ECE5590: Model Predictive Control 4–1

Model Predictive Control Problem Formulation

The objective of a model predictive control strategy is to:

Compute a trajectory of future control inputs that optimizes the future
behavior of plant output, where the optimization is carried out within a
limited time window

An Application Example

! Let’s examine how this might work for a golf game.

– Prior to tee-off, we analyze the course and game-time weather
conditions.
– Based on this knowledge - and the physical principles governing
ball trajectories as a function of club - we develop a model.
– Using this model, we develop a strategy for the next n golf strokes
that will land the ball in the hole.
– We execute the next shot in this optimal sequence.
– We then assess our success and use information gained to plan
the next series of shots.
! Of course a key difference in golf is that our horizon is ever-
shortening (we hope!)
! In practice, our actual play sequence will differ from the original set of
n shots in the optimal sequence.
c 2010-2015, M. Scott Trimboli
Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–2

! A part of this difference is due to unknowns (like how close to our

potential we perform).
N OTE : In practice, we will have limitations (e.g., known injuries,
maximum shot distance, etc.) that we will build in as constraints during
pre-game analysis.
Key principles of MPC demonstrated by this example include:
1. Moving prediction horizon window
2. Receding horizon control that only executes the next step
3. Need for current-time information to make the next prediction
4. Model that describes system dynamics used for prediction
5. Objective criterion based on measured difference between desired &
actual response

Model Choice for MPC

! Three popular approaches to system modeling for predictive control

design are:
– Finite Impulse Response (FIR)
– Step Response (SR)
– Transfer Function (TF)
– State-Space (SS)
! Each approach exhibits certain advantages and disadvantages - the
development in this course will focus on state-space models
! FIR, SR and TF models are of historical interest and will be presented
for completeness
! All models will assume a discrete-time setting
c 2010-2015, M. Scott Trimboli
Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–3
Finite Impulse Response

! Also known as weighting sequence, the output is related to the input

via the convolution operator
! For the single-input-single-output (SISO) case, we write
X
N
yk D gi uk#i
i D1

where uk and yk denote the input and output sequence elements,

respectively.
! The sequence gi comprises the first N coefficients of an infinite
Taylor series in the unit time delay operator, z #1 , representing the
dynamic process g.z/:
X
1
g.z/ D gi z #1
iD0

– This sequence will always decay to zero for the stable case.

! The first N values of the infinite series give a finite impulse response
which will capture the main process dynamics.
! A FIR model can be viewed as a representation of the discrete-time
pulse response (expected value of the output y in response to a
single unit pulse u).
! This model generalizes easily to the multi-input-multi-output (MIMO)
case by use of appropriate matrices.

Step Response

! Integrating the unit pulse response will give the unit step response,
c 2010-2015, M. Scott Trimboli
Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–4

X
1
h.z/ D hi z #1
i D1

where
g.z/
h.z/ D
4.z/
and 4.z/ is the one-step difference operator, 1 # z #1.
! The following plot shows the relationship between the impulse and
step response.

h4 = h3 + g 4

h3 g4

Transfer Function

! Again considering the SISO case we can write

a.z/yk D b.z/uk
b.z/
where the form gives a transfer function representation for the
a.z/
output-input ratio of the corresponding z-transforms.

c 2010-2015, M. Scott Trimboli

Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–5

! The polynomials assume the forms:

a.z/ D 1 C a1z #1 C a2z #2 C $ $ $ C anz #n
b.z/ D b0 C b1z #1 C b2z #2 C $ $ $ C bmz #m

N OTE: Performing the long division b.z/=a.z/ will yield the corresponding
unit pulse response g.z/.

State-Space

! We shall assume a linearized, discrete-time, state-space model of the

plant:
x.k C 1/ D Ax.k/ C Bu.k/
y.k/ D Cy x.k/
z.k/ D Cz x.k/
where x is an n-dimensional state vector, u is an `-dimensional input
vector, y is an my -dimensional vector of measured outputs, and z is an
mz -dimensional vector of outputs which are to be controlled.
! Since there is always finite delay between measuring y.k/ and
applying u.k/, there is no direct feed-through term.
– Hence our model is strictly proper.
! For simplicity, we shall often assume y % z in which case C will
denote both Cy and Cz .
Relationship to FIR Model

We can use the discrete-time state equations to determine the sequence

parameters resulting from a unit pulse input:

x.0/ D 0
c 2010-2015, M. Scott Trimboli
Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–6
x.1/ D B
x.2/ D AB
x.3/ D A2B
:::
x.n/ D An#1B

y.0/ D C x.0/ C Du.0/

DD
y.1/ D CB
y.2/ D CAB
y.3/ D CA2B
:::
y.n/ D CAn#1B
! Hence the FIR model sequence is given by:
g.z/ D D C CBz #1 C CABz #2 C CA2Bz #3 C CAn#1Bz #n
Relationship to Transfer Function Model

! Applying the z-transform to the state-space model,

zX.z/ D AX.z/ C BU.z/
.zI # A/X.z/ D BU.z/
X.z/ D .zI # A/#1BU.z/
and
Y.z/ D C.zI # A/#1BU.z/
! So that,
Y.z/
G.z/ D D C.zI # A/#1B
U.z/
c 2010-2015, M. Scott Trimboli
Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–7
State-Space Model with Embedded Integrator

! Returning to our state-space representation, we consider the SISO

system
xm.k C 1/ D Amxm.k/ C Bmu.k/
y.k/ D Cmxm.k/

! Recall from your study of feedback control that in order to eliminate

steady-state error from a control system we need to include integral
action.
! For model predictive controllers, a convenient way to introduce
integral action is by modifying the state-space model.
! Taking a difference operation on both sides of the state equation we
get
x m.k C 1/ # x m.k/ D Am.x m.k/ # xm.k # 1// C Bm.u.k/ # u.k # 1//

! Denote the difference of the state variable by

4x m.k C 1/ D xm.k C 1/ # xm.k/
4x m.k/ D xm.k/ # xm.k # 1/
and of the control variable by
4u.k/ D u.k/ # u.k # 1/

! Then the difference of the state equation is:

4xm.k C 1/ D Am4xm.k/ C Bm4u.k/

! We now have to compute the corresponding expression for the output

y.k/.
! Define a new augmented state vector

c 2010-2015, M. Scott Trimboli

Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–8
" #
4xm.k/
x.k/ D
y.k/
! Note that
y.k C 1/ # y.k/ D Cm.xm.k C 1/ # x m.k// D Cm4xm.k C 1/
D CmAm4x m.k/ C CmBm4u.k/

! Putting the above expressions together we get the augmented model:

" # " #" # " #
T
4x m.k C 1/ Am 0m 4x m.k/ Bm
D C 4u.k/
y.k C 1/ CmAm 1 y.k/ CmBm
" #
h i 4x .k/
m
y.k/ D 0m 1
y.k/
or simply,

x.k C 1/ D Ax.k/ C B4u.k/

" #
4x m.k/
y.k/ D C
y.k/
Example (Wang, pg. 34)

! Given a discrete-time system where

" # " #
1 1 0:5 h i
Am D Bm D Cm D 1 0
0 1 1
! Find the matrices of the augmented system .A; B; C / and calculate
the eigenvalues of A.

S OLUTION .
h i
! We have n D 2 and 0m D 0 0 .
c 2010-2015, M. Scott Trimboli
Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–9
! Then,
x.k C 1/ D Ax.k/ C B4u.k/
y.k/ D C x.k/
where

2 3
" 1 1 0 #
6
Am 0Tm 7
AD D40 1 05
CmAm I
1 1 1
2 3
" # 0:5
Bm 6 7
BD D4 1 5
CmBm
0:5
h i h i
C D 0m 1 D 0 0 1

! The eigenvalues of A are found as the roots of the characteristic

equation, .! # 1/3 D 0; or !1 D !2 D !3 D 1.

Predictive Control: SISO Case

! We will first develop the case for a simple SISO system for which the
states are directly measureable.

– The more general case will be discussed later.

! Future control inputs are denoted by:

4u.ki /; 4u.ki C 1/; : : : ; 4u.ki C Nc # 1/
where ki represents a particular time step and Nc is the control horizon.
! Information x.ki / permits prediction up to Np samples ahead; Np is
called the prediction horizon.
c 2010-2015, M. Scott Trimboli
Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–10
! Future state variables are given by:
x.ki C 1jki /; x.ki C 2jki /; : : : x.ki C mjki /; : : : ; x.ki C Np jki /
N OTE : Nc & Np .

P REDICTION .
! Future state variables:
x.ki C 1jki / D Ax.ki / C B4u.ki /
x.ki C 2jki / D Ax.ki C 1jki / C B4u.ki C 1/
D A2x.ki / C AB4u.ki / C B4u.ki C 1/
:::
x.ki C Np jki / D ANp x.ki / C ANp #1B4u.ki / C ANp #2B4u.ki C 1/
C : : : C ANp #Nc B4u.ki C Nc # 1/

! Predicted outputs:
y.ki C 1/ D CAx.ki / C CB4u.ki /
y.ki C 2/ D CA2x.ki / C CAB4u.ki / C CB4u.ki C 1/
y.ki C 3/ D CA3x.ki / C CA2B4u.ki / C CAB4u.ki C 1/
C CB4u.ki C 2/
:::
y.ki C Np jki / D CANp x.ki / C CANp #1B4u.ki / C CANp #2B4u.ki C 1/
C : : : C CANp #Nc B4u.ki C Nc # 1/

! Gathering up the the predicted output sequence into the vector

c 2010-2015, M. Scott Trimboli

Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–11

2 3
y.ki C 1/ j ki
6 7
6 y.ki C 2/ j ki 7
6 7
Y D6
6 y.ki C 3/ j ki
7
7
6 ::: 7
4 5
y.ki C Np / j ki
! Gather future input increments into the vector
2 3
4u.ki /
6 7
6 4u.ki C 1/ 7
6 7
4U D 6 4u.ki C 2/ 7
6
7I
6 ::: 7
4 5
4u.ki C Nc # 1/
! Now we can write
Y D F x.ki / C ˆ4U
where
2 3 2 3
CA CB 0 0 $$$ 0
6 7 6 7
6 CA2 7 6 CAB CB 0 $$$ 0 7
6 7 6 7
6 3 7
F D 6 CA 7 I ˆD6 CA2B $$$ 7
6 CAB CB 0 7
6 :: 7 6 ::: 7
4 : 5 4 5
CANp CANp #1B CANp #2B CANp #3B $ $ $ CANp #Nc B
O PTIMIZATION .
! For a given set-point signal r.ki / at sample time ki , the objective is to
find the best control parameter 4U such that an error function is
minimized.
! Define

c 2010-2015, M. Scott Trimboli

Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–12

2 3
1
6 7
617
Rs D 6 7
6 ::: 7 r.ki /
4 5
1
! Then we can define a cost function J as
J D .Rs ! Y /t .Rs ! Y / C 4U tR4U

! Optimal 4U that minimizes J is found from the solution of

@J
D0
@4U
! Expand J as

J D .Rs #F x.ki //t .Rs #F x.ki //#24U t ˆT .Rs #F x.ki //C4U t .ˆT ˆCR/4U
to give
@J
D #2ˆT .Rs # F x.ki // C 2.ˆT ˆ C R/4U D 0
@4U
! Solving,

4U " D .ˆT ˆ C R/#1ˆT .Rs # F x.ki //

Example

! Consider a simple first-order system:

xm.k C 1/ D axm.k/ C bu.k/
y.k/ D cxm.k/
where, a D 0:8; b D 0:1 and c D 1:0. Let Np D 10 and Nc D 4.
! Assume ki D 10, andh at this value,
i the reference r.ki / D 1 and the
t
state vector x.ki / D 0:1 0:2 .

c 2010-2015, M. Scott Trimboli

Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–13

! Determine the optimal solution 4U ' for the cases: 1) R D Œ0" and 2)
R D 10 $ I , where R is the weighting on the control input.

S OLUTION .
! Form the augmented state-space system:
" # " #" # " #
4x m.k C 1/ a 0 4x m.k/ b
D C 4u.k/
y.k C 1/ a 1 y.k/ b
" #
h i 4x .k/
m
y.k/ D 0 1
y.k/
! Therefore, F and ˆ take the following form:
2 3 2 3
CA CB 0 0 0
6 7 6 7
6 CA2 7 6 CAB CB 0 0 7
6 7 6 7
6 CA3 7 6 CA2B CAB 7
6 7 6 CB 0 7
6 CA4 7 6 CA3B CA2B 7
6 7 6 CAB CB 7
6 7 6 7
6 CA5 7 6 CA4B CA3B CA2B CAB 7
F D6 7 6
6 CA6 7 I ˆ D 6 CA5B CA4B
7
7
6 7 6 CA3B CA2B 7
6 7 6 7
6 CA 7
7 6 CA6B CA5B CA4B CA3B 7
6 7 6 7
6 CA8 7 6 CA7B CA6B CA5B CA4B 7
6 7 6 7
6 9 7 6 7
4 CA 5 4 CA8B CA7B CA6B CA5B 5
CA10 CA9B CA8B CA7B CA6B
! Note that the elements of ˆ comprise the discrete pulse response of
the system arranged in the form of a Toeplitz matrix.
! Substituting values we obtain:

c 2010-2015, M. Scott Trimboli

Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–14

2 3
1:1541 1:0407 0:91160:7726
6 7
6 1:0407 0:9549 0:84750:7259 7
ˆT ˆ D 6
6 0:9116 0:8475
7
4 0:76750:6674 7
5
0:7726 0:7259 0:66740:5943
2 3 2 3
9:2325 3:2147 3:2147
6 7 6 7
6 8:3259 2:7684 7 6 2:7684 7
T
ˆ F D66 7I T 6
ˆ Rs D 6 7
7 7
4 7:2927 2:3355 5 4 2:3355 5
6:1811 1:9194 1:9194
! For case 1) where R D Œ0", the optimal input sequence is
2 3
7:2
6 7
6 #6:4 7
4U D .ˆT ˆ/#1.ˆT Rs # ˆT F x.ki // D 6 6 0 7
7
4 5
0

c 2010-2015, M. Scott Trimboli

Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–15

! For case 2) where R D 10 $ I , we obtain

2 3
0:1269
6 7
6 0:1034 7
4U D .ˆT ˆ C 10 $ I /#1.ˆT Rs # ˆT F x.ki // D 6 7
6 0:0829 7
4 5
0:065

c 2010-2015, M. Scott Trimboli

Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–16

Receding Horizon Control

! Although the optimal parameter vector 4U contains the control

values 4u.ki /; 4u.ki C 1/,4u.ki C 2/,: : : 4u.ki C NC # 1/ , we only
implement the first sample of the sequence.
! Consider again the first-order system with state-space description
x m.k C 1/ D 0:8x m.k/ C 0:1u.k/
h it
! Let rw D 0; initial conditions are x.10/ D 0:1 0:2 and u.9/ D 0.
! As found previously, at sample time ki D 10, the optimal control was
found to be 4u.10/ D 7:2:
! Assuming 4u.9/ D 0, then we have the control signal to the plant is
u.10/ D u.9/ C 4u.10/ D 7:2.
! Compute the next plant state variable:
c 2010-2015, M. Scott Trimboli
Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–17

x m.11/ D 0:8x m.10/ C 0:1u.10/ D 0:88

! At ki D 11, new plant information is

4x m.11/ D 0:88 # 0:2 D 0:68
y.11/ D 0:88
which gives
" #
0:68
x.11/ D
0:88
! We can then calculate:
2 3
#4:24
6 7
6 #0:96 7
T #1 T T 6
4U D .ˆ ˆ/ .ˆ Rs # ˆ F x.11// D 6 7
7
4 0 5
0
and thus the optimal control input u.11/ D u.10/ C 4u.11/ D 2:96.
! Propagate the state again,
x m.12/ D 0:8x m.11/ C 0:1u.11/ D 1:0

! At ki D 12, the new plant information is

4x m.12/ D 1:0 # 0:88 D 0:12
y.12/ D 1:0
which gives
" #
0:12
x.12/ D
1:0
and

c 2010-2015, M. Scott Trimboli

Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–18

2 3
#0:96
6 7
6 0 7
4U D .ˆT ˆ/#1.ˆT Rs # ˆT F x.12// D 6
6
7
4 0 7 5
0
! Continuing,
x m.13/ D ax m.12/ C bu.12/ D 1:0
y.13/ D 1:0

4x m.13/ D 1:0 # 1:0 D 0

y.13/ D 1:0
2 3
0
6 7
607
4U D .ˆT ˆ/#1.ˆT Rs # ˆT F x.13// D 6
607
7
4 5
0
Stacking
h the the computed
i values of 4U for time instants
ki D 10 11 12 13 column-wise we obtain
2 3
7:2 #4:24 #0:96 0
6 7
6 #6:4 #0:96 0 0 7
4U ALL D 6
6 0
7
4 0 0 075
0 0 0 0
where the optimal control input sequence is given by the first row.
The optimal response follows:

c 2010-2015, M. Scott Trimboli

Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–19

Closed Loop Control System

! We’ve shown that at a given time ki , the optimal control input vector
4U is solved from
4U D .ˆT ˆ C R/ N #1.ˆT Rs # ˆT F x.ki //

N #1ˆT Rs corresponds to the

– Here, the first term, .ˆT ˆ C R/
set-point change
– The second term, .ˆT ˆ C R/ N #1ˆT F corresponds to state
feedback control (within the context of predictive control)

! We take just the first element of 4U at time ki as the incremental

control,
h i
4U .ki / D 1 0 $ $ $ 0 .ˆT ˆ C R/ N #1.ˆT RN s r.ki / # ˆT F x.ki //

D Ky r.ki / # Kmpc x.ki /

c 2010-2015, M. Scott Trimboli
Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–20
N #1ˆT Rs and Kmpc is the
where Ky is the first element of .ˆT ˆ C R/
N #1ˆT F .
first row of .ˆT ˆ C R/
! These relationships depict linear time-invariant state feedback control.
! The closed-loop system is found by using the augmented system
equation to write
x.k C 1/ D Ax.k/ # BKmpc x.k/ C BKy r.k/
D .A # BKmpc /x.k/ C BKy r.k/
– Closed-loop eigenvalues are then the roots of the closed-loop
characteristic equation:
detŒ!I # .A # BKmpc /" D 0

Predictive Control with Unmeasured States

State Estimation
! In the design of model predictive controllers, we initally assume that
the information x.ki / is available at time ki , which assumes all state
variables are measurable.
! In practice this is not the case and we must instead estimate state
values from the available process measurements.
– The general construction we devise for this task is termed an
observer.
! Assume a plant state-space model,
x m.k C 1/ D Amx m.k/ C Bmu.k/
! We can then calculate an estimate of the state variable,
xO m.k/; k D 1; 2; : : :, with an initial condition xO m.0/ and input signal
u.k/ as
c 2010-2015, M. Scott Trimboli
Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–21

xO m.k C 1/ D AmxO m.k/ C Bmu.k/

! What’s the problem with this estimate model?
– open loop prediction
– works only if initial condition is accurate
– requires stable plant - convergence only if eigenvalues of A inside
unit circle
– can’t regulate convergence rate
! Better approach ! Use feedback that relies on an error measure to
improve the estimate:
! "
xO m.k/ D AmxO m.k/A C Bmu.k/ C Kob y.k/ # CmxO m.k/
where Kob is an observer gain matrix.
! First term propagates the original model
! Second term is a correction
– Correction based on the error between the measured output and
the predicted value based on the state estimate.
! How do we choose the observer gain, Kob ?
– Define the error state to be x O m.k/; then
f .k C 1/ D x .k/ # x
m m

f .k C 1/ D A x
x m m fm .k/ # Kob Cm x
f .k/
m
D .Am # Kob Cm/x
f .k/
m

– With an initial error state xm.0/, we have

f .k/ D .A # K C /k x
x m m ob m fm .0/

c 2010-2015, M. Scott Trimboli

Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–22
! Unlike the open-loop estimator case, now we can exploit the degrees
of freedom in the observer gain matrix to affect the estimator
convergence rate.
! One way to do this is to place the closed-loop poles of the error
system matrix at desired locations in the complex plane.
Example
! The linearized equation of motion of a simple pendulum is:
d 2# 2
C ! # Du
dt 2
! Design an observer that reconstructs the pendulum angle, #, from
measurements of its angular velocity, d#=dt . Assume ! D 2 rad/sec
and sampling interval 4t D 0:1 sec.
! Select states, x1.t/ D # and x2.t / D #P ; the continuous time
state-space model is,
" # " #" # " #
xP 1.t/ 0 1 x1.t/ 0
D C u.t/
xP 2.t/ #! 2 0 x2.t/ 1
" #
h i x .t/
1
y.t/ D 0 1 :
x2.t/
! The discrete-time model is found from the Matlab function [sysd] =
c2d(sysc, .1):
" # " #" # " #
x1.k C 1/ 0:9801 0:0993 x1.k/ 0:0050
D C u.k/
x2.k C 1/ #:3973 0:9801 x2.k/ 0:0993
" #
h i x .k/
1
y.k/ D 0 1
x2.k/

c 2010-2015, M. Scott Trimboli

Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–23

! What if we just use the pendulum model alone for (open-loop)

prediction?

– Assume u.k/ D 0, and the initial conditions, #.0/ D x1.0/ D 1 and

#.0/ D xP2.0/ D 0:
– Now let’s assume we take a guess at the initial conditions of the
estimator as xO 1.0/ D 0:3 and xO 2.0/ D 0.
– Results are as follows:

c 2010-2015, M. Scott Trimboli

Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–24

– It’s easy to see that the open-loop estimation was unable to

recover from an error in the initial conditions.

! Now let us design and implement an observer to predict the angle of

the pendulum.
h it
– Assume the observer gain is given by Kob D $1 $2 .
– The close-loop characteristic polynomial of the observer is
det .!I # ŒAm # Kob Cm"/ D .!#0:9801/.!C$2#:9801/#:3973(.$1#:0993/

– Let’s set the observer poles at 0:1 and 0:2 for good convergence
speed.
– Equating the above expression to the desired closed cloop
characteristic equation for the observer, namely
.! # 0:9801/.! C $2 # :9801/ # :3973 ( .$1 # :0993/ D .! # 0:1/.! # 0:2/

c 2010-2015, M. Scott Trimboli

Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–25

we can solve for the unknown observer gains

" #" # " #

0 1 $1 1:6601
D
#:3973 #:9801 $2 #0:9800
" # " #
$1 #1:6284
D
$2 1:6601
" #" # " #
0 1 $1 1:6601
D
#:3973 #:9801 $2 #0:9800
" # " #
$1 #1:6284
D
$2 1:6601
– Implementing the estimator with the computed observer gives the
results depicted below.

c 2010-2015, M. Scott Trimboli

Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–26

– Here we see the observer quickly converges to the true value in

approximately three time steps.

Predictive Control Using State Estimates

! In cases where the state variable x.ki / at time ki is not measurable,

we use an observer in the implementation of predictive control.
O i C 1/ D Ax.k
x.k O i / C B4u.ki / C Kob .y.ki / # C x.k
O i //
! "
! Note that the state matrices, A; B; C represent the augmented
model used for predictive controller design.
! The corresponding modified cost function J can then be written
O i //T .RN s r.ki /#F x.k
J D .Rs #F x.k O i //#24U T ˆT .Rs #F x.k
O i //C4U T .ˆT ˆC

– The optimal solution is thus:

c 2010-2015, M. Scott Trimboli

Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–27

N #1ˆT .Rs # F x.k

4U ' D .ˆT ˆ C R/ O i //

! Application of receding horizon control leads to the optimal solution at

time ki :
O i/
4u.ki / D Ky r.ki / # Kmpc x.k

– This is standard state feedback control with estimate x.k

O i /.

! Let’s now examine the close-loop properties of this system.

! First, for the closed-loop state equations,
x.k C 1/ D Ax.k/ C B4u.k/
O
D Ax.k/ C BKy r.k/ # BKmpc x.k/

! The closed-loop observer error equation is:

Q C 1/ D .A # Kob C /x.k/
x.k Q

where
Q
x.k/ D x.k/ # x.k/
O

! Substituting for x.k/

O we can write:
Q
x.k C 1/ D .A # BKmpc /x.k/ # BKmpc x.k/ C BKy r.k/

! Combining the error and system state equations,

" # " #" # " #
Q C 1/
x.k A # Kob C 0n(n Q
x.k/ 0n(m
D C r.k/
x.k C 1/ #BKmpc A # BKmpc x.k/ BKy
– Since the system matrix is lower block diagonal, the eigenvalues
are given by the eigenvalues of the diagonal blocks, A # Kob C and
A # BKmpc respectively (separation principle).

Example
c 2010-2015, M. Scott Trimboli
Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–28

The augmented model for a double integrator plant is given by

x.k C 1/ D Ax.k/ C B4u.k/

y.k/ D C x.k/
where
2 3 2 3
1 1 0 0:5 h i
6 7 6 7
A D 4 0 1 0 5; B D 4 1 5; C D 0 0 1
1 1 1 0:5
! Design a state estimate predictive control system and simulate the
closed-loop response for a set-point change.
! Design specifications:
Nc D 5
Np D 30
rw D 10

! Place the observer poles at: 0:01, 0:0105, 0:011 to give fast dynamic
response.
! The open-loop plant has three eigenvalues at one (by inspection), two
from the double-integrator and one from the embedded integrator
structure. Using the Matlab function place, we find the observer gains
from:

h i
Pole D 0:01 0:0105 0:011 I
! "
Kob D place A C Pole 0I
0 0

which gives,

c 2010-2015, M. Scott Trimboli

Lecture notes prepared by M. Scott Trimboli. Copyright "
ECE5590, Model Predictive Control Problem Formulation 4–29

2 3
1:9685
6 7
Kob D 4 0:9688 5
2:9685
! Using the given performance parameters, the state feedback
controller gain is computed to be,
h i
Kmpc D 0:8984 1:3521 0:4039

! The resulting closed-loop eigenvalues are at 0:3172 ˙ j 0:4089 and

0:3624.

c 2010-2015, M. Scott Trimboli

Lecture notes prepared by M. Scott Trimboli. Copyright "
(mostly blank)
(mostly blank)
(mostly blank)