . . . . . 741
C.2.3 Convex Problems: Polyhedral Constraint Set . . . 743
C.2.4 Nonconvex Problems . . . . . . . . . . . . . . . . . . 745
C.2.5 Tangent and Normal Cones . . . . . . . . . . . . . . 746
C.2.6 Constraint Set Defined by Inequalities . . . . . . . 750
C.2.7 Constraint Set; Equalities and Inequalities . . . . . 753
C.3 Set-Valued Functions and Continuity of Value Function . 755
C.3.1 Outer and Inner Semicontinuity . . . . . . . . . . . 757
C.3.2 Continuity of the Value Function . . . . . . . . . . . 759
C.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 767
List of Figures
xxi
xxii List of Figures
5.1 State estimator tube. The solid line x̂(t) is the center of
the tube, and the dashed line is a sample trajectory of x(t). 342
5.2 The system with disturbance. The state estimate lies in
the inner tube, and the state lies in the outer tube. . . . . . 343
xxiv List of Figures
p p p+1 p+1
6.1 Convex step from (u1 , u2 ) to (u1 , u2 ). . . . . . . . . . 386
6.2 Ten iterations of noncooperative steady-state calculation. . 403
6.3 Ten iterations of cooperative steady-state calculation. . . . 403
6.4 Ten iterations of noncooperative steady-state calculation;
reversed pairing. . . . . . . . . . . . . . . . . . . . . . . . . . . 404
6.5 Ten iterations of cooperative steady-state calculation; re-
versed pairing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404
6.6 Cooperative control stuck on the boundary of U under
coupled constraints . . . . . . . . . . . . . . . . . . . . . . . . 412
6.7 Cost contours for a two-player, nonconvex game. . . . . . . 422
6.8 Nonconvex function optimized with the distributed gra-
dient algorithm. . . . . . . . . . . . . . . . . . . . . . . . . . . . 425
6.9 Closed-loop state and control evolution with (x1 (0), x2 (0)) =
(3, −3). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431
6.10 Contours of V (x(0), u1 , u2 ) for N = 1. . . . . . . . . . . . . . 432
6.11 Optimizing a quadratic function in one set of variables at
a time. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 440
6.12 Constrained optimality conditions and the normal cone. . 445
8.8 Direct single shooting solution for (8.63) without path con-
straints. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 589
8.9 Open-loop simulation for (8.63) using collocation. . . . . . . 592
8.10 Gauss-Newton iterations for the direct multiple-shooting
method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 594
xxvii
xxviii List of Examples and Statements
Mathematical notation
∃ there exists
∈ is an element of
∀ for all
=⇒ ⇐= implies; is implied by
=6 ⇒ ⇐=
6 does not imply; is not implied by
a := b a is defined to be equal to b.
a =: b b is defined to be equal to a.
≈ approximately equal
V (·) function V
V :A→B V is a function mapping set A into set B
x , V (x) function V maps variable x to value V (x)
x+ value of x at next sample time (discrete time system)
ẋ time derivative of x (continuous time system)
fx partial derivative of f (x) with respect to x
∇ nabla or del operator
δ unit impulse or delta function
|x| absolute value of scalar; norm of vector (two-norm unless
stated otherwise); induced norm of matrix
x sequence of vector-valued variable x, (x(0), x(1), . . .)
kxk sup norm over a sequence, supi≥0 |x(i)|
kxka:b max a≤i≤b |x(i)|
tr(A) trace of matrix A
det(A) determinant of matrix A
eig(A) set of eigenvalues of matrix A
ρ(A) spectral radius of matrix A, max i |λi | for λi ∈ eig(A)
A−1 inverse of matrix A
A †
pseudo-inverse of matrix A
A0 transpose of matrix A
inf infimum or greatest lower bound
min minimum
sup supremum or least upper bound
max maximum
xxxix
xl Notation
Symbols
A, B, C system matrices, discrete time, x + = Ax + Bu, y = Cx
Ac , Bc system matrices, continuous time, ẋ = Ac x + Bc u
Aij state transition matrix for player i to player j
Ai state transition matrix for player i
ALi estimate error transition matrix Ai − Li Ci
Bd input disturbance matrix
Bij input matrix of player i for player j’s inputs
Bi input matrix of player i
Cij output matrix of player i for player j’s interaction states
Ci output matrix of player i
Cd output disturbance matrix
C controllability matrix
C∗ polar cone of cone C
d integrating disturbance
E, F constraint matrices, F x + Eu ≤ e
f,h system functions, discrete time, x + = f (x, u), y = h(x)
fc (x, u) system function, continuous time, ẋ = fc (x, u)
F (x, u) difference inclusion, x + ∈ F (x, u), F is set valued
G input noise-shaping matrix
Gij steady-state gain of player i to player j
H controlled variable matrix
I(x, u) index set of constraints active at (x, u)
0
I (x) index set of constraints active at (x, u0 (x))
k sample time
K optimal controller gain
`(x, u) stage cost
`N (x, u) final stage cost
L optimal estimator gain
m input dimension
M cross-term penalty matrix x 0 Mu
M number of players, Chapter 6
M class of admissible input policies, µ ∈ M
n state dimension
N horizon length
O observability matrix, Chapters 1 and 4
O compact robust control invariant set containing the origin,
Chapter 3
p output dimension
xlii Notation
Greek letters
xlv
Another random document with
no related content on Scribd:
* Agrostemma githago.—The
common corn cockle (cockle;
mullein pink) is a weed common to
both the United States and Europe.
Poultry and household animals are
occasionally poisoned by eating the
seeds or the bread made from wheat
contaminated with the seeds.
MAGNOLIACEÆ (MAGNOLIA
FAMILY).
Illicium floridanum.—The
leaves of this species of anisetree are
supposed to be poisonous to stock.
RANUNCULACEÆ (CROWFOOT
FAMILY).
* Aconitum napellus.—Aconite
Fig. 77.—Slender nettle (monkshood; wolfsbane) is very
(Urtica gracilis). commonly cultivated in gardens, and
is therefore capable of doing great
damage to stock. Horses and cattle have frequently been poisoned by
eating the leaves and flowering tops.
* Aconitum columbianum.—The Western aconite, or
monkshood, is native in the north-western portion of America, where
it sometimes poisons sheep.
Anemone quinquefolia.—The common wind flower, which
grows throughout most of the United States, is extremely acrid and
poisonous. Cattle seldom touch it. The plant loses most of its poison
in drying.
Fig. 78.—Pokeweed (Phytolacca decandra), one-half natural size.
LARKSP
UR
POISONI
NG IN
SHEEP.[3]
3. The
following
account is
summarise
d from a
bulletin of
the
Montana
Experiment
Station by
Dr. Wilcox.
Severe
losses
have
from time
Fig. 79.—Corn cockle to time Fig. 80.—Aconite
(Agrostemma githago). a, been (Aconitum
Sprays showing flowers and recorded, columbianum). a,
seed capsule, one-third especially Flowering plant; b, seed
natural size; b, seed, natural in capsule—both one-third
size; b′, seed, four times America, natural size.
natural size. from
larkspur
poisoning, the number of animals lost
amounting to thousands. The first signs of poisoning are slight
general stiffness and straddling gait, especially of the hind legs. The
stiffness becomes more and more pronounced, until walking is
difficult and evidently painful. Soon there are manifested various
involuntary twitchings of the muscles of the legs and sides of the
body, and loss of control or co-ordination of the muscles. Ordinarily
there is no increase in the quantity of the saliva, no dribbling of
saliva from the mouth, no champing of the jaws or attempts at
swallowing. The sheep manifest none of the mental disturbances
frequently seen in cases of poisoning from other sources, as for
example loco weed and lupine. There is no impairment of the special
senses. The sheep seem to hear and see as well and as correctly as
under normal conditions of health.
No indications of any disturbances of the digestive functions are to
be seen. The appetite remains good, and the sheep eat up to the very
last. They were observed eating industriously during the intervals
between the attacks of spasms which they have during the last stages.
At first the frequency of the pulse and of the respiratory
movements is lessened and the temperature is lowered. The pulse
remains very weak, but in the later stages becomes very rapid, in
some cases 130 per minute. Toward the last also the respiration is
very shallow and rapid. During the final convulsions the respiration
is sometimes 120 per minute, but so shallow that the air is simply
pumped up and down the windpipe. The air in the lungs is therefore
not renewed, and the animal dies by asphyxia or suffocation.
So long as the sheep can stand on its feet, or walk, it keeps up with
the flock as nearly as possible. The exercise, however, excites it,
makes its respiration more rapid, and it has frequently to lie down
for a moment and then get up and hobble along after the flock. The
worst cases can thus easily be detected, since they straggle behind
the rest of the flock.
The later stages follow rather rapidly. The involuntary movements
become more frequent and more severe. All four legs tremble and
shake violently. In fact, all the muscles of the body contract
spasmodically until the animal totters over on its side and dies in the
most violent spasms.
Larkspur has the effect of arresting the heart’s action and
respiration and of paralysing the spinal cord.
Treatment. Place the animal by itself in a cool, quiet, shaded
place and avoid all excitement. Of the drugs tested, atropine sulphate
dissolved in camphor water has given the best results. Wilcox (Bull.
15, Montana Ex. Station) recommends for sheep from ¹⁄₂₀ to ¹⁄₁₅
grain in the earlier, and ⅙ to ¼ grain in the later convulsive stages.
Cattle require from four to five times these doses. Inhalations of
ammonia vapour, and small doses of alcohol and ether, are also
useful.
In other cases very good results have been obtained from giving
permanganate of potash in the form of a drench: 5 to 10 grains for an
adult sheep or pig, 15 to 20 grains for a horse, and 30 to 50 grains for
an ox, dissolved in a pint or two pints of water.
POISONING BY HELLEBORE.
* Ranunculus sceleratus.—
The cursed crowfoot, or celery-
leafed crowfoot, is found
throughout the eastern half of the
United States and also in Europe.
Cattle generally avoid all of the
buttercups, but fatal cases of
poisoning from this plant are
recorded in European literature.
When dried in hay, the plant
appears to be non-poisonous. The
bulbous crowfoot (R. bulbosus) and
the tall crowfoot (R. acris) are well-
Fig. 86.—Mandrake known to be very acrid in taste, and
(Podophyllum peltatum). it is probable that all of the species
which grow in water or in very
marshy land are poisonous.
POISONING BY RANUNCULACEÆ.
Poisoning only occurs when the green plants are eaten. Drying
causes certain essences contained in them to disappear, and thus
destroys their toxicity.
This form of poisoning is indicated by yawning, colic, blackish,
fœtid diarrhœa, and rapid loss of strength.
The animals suffer from stertorous breathing, weakness of the
pulse, and aberration of vision. They die in convulsions.
POISONING BY POPPIES.