TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 13, No. 3, March 2
015,
pp. 497 ~ 50
2
DOI: 10.115
9
1
/telkomni
ka.
v
13i3.721
5
497
Re
cei
v
ed Au
gust 2, 201
4; Re
vised Sept
em
ber
18, 20
14; Accepted
Octob
e
r 16, 2
014
A Nonlinear System of Generalized Predictive Control
Jingfan
g Wa
ng
Schoo
l of Information Sci
enc
e & Engin
eer
in
g, Huna
n Inter
natio
nal Ec
ono
mics Univers
i
t
y
,
Cha
ngsh
a
, Chi
na, postco
de: 410
20
5
E-mail: matlab
_b
ysj@
12
6.co
m
A
b
st
r
a
ct
Genera
l
i
z
e
d
pr
edictiv
e co
ntrol
(GPC) a
l
gor
ithm h
a
s
b
e
e
n
app
lie
d to
al
l k
i
nds
of i
n
d
u
stry contro
l
systems. But systemic
and
effective
metho
d
for non
lin
ear
sy
stem h
a
s not b
een fo
und. To this pro
b
le
m, thi
s
pap
er inte
grat
es the ch
aract
e
ristics of PID
techno
l
ogy
a
nd GPC, pres
ent a PID g
e
n
e
rali
z
e
d
pred
ic
tive
control a
l
gor
ith
m
for a class of
nonl
ine
a
r system,
a
nd i
m
pr
o
v
es the contro
l
qua
lity of the system.
Ke
y
w
ords
: ge
nera
l
i
z
e
d
pre
d
i
c
itive contro
l,
PID, nonli
n
e
a
r system, si
mul
a
ti
on
Copy
right
©
2015 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
Since the 19
90s, the mo
dern in
du
stri
al bec
ome q
u
ickly to co
mplex, large
-
scale and
automation d
e
velopme
n
t, therefo
r
e ma
n
y
industria
l control sy
stem
s with a high
degre
e
of non-
linearity, cou
p
ling, whe
n
the variability and large ti
me-del
ay cha
r
a
c
teri
stics,
and
the existence of
deman
dingth
e
con
s
trai
nt condition
s. Th
ese requi
re
t
he automati
c
control te
ch
nology to pro
v
ide
importa
nt technical gu
aran
tee for th
e re
alizatio
n
of ef
ficient, safe,
high-quality
mass p
r
o
duct
i
on.
Only rely
on t
r
adition
al
con
t
rol techniq
u
e
s
su
ch a
s
PI
D
control al
g
o
rithm
ca
n n
o
t solve
all th
ese
probl
em
s. T
herefo
r
e,
we
mu
st see
k
more
adva
n
ce
d
cont
rol
metho
d
s to
meet th
e
high
requi
rem
ents
of modern ind
u
strial
autom
atic co
ntrol te
chn
o
logy
. Th
e contin
uou
s
developm
ent
of
comp
uter te
chnolo
g
y bro
u
ght great cha
nge
s to the
control system
hardware,
a
nd on
this
ba
sis,
the new
control algorith
m
- predi
ct
ive co
ntrol slo
w
ly d
e
velope
d.
Gene
rali
zed
predi
ctive control algo
rithm
with a feedback
corre
c
tion, multi-ste
p
predi
ction, ro
lling optimiza
t
ion cont
rol
method,
an
d
thus control
the effect of good, stro
n
g
robu
stne
ss, this control al
gorithm
can
be used to
control
compl
e
x industri
a
l p
r
ocesse
s [1], or
difficult to est
ablish preci
s
e
the mathem
a
t
ical mo
d
e
l, a
nd ha
s b
een
use
d
in
the control system
of
the ch
emical, petrole
um,
metallurgy, machi
n
e
r
y a
nd othe
r ind
u
strial
se
cto
r
s, and
refle
c
ts the
cha
r
a
c
teri
stics of the a sup
e
rio
r
tradition
al cont
rol system, is a pro
m
ising n
e
w
cl
ass of comp
u
t
er
control algo
rit
h
ms.
From the a
b
o
ve, we can
see that, alth
ough
p
r
edi
cti
v
e control ha
s many adva
n
tage
s,
however, the
traditional P
I
D cont
rol al
gorithm b
e
ca
use of its st
ructure is
sim
p
le, clea
r and
con
c
i
s
e alg
o
ri
thm robu
stne
ss, lo
w co
ntro
l algor
ithm m
odel a
c
cura
cy requirement
s and o
peration
easily a
c
cept
ed in actu
al indu
strial p
r
o
c
ess co
nt
rol o
r
occu
py a dominant po
sit
i
on, con
s
train
ed
gene
rali
zed p
r
edi
ctive cont
rol to the appl
icat
ion of the
actual in
du
strial field.
Relatively According
to th
e above,
we
can
see tha
t
the PID co
ntrol te
chn
o
l
ogy with
predi
ctive
co
ntrol
com
b
in
ation i
s
a
ne
w
re
sea
r
ch
dire
ction; the
integration
of the
re
spe
c
tive
cha
r
a
c
teri
stics of this ne
w control algo
rit
h
m has
ce
rtai
n signifi
can
c
e
.
2. Generalized Predictive Control
2.1. Basic Pr
inciple of Ge
neralize
d
Predictiv
e
Control
Gene
rali
zed
Predi
ctive Co
ntrol (GP
C
) [
2
] wa
s p
r
op
o
s
ed in 19
84
by Clarke et al. GPC
based on
ge
nerali
z
e
d
min
i
mum varia
n
ce co
ntrol, th
e
introdu
ction
of a multi-ste
p
pre
d
ictio
n
i
dea
to make it a rando
m noi
se, sig
n
ifica
n
t
ly impr
oved
the ability o
f
anti-distu
r
b
ance and de
lay
variation. GP
C's
ba
sic
structure a
s
sh
own
by
the p
r
edi
ction m
o
d
e
l, GPC is
ro
lling optimi
z
a
t
ion
and feed
ba
ck corre
c
tion co
nsi
s
ts of thre
e parts.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 13, No. 3, March 2
015 : 497 – 5
0
2
498
Figure 1. GPC structu
r
e schem
atic
2.2. Generali
zed Predic
ti
v
e
Control M
odel
In 1987, Cla
r
ke et al gen
e
r
alized pa
ram
e
tric mo
del b
a
se
d co
ntrol. Its essen
c
e i
s
ba
sed
on gen
erali
z
ed minimum
variance introdu
ce
d the i
dea of the multi-ste
p
predictio
n, ran
d
o
m
noise, anti-lo
ad distu
r
ba
n
c
e an
d dela
y
variat
ion capability has been sig
n
ificantly impro
v
ed
robu
stne
ss suitable for o
p
en-lo
op u
n
sta
b
le, non
-m
ini
m
um pha
se
delay syste
m
. The ba
sic f
o
rm
of the al
gorith
m
, the follo
wi
ng di
screte
di
fferentia
l e
q
u
a
tions to d
e
scrib
e
the
mat
hematical m
o
del
of the controlled obje
c
t:
A
z
y
t
B
z
μ
t1
C
z
ω
t
/
∆
(1)
Whe
r
e
A
z
、
B
z
、
C
z
are th
e ba
ckward
shift op
erato
r
z
polynomial.
μ
t
and
yt
is
respe
c
tively rep
r
e
s
ent th
e input a
n
d
output of the controll
ed
obje
c
t,
=
z
, is
differenc
e
operator.
ω
t
is
not rel
a
ted t
o
the
seq
u
e
n
ce
of ra
ndo
m variabl
es.
C
on
stant
n
an
d
n
is not
related to the
seq
uen
ce of random va
riab
les.Con
s
tant
C
z
=1.
3. A Class o
f
Nonlinear G
e
neralized Predictiv
e
Control
3.1. Nonline
a
r Sy
stem Model
Gene
ral no
nli
near
system
can b
e
used
to describ
e the followi
ng I / O model with first-
orde
r del
ay:
y
t
f
yt
1
,
…
,
y
t
n
,
u
t1
,
…
,
ut
m
(
2
)
Whe
r
ein m,
n, resp
ectivel
y
, is known as
a sy
stem
input output gradatio
n times o
r
the
ceiling in order: f () is the unknown, and is
y
t1
,
…
,
y
t
n
,
u
t
1
,
…
,
u
t
m
‘s
nonlin
ear fun
c
tion.The foll
owin
g co
ndition is satisfied
:
a)
f(0,0, …0)=0,
b)
f() is abo
ut
,
…
,
,
,
…
,
continuou
sly
differentiabl
e, and variou
s
partial de
rivat
i
ves bou
nde
d
.
3.2. Equiv
a
le
nt Time-v
ar
y
i
ng Linear Sy
stems
In order to fa
cilitate
the
an
alysis,
we
co
ns
id
er only the input-o
utpu
t system, the
system
eq. (2), we ob
tain the following theorem b
y
analyzing:
Theorem
3.
1
[3]: sati
sfy the
con
d
itio
n (1
),
(2) n
on-lin
ea
r
system eq.
(2
)
can
be
equivalently e
x
presse
d as
f
o
llows
when denatu
r
ed system:
y
t
f
y
t1
,
…
,
y
tn
,
u
t1
,
…
,
u
tm
(3)
=
a
t
y
t1
⋯
a
t
y
tn
b
t
u
t
1
⋯
b
t
u
t
m
Formula’
s a
i
,b
j
is bou
nded
time-varying
coeffici
ents.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Nonline
a
r
System
of Generali
z
e
d
Pre
d
ictive Co
ntro
l (Jingfa
ng Wang)
499
3.3. Time-v
ar
y
i
ng Parameter Iden
tifica
tion
Theo
rem 3.1
only given t
he existe
nce
of t
he equiv
a
lent linea
r
system of the
origin
al
nonlin
ear sy
stem, did n
o
t
give the
spe
c
ific
par
amet
ers of the
co
rre
sp
ondi
ng l
i
near sy
stem
in
every mome
nt. Since the nonlinea
r fu
nction f () of
the partial d
e
rivative is u
n
kn
own, i.e.
the
linear
rep
r
e
s
entation of th
e re
spe
c
tive coeffici
ents i
n
the Equatio
n (3) i
s
un
kn
own. And
ca
n be
see
n
from
th
e above
de
ri
vation, the co
efficient
a
i1
,
…n
,
b
j
1
,
…m
is
not only with
the no
nline
a
r functio
n
f
(), also
a
s
soci
ated
with
the
input
and
o
u
tput of the
system, th
at
is
con
s
tantly ch
angin
g
with ti
me. Ne
ed o
n
line ide
n
tifi
ca
tion
of
the
s
e time-varying
coeffici
ents, on
this ba
sis, re
-desi
gn of the controlle
r.
Seen by The
o
rem 3.1, the
system
Equa
tion (2)
can b
e
expre
s
sed
as:
y
t
ϕ
t
θ
t
ξ
t
(4)
Whe
r
e
ϕ
t
y
t1
,
…
,
y
tn
,
u
t1
,
…
,
u
tm
(
5
)
θ
t
a
t
,
a
t
,
…a
t
,
b
t
,
b
t
,
…
,
b
t
3.4. Contr
o
ller Desig
n
By the formul
a (3
)
sho
w
s,
the w.
Cha
r
g
ed o
b
ject typ
e
eq.(2) is
e
quivalent lin
e
a
r time
-
v
a
ry
ing sy
st
e
m
s:
y
t
a
t
y
t1
⋯
a
t
y
tn
b
t
u
t1
⋯
b
t
u
t
m
+
ξ
t
(6)
Theo
rem 3.1
sho
w
s that
|
a
|
、
b
r
.
A
ssu
me 1:
a
t
∑
a
、
F
t
∞
i
0
,
…n
b
t
∑
b
、
F
t
∞
i
0
,
…m
Here
a b
a
si
s functio
n
a
p
p
r
oximation
wi
th va
ria
b
le coefficient
s.
A
c
tual appli
c
at
ion,
the
interception o
f
a finite number of term
s, that:
a
t
∑
a
、
F
t
i
1
,
…
n
b
t
∑
b
、
F
t
i
1
,
…m
Y
t
a
,
F
ty
t1
⋯
a
,
F
ty
t1
⋯
a
,
F
ty
t
n
…
a
,
F
ty
tn
b
,
F
tu
t1
⋯
b
,
F
tu
t
1
…
b
,
F
tu
tm
⋯
b
,
F
tu
tm
ε
t
(7)
y
t
ϕ
t
θ
ε
t
(8)
Whe
r
e,
θ
a
,
,
…,a
,
,
…
,
a
,
,
…
,
a
,
,
b
,
,
…,b
,
,
…
,
b
,
,
…
,
a
,
ϕ
t
y
t
1
F
t
,…,
y
t
1
F
t
,
…y
t
n
F
t
…
y
t
n
F
t
ut
1
F
t
,
…
,
u
t
1
F
t
,
…
u
t
m
F
t
…
u
t
m
F
t
]
Here is in
clud
ing noi
se an
d
unmodel
ed d
y
namics.
A
ssu
me 2:
ε
t
is bound
ed,
|
ε
t
|
k
,
k
is the no
rmal
numbe
r. For
solving
cont
rol law
conve
n
ien
c
e,
(3-6
) can be
expre
s
sed a
s
:
A
t
,
z
y
t
B
t
,
z
u
t1
ε
t
(9)
A
t
,
z
a
,
F
t
⋯
a
,
F
t
z
⋯
a
,
F
t
⋯
a
,
F
t
z
B
t
,
z
b
,
F
t
⋯
b
,
F
t
z
⋯
b
,
F
t
⋯
b
,
F
t
z
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 13, No. 3, March 2
015 : 497 – 5
0
2
500
So the time-varying pa
ra
meter e
s
tima
tion in
the o
r
iginal
syste
m
Equation
(3) int
o
Equation
(4
)
in the fixed
-
length p
a
ram
e
ters e
s
timat
ed. Paramete
r ide
n
tificatio
n
u
s
ing
ada
p
t
ive
algorith
m
as f
o
llows:
θ
t
P
θ
t1
ϕ
ϕ
ϕ
e
t
y
t
ϕ
t
1
θ
t
1
The above fo
rmula, P
r
is p
r
oje
c
tion op
erand,an
d is u
s
ed to locate
)
(
t
ˆ
on a co
mpa
c
t
set
c.
A
t
,
z
y
t
B
t
,
z
u
t1
ε
t
(10)
Estimation m
odel Equ
a
tio
n
(10
)
can
b
e
used a
s
th
e GPC
plant
model, the
desi
gn of
gene
rali
zed p
r
edi
ctive cont
rolle
r, as the
cont
rolle
r of the origi
nal no
nlinea
r syste
m
.
4. The Gener
a
lized Predic
tiv
e
of a Class of
Nonlinear Sy
stems PID Control
Algorithm
4.1. Plant model
By theorem
3
.
1, spline
ba
sis fun
c
tion
s [4] better tha
n
other
polyno
m
ial ba
sis fu
nction
s
smooth a
nd
simple
cal
c
ul
ation, usin
g this time
-va
r
ying co
efficient
s of the cubi
c spli
ne ba
si
s
function a
pproximation. Finally, we
ca
n cal
c
ulate the
estimated m
o
del:
A
t
,
z
y
t
B
t
,
z
u
t1
ε
t
(11)
Whe
r
e eq.(11
)
can be used
as
the GPC cha
r
ge
d li
ke
a mod
e
l, the
desi
gn
of gen
erali
z
e
d
predi
ctive co
ntrolle
r and
controlle
r a
s
a
n
element no
nlinea
r syste
m
s.
4.2. Contr
o
ller Desig
n
Formul
a
(11
)
wa
s
cha
r
ge
d
with the
obj
e
c
t mo
del, mo
re than
on
e of
a
cla
s
s of n
o
nlinea
r
system
s GPC princi
ple de
si
gn PIDGPC
controlle
r. The
performan
ce
index functio
n
we get:
J
t
E
∑
k
∆
e
tj
k
e
tj
k
∆
e
tj
∑
λ
j
∆
ut
j
12et
∆
et
0
e
t
j
ω
t
j
y
t
j
(12)
Whe
r
e E is th
e mathemati
c
al expecta
tio
n
, N is the predictio
n hori
z
on,
N
is cont
rol
ling a
time domain,
λ
j
is a wei
ghtin
g coeffici
ent, we set the co
nstant
λ
、
k
、
k
、
k
, res
p
ec
tively, for
the coefficie
n
t
of the proportional term,
the int
egral term coeffici
e
n
t and deriva
t
ive coefficie
n
t,
y
t
j
is
y
t
forward j
–
th step p
r
e
d
i
ction, the se
t value of the
ω
t
j
for a given
softenin
g
seq
uen
ce.
ω
t
y
t
ω
t
j
ηω
tj
1
η
y
t
j
j1
,
2
,
…
,
N
Whe
r
ein
yt
is the set value of
the time point t,
η
0
1
is
soften fac
t
or.
Introdu
ction o
f
Diopha
ntine
equation
s
.
1
E
A
∆
Z
F
E
B
G
Z
H
(13)
Whe
r
ein, E
j
,
F
j
,
G
j
,
H
j
is a polynomi
a
l for Z
1
. Through a
se
ries of cal
c
ul
ations a
nd
derivation
s
e
v
entually cam
e
to the following formula:
∆
u
t
P
T
y
r
t
P
T
F
j
η
j
y
t
P
T
H
j
∆
u
t
1
(
1
4
)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Nonline
a
r
System
of Generali
z
e
d
Pre
d
ictive Co
ntro
l (Jingfa
ng Wang)
501
u
t
u
t
1
∆
u
t
(15)
Be seen fro
m
the above derivation of the plant
para
m
eters are
known,
the PID indirect
gene
rali
zed p
r
edi
ctive cont
rolle
r is de
sig
ned a
s
follows:
To a
spe
c
ifie
d fore
ca
st do
main
N, co
ntrol of time-d
o
m
ain
N
u
, the weighted
co
nst
ant
,
as well as the
numbe
r of PID co
ntrol pa
rametersk
p
、
k
i
、
k
d
.
(1) T
he linea
rization of no
n
linear
system
s,
have to the GPC model
para
m
eters
matrix A,
B;
(2) T
he polyn
omial E
j
,
F
j
,
G
j
,
H
j
is solv
ed by the the Diop
hantine
equatio
n;
(3) T
he control amount is
solved by the
formula (1
4),
(15);
(4)
t
t
1, and returns to (1
).
5. Simulation and Con
c
lusions
Gene
rali
zed
Predi
ctive Co
ntrol u
s
ing
m
u
lti-step
pred
iction, do
mai
n
N, a
s
well
as th
e
control domai
n N
u
foreca
st incre
a
se of these two pa
ram
e
ters
with a single step p
r
e
d
iction, PID-
GPC adde
d para
m
eter
k
p
、
k
i
、
k
d
,
these para
m
e
t
ers and cont
rol weig
hting con
s
tant
s
λ
,s
often
the sele
ction
factor
η
control
performance w
ill have an i
m
portant impact.
Controlled o
b
j
ect:
y
t
5y
t
1
y
t
2
1
y
t
1
2
y
t
2
2
y
t
3
2
u
t
1
1
.
1u
t
2
y
1
y
2
0
;
y
3
1
,
u
1
1
,
u
2
1
The GPC pa
ramete
rs ta
ken pre
d
iction
field N = 20, control the time domain N
u
1,
η
0
.
95
,
λ
0
.
3,
ρ
0
.
6
,
PID parameters is
k
p
0
.
5
,
k
i
0
.
8
,
k
d
5
.
6.
The followin
g
plans were the GPC and PID-GP
C tracking
y
r
t
1 step
re
spo
n
se
simulat
i
o
n
re
sult
s.
Figure 2. Wh
en
λ
0
.
3, The simu
lation re
sults
of the conven
tional GPC
Figure 3. Wh
en
λ
0
.
3, k
p
0
.
5
,
PID-GPC
simulat
i
o
n
re
sult
s
The
simulat
i
o
n
re
sult
s
sho
w
t
hat
t
h
e
i
m
pr
ove
d
con
t
roller PID-G
P
C the
co
ntrol effect
than tradition
al GPC cont
rol effect, ca
n effect
ively sho
r
ten the t
r
ack time, a
nd enh
an
ce
the
robu
stne
ss of
the co
ntrol,
sup
p
re
ss ove
r
sh
oot
smoot
her
cont
rol a
c
tion, to a
c
hi
eve the control
effec
t.
Referen
ces
[1]
Xi
ao-N
i
n
g
Du,
Yu-Geng
Xi, Shao-Y
u
a
n
. Predict
iv
e contro
l of compl
e
x
s
y
stems res
e
a
r
ch.
Electric
Autom
a
tion
. 20
05; 22(5): 4-1
0
.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 13, No. 3, March 2
015 : 497 – 5
0
2
502
[2]
Rou
han
i R, R
K
Mehra. Mo
d
e
l Al
gorithm
ic
Contro
l (MAC). Basic T
heoret
ical Pro
perti
es.
Autom
a
tion
.
199
5; 18(4): 40
1-41
4.
[3]
Guo Ji
an. Ge
nera
lize
d
Pre
d
i
ctive C
ontro
l
of a cl
ass
of
non
lin
ear s
y
st
ems. [Nan
jin
g
Univ
ersit
y
of
Scienc
e an
d T
e
chn
o
lo
g
y
Doc
t
oral Diss
e
rtati
on. 200
2: 13-3
8
.
[4]
Gao Qinhe, W
ang Sun
an,
Huan
g Xi
an
xian
g. Gener
al
i
z
ed Pred
ictive
Control of var
y
in
g s
y
ste
m
param
eters.
Computer Si
mul
a
tion.
20
08; 25
(2): 181-1
83.
[5]
Clarke DW
, et al. Genera
lize
d
Productive C
o
ntrol.
Autom
a
tic.
1987; 23(
2): 137-
162.
[6]
Rou
han
i R, Metha RK. Mo
d
e
l Alg
o
rithmic
Contro
l (MAC)
. Basic T
heoretical Pro
perti
es Automatic
.
198
2; 18(3): 40
1-41
4
.
Evaluation Warning : The document was created with Spire.PDF for Python.