Int
ern
at
i
onal
Journ
al of Ele
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
11
,
No.
1
,
Febr
uar
y
2021
,
pp.
300
~
318
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v11
i
1
.
pp
300
-
318
300
Journ
al h
om
e
page
:
http:
//
ij
ece.i
aesc
or
e.c
om
Des
i
gn o
f a mode
l
refer
ence adapt
ive PID
co
nt
rol
algorith
m
for a t
ank system
Yohan
Da
rc
y Mf
ou
m
boul
ou
Depa
rte
m
ent of
El
e
ct
ri
ca
l
,
Elec
tr
onic
,
and
Com
pute
r Engineering
,
Cape
Pen
insula
Univer
sit
y
of
T
e
chnol
og
y
,
South
Afric
a
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
l
e hist
or
y:
Re
cei
ved
M
a
r 31,
2020
Re
vised
Jun
2
0
,
20
20
Accepte
d
J
ul
6
, 2020
Thi
s
pape
r
desc
ribe
s
the
design
of
an
ada
pti
v
e
cont
roller
base
d
on
m
odel
ref
ere
n
ce
ad
apt
i
ve
PID
cont
rol
(
MRA
PID
C)
to
s
ta
biliz
e
a
two
-
tank
proc
ess
when
la
rge
va
ria
ti
ons
of
par
amete
rs
and
e
xte
rna
l
disturbance
s
aff
e
c
t
the
c
losed
-
loop
s
y
stem.
To
ac
hi
eve
tha
t
,
a
n
innova
t
ive
s
truc
tur
e
of
the
ad
apt
iv
e
PI
D
cont
roller
is
d
efi
ned
,
an
additi
onal
PI
is
desig
ned
to
m
ake
sure
tha
t
the
r
efe
ren
ce
m
odel
produc
es
stabl
e
output
signals
and
thre
e
ad
aptive
ga
ins
are
in
cl
uded
to
guar
an
te
e
stabi
lit
y
and
robustness
of
the
c
losed
-
loop
s
y
stem.
The
n
,
the
p
erf
orm
an
ce
of
the
m
odel
r
efe
ren
ce
ada
pt
ive
PID
cont
roll
e
r
on
the
beha
viour
of
the
cl
osed
-
loo
p
sy
st
em
is
compare
d
to
a
PI
cont
roll
e
r
designe
d
on
MA
TL
AB
w
h
en
both
cl
osed
-
loop
s
y
st
ems
are
under
var
ious
conditio
ns.
The
result
s
demons
tra
te
tha
t
th
e
MRA
P
IDC
per
form
s
signifi
c
ant
l
y
be
t
te
r
tha
n
th
e
co
nvent
ion
al
PI c
ontroller
.
Ke
yw
or
d
s
:
Ad
a
ptive
Lineariza
ti
on
MIT
MR
AP
I
DC
Nonlinea
r
Param
et
ers
Stabil
it
y
Thi
s is
an
open
acc
ess arti
cl
e
un
der
the
CC
B
Y
-
SA
l
ic
ense
.
Corres
pond
in
g
Aut
h
or
:
Yoha
n Darcy
Mfo
um
bo
ulou,
Dep
a
rtem
ent o
f
Ele
ct
rical
, El
ect
ronic an
d
C
om
pu
te
r
E
ng
i
ne
erin
g,
Ca
pe
Pe
nin
s
ul
a Unive
rs
it
y of Tech
nolo
gy,
Be
ll
vill
e Ca
m
p
us
, P.
O
B
ox 75
30, So
uth
A
fr
i
ca.
Em
a
il
: fab
olou
s8
6y
o@
ya
hoo.fr
1.
INTROD
U
CTIO
N
Ad
a
ptive
c
on
tr
ol
of
uncertai
n
processes
has
beco
m
e
m
or
e
and
m
or
e
im
po
rtant
in
in
du
st
r
y.
Ad
a
ptive
con
t
ro
ll
ers
dif
f
er
from
or
di
na
ry
on
e
s,
be
cau
se
th
ei
r
pa
ram
et
ers
are
va
ria
ble,
an
d
the
re
is
a
m
echan
ism
fo
r
adjustin
g
th
ese
par
am
et
ers
on
li
ne
base
d
on
s
ign
al
s
in
the
s
yst
e
m
[1
]
.
The
desig
n
of
an
a
dap
ti
ve
PI
co
nt
ro
ll
er
to
sta
bili
ze
a
m
ass
da
m
per
-
s
pr
i
ng
syst
em
u
nd
e
r
pa
ram
et
er
s’
uncertai
nties
was
pro
po
se
d
in
[2
]
.
Th
e
des
ign
e
d
adap
ti
ve
PI
c
ontr
oller
ad
j
ust
s
to
par
am
et
ers’
var
ia
ti
ons,
a
nd
the
outp
ut
of
the
process
fol
lows
the
set
points,
reg
a
rd
le
ss
of
t
he
valu
es
of
th
e
par
am
et
ers.
But
it
do
es
not
gu
a
ran
te
e
sta
bi
li
t
y
wh
en
e
xtern
al
dist
urban
c
es
an
d
la
rg
e
var
ia
ti
o
ns o
f param
et
ers
occ
ur.
In
[3
]
,
t
he
de
s
ign
of
a
PID
con
t
ro
ll
er
on
MATLAB
t
o
m
ai
ntain
the
le
vel
of
li
qu
i
d
const
ant
in
a
co
up
le
d
-
ta
nk
syst
e
m
(CTS)
was
pro
posed
.
T
he
c
on
tr
ol
par
am
et
ers
we
re
f
ound
us
i
ng
the
tria
l
a
nd
erro
r
m
et
ho
dolo
gy
and
t
he
re
su
l
ts
wer
e
a
na
l
yse
d
in
M
A
TLAB/Si
m
ulink
en
vir
onm
ents.
Propo
rtiona
l
(P
),
pro
portion
al
in
te
gr
al
(PI),
pro
portio
nal
der
iv
at
ive
(P
D
)
an
d
pr
op
or
ti
onal
integ
ral
der
iv
at
ive
(PID
)
co
ntr
ollers
wer
e
ap
plied
on
the
pr
oces
s
an
d
t
heir
pe
rfor
m
ances
we
re
c
om
par
ed
t
o
sel
ect
t
he
m
os
t
su
i
ta
ble
con
t
ro
l
so
luti
on.
T
he
PI
D
c
ontr
oller
sh
owe
d
supe
r
ior
res
ults,
but
it
did
no
t
gu
a
ran
te
e
sta
bili
ty
to
disturba
nce
s
and
var
ia
ti
ons
of p
l
ant p
a
ram
et
ers.
A
dap
ti
ve
c
on
trolle
rs,
as
oppo
s
ed
to
c
onven
ti
onal
co
nst
ant
gain
co
nt
ro
ll
ers
(
PID
c
on
t
ro
ll
ers
),
are
v
ery
e
ff
ect
ive
in
ha
nd
li
ng
sit
uatio
ns
w
her
e
the
var
ia
t
ion
s
of
par
am
et
ers
an
d
e
nvir
onm
ental
chang
es
are
ver
y
f
reque
nt
with
the
ap
plica
ti
on
of
m
od
el
ref
ere
nce
ad
aptive
co
ntr
ol
schem
e
in
a
fi
rst
order
syst
e
m
[4
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
8
8
-
8708
Desig
n of
a model ref
erenc
e
adapti
ve PI
D
c
ontrol
…
(
Y
ohan
Darcy Mf
oum
boulou
)
301
They
no
ti
ce
d
that
the
new
e
r
adap
ti
ve
c
on
t
r
ol
sc
hem
es
cou
ld
not
co
pe
w
it
h
dr
ast
ic
cha
ng
e
s
in
loads,
inerti
as
and
f
or
ces
,
un
pr
e
dicta
ble
an
d
s
udde
n
fa
ults,
or
f
reque
nt,
or
unf
or
esee
n
disturba
nces.
Most
co
nv
e
ntion
al
P
I
D
con
t
ro
ll
ers
wit
h
co
ns
ta
nt
gai
n
wer
e
al
s
o
unable
to
co
pe
with
these
pro
blem
s.
Fo
r
this
r
easo
n,
the
a
uthors
dev
el
op
e
d
a
c
on
t
ro
l
te
c
hn
i
que
t
o
s
olv
e
the
se
pro
blem
s
and
ad
de
d
a
n
a
dap
ta
ti
on
gain
to
s
how
t
he
e
f
fects
on
the syst
em
p
erfor
m
ance.
An
ada
ptive
c
on
t
ro
l
al
gorith
m
of
a
wate
r
t
ank
m
od
el
wa
s
sim
ulate
d
in
[5
]
.
It
was
c
oncl
ud
e
d
th
at
,
com
par
ed
t
o
the
one
de
gr
ee
of
f
ree
do
m
(1DOF)
al
gorith
m
,
the
two
de
gr
ees
of
f
ree
dom
a
lgo
rithm
(2DOF
)
reduce
d
the
c
ontr
ol
input
de
m
and
s,
w
hich
was
im
po
rtant
from
the
pr
act
ic
al
po
i
nt
of
view.
B
ut
the
2DOF
ha
d
slow
e
r ou
t
pu
t
r
esp
on
se
co
m
par
ed
to
t
he 1DO
F.
A
r
obus
t
op
ti
m
a
l
adap
ti
ve
con
t
ro
l
strat
e
gy
was
de
velo
pe
d
in
[
6]
to
de
a
l
with
tracki
ng
prob
le
m
of
a
qu
a
droto
r
un
m
ann
ed
ae
rial
veh
ic
le
(UA
V)
.
T
he
c
on
tr
oller
has
a
pr
om
inent
abili
ty
to
sta
bili
ze
nonlinear
dynam
ic
sys
tem
of
qu
a
droto
r,
f
or
ce
the
sta
te
s
to
fo
ll
ow
de
sired
re
fe
re
nc
e
sign
al
s,
an
d
fin
d
op
ti
m
al
s
olu
ti
on
for
the
trac
king
pro
blem
wit
hout
c
on
t
ro
l
i
nput
sat
ur
at
i
on.
The
pe
rfo
rm
a
nce
a
naly
sis
of
a
c
onve
ntio
na
l
PID
con
t
ro
ll
er
a
nd
a
MR
AC
w
a
s
done
in
[7
]
.
Cy
li
nd
rical
ta
nk
interact
in
g
a
nd
no
nin
te
racti
ng
syst
em
s
we
re
sel
ect
ed
as
pro
cesses
to
be
c
ontr
olled.
T
he
r
esults
showe
d
that
the
MR
A
C
has
bette
r
overs
hoot,
set
tl
ing
ti
m
e
and set
-
po
i
nt tr
ackin
g per
form
ance tha
n
t
he
c
onve
ntion
al
PID c
on
t
ro
ll
er.
The
de
velo
pme
nt
of
direct
and
in
direct
ada
ptive
co
ntr
ol
m
et
ho
ds
to
c
ontr
ol
the
powe
r
in
a
TRI
G
A
MARK
II
reac
tor
was
pro
pos
ed
in
[
8].
T
he
analy
sis
showe
d
that
t
he
a
dapt
ive
al
gorithm
offer
s
over
al
l
bette
r
resu
lt
s
t
han
th
e
fee
db
ac
k
co
ntr
ol
al
gorithm
.
T
he
a
da
ptive
al
gorithm
reduced
the
set
tl
ing
ti
m
e
up
to
25%
of
the nom
inal set
tl
ing
ti
m
e.
A
m
od
el
re
fer
e
nce
a
dap
ti
ve
c
on
t
ro
ll
er w
it
ho
ut
integ
ral
(M
RAC
WI
)
para
m
et
er
for
po
sit
ion
c
ontr
ol
of
a
DC
Moto
r
was
de
sig
ned
in
[
9].
The
c
ontr
oller
pro
du
ced
bette
r
pe
r
form
ance
in
te
rm
s
of
set
tl
ing
tim
e,
per
ce
ntage
ov
e
rsho
ot
and
m
e
an
square
e
rror
as
com
par
ed
wit
h
PID
co
ntr
oller,
sta
ndar
d
MR
AC
and
M
RAC
with
a
sigm
a
m
od
ific
at
ion
.
A
draw
bac
k
of
this
al
gorithm
is
that
it
s
per
f
or
m
ance
to
va
r
ia
ti
on
s
of
pa
ra
m
et
ers
and exte
rn
al
di
sturbance
s is
unkn
own.
A
c
om
par
iso
n
of
t
he
ti
m
e
s
pecifica
ti
on
pe
rfor
m
ance
bet
ween
a
c
onve
nti
on
al
PID
c
on
t
ro
ll
er
an
d
a
m
od
er
n
sli
di
ng
m
od
e
co
nt
ro
ll
er
(
SMC
)
for
a
no
nline
ar
in
ver
te
d
pe
ndulu
m
syst
e
m
was
done
i
n
[
10]
.
The
pe
rfor
m
ances
of
both
con
t
ro
l
strat
eg
ie
s
wer
e
asse
ssed
to
see
wh
ic
h
one
ha
d
bette
r
ha
ndli
ng
of
pend
ulu
m
’s
an
gle
a
nd
cart’
s
posit
i
on.
T
he
overall
res
ults
of
t
he
a
naly
sis
sho
wed
th
at
the
sli
di
ng
m
od
e
con
t
ro
ll
er
ha
d
faster
risin
g
ti
m
e,
bette
r
set
tl
ing
ti
m
e
and
a
m
uch
bette
r
pe
rcen
ta
ge
of
overs
hoot
c
om
par
ed
t
o
the
conve
ntional
PI
D.
Both
c
on
t
ro
ll
ers
di
d
no
t
ha
ve
any
s
te
ady
sta
te
err
or
s
.
Since
the
inv
e
rted
pe
ndul
um
is
a
highly
nonl
inear
syst
em
,
this
researc
h
sho
wed
t
w
o
draw
bac
ks
.
The
a
uthors
did
no
t
in
ve
sti
gate
the
perform
ance
of
the
co
nt
ro
ll
ers
wh
e
n
e
xter
nal
disturb
ances
an
d
va
ri
at
ion
s
of
pa
ra
m
et
ers
occu
r
.
These
stud
ie
s
w
ou
l
d hav
e
m
ade
the
inv
est
igati
on
m
or
e reali
sti
c.
Adva
nced
P
ID
are
al
so
us
ed
in
the
m
edical
sect
or
,
[
11
]
propose
d
a
fr
act
ion
al
orde
r
PID
con
tr
olle
r
and
a
n
intege
r
order
P
ID
c
on
t
ro
ll
er for
s
upre
ssing
e
pilepti
c act
ivit
ie
s.
Both co
ntr
ollers show
e
d
great
r
es
ults to
sta
bili
ze
the
patie
nt,
bu
t
the
fr
act
io
nal
ord
er
PID
co
ntr
ol
le
r
is
m
or
e
su
i
ta
ble
for
i
m
ple
m
entat
ion
in
FPGA
because
it
us
e
s
le
ss
flip
-
fl
ops.
U
nfor
t
un
at
el
y,
the
stu
dy
did
not
ta
ke
in
c
on
si
der
at
i
on
sud
den
ab
norm
al
act
ivit
ie
s
of
t
he
brai
n
cel
ls
t
o
e
valuate
t
he
tim
e
res
pons
e
ta
ken
by
the
con
t
ro
ll
er
to
s
ta
bili
ze
the
pa
ti
ent.
This st
ud
y i
s c
r
ucial
to bri
ng the
patie
nt
back to a
good
heal
th con
diti
on in t
he
s
hortest
ti
m
e p
os
si
ble.
In
[
12
]
,
a
nove
l
data
-
dri
ve
n
si
gm
oid
-
base
d
P
I
co
ntr
oller
wa
s
desi
gn
e
d
to
track
t
he
a
ngul
ar
ve
l
ocity
of
dc
m
oto
r
power
e
d
by
a
dc/
dc
bu
c
k
co
nve
rter.
The
res
ult
s
of
t
he
in
vesti
gations
s
howe
d
that
the
data
-
dr
i
ven
sigm
oid
-
base
d
PI,
wh
ic
h
is
tun
e
d
us
i
ng
glo
bal
sim
ultaneou
s
per
t
urbati
on
stoc
hastic
appr
ox
im
at
ion
,
yi
el
ds
a
bette
r
angula
r
velocit
y
trackin
g
as
com
par
ed
to
c
onve
ntion
al
P
I
an
d
PI
-
Fuzz
y.
A
dra
wb
ac
k
of
this
stud
y
is
that, the
pe
rform
ance of th
e
si
gm
oid
-
base
d
c
on
t
ro
ll
er
was
not eval
uated
fo
r dist
urba
nce rejecti
on.
In
[13],
the
perform
ance
of
th
e
fr
act
io
nal
order
pro
portiona
l
-
integr
al
-
de
ri
vative
(FOP
I
D
)
co
ntr
ollers
desig
ne
d
by
usi
ng
arti
fici
al
be
e
colon
y
(
AB
C)
for
fr
act
io
na
l
or
de
rs
syst
em
s
is
co
m
par
ed
to
co
nv
e
ntio
nal
PI
D
con
t
ro
ll
er
op
ti
m
iz
ed
by
the
ABC
colo
ny
al
gorithm
.
The
resu
lt
s
of
the
s
i
m
ulati
on
s
sho
wed
t
hat
the
F
OP
I
D
con
t
ro
ll
ers had
sign
ific
a
ntly
b
et
te
r
per
f
orm
a
nce co
m
par
ed
t
o
the con
ven
ti
on
al
P
ID
c
on
t
r
ollers. Un
fortu
natel
y,
there
was n
o
st
ud
y m
ade to
ev
al
uate the
perf
or
m
ance of the
contr
oller whe
n dist
urba
nces
occur.
An
a
da
ptive
sa
fe
ex
per
im
entat
ion
dy
nam
ic
s
(A
S
ED
)
f
or
da
ta
dr
ive
n
neur
oend
ocr
i
ne
-
P
I
D
co
ntr
ol
of
MIM
O
Syst
em
s
was
desig
ned
in
[
14]
.
The
pe
rfor
m
ance
of
the
A
SED
ba
sed
m
et
hod
was
co
m
par
ed
to
the
sta
nda
rd
sa
fe
e
xp
e
rim
enta
ti
on
dynam
ic
s
(S
E
D)
an
d
sim
ultaneo
us
pe
rt
urbati
on
st
ochast
ic
appr
ox
im
at
ion
(S
PS
A
)
based
m
et
ho
ds.
T
he
r
esults
of
t
he
si
m
ula
ti
on
s
s
howed
that
t
he
A
SED
an
d
SED
base
d
m
et
ho
ds
ha
ve
su
ccess
fu
ll
y
sol
ved
the
unsta
bl
e
con
ve
r
gen
ce
issue
in
the
existi
ng
neur
oendo
c
rine
-
P
ID
ba
sed
sta
ndar
d
SP
S
A
.
More
ov
e
r,
the
pr
ese
nted
A
S
ED
base
d
al
gorithm
ou
tper
f
or
m
s
the
SED
and
the
SPS
A
base
d
m
e
th
od
s
i
n
the p
er
sp
ect
ive
o
f
the c
ontrol
perform
ance accur
acy
in
te
r
m
s
o
f
lowe
r object
ive fu
nctio
n,
total
nor
m
error
a
nd
total
no
rm
inp
ut.
A
draw
bac
k
of
this
resea
r
ch
is
that,
the
auth
or
s
did
not
per
f
or
m
plant’
s
par
am
et
ers
c
hanges
to
see
t
he
i
nf
lu
ence
of
the
ad
aptive
gain
of
the
A
SE
D
c
ontrolle
r.
The
re
s
earch
ga
p
an
d
m
erit
of
t
he
a
da
ptiv
e
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
11
, No
.
1,
Febr
uar
y
2021
:
300
-
318
302
PI
D
c
ontrolle
r
dev
el
ope
d
in
this
pap
e
r
co
m
par
ed
to
the
oth
er
a
dv
a
nc
ed
PID
co
ntr
ol
le
rs
rev
ie
we
d
is
that,
the
pro
posed
con
t
ro
ll
er
ca
n
sta
bili
ze
the
cl
os
ed
-
lo
op
sy
stem
wh
en
va
r
ia
ti
on
s
of
para
m
et
ers
and
s
udde
n
rand
om
d
ist
ur
ba
nces
occ
ur
sim
ul
ta
neo
usl
y.
None o
f
the
r
e
viewe
d pap
e
rs exp
l
or
e
d
t
his s
cenari
o.
The
m
ai
n
co
ntributi
on
of
this
pa
per
is
t
hat
it
pr
e
sents
t
he
de
sign
of
a
m
odel
ref
e
ren
ce
ad
aptive
P
I
D
con
t
ro
ll
er
(MRAPIDC)
base
d
o
n
t
he
MIT
a
ppr
oac
h
to
sta
bi
li
ze
and
opti
m
i
ze
the
li
near
iz
e
d
m
od
el
of
the
two
-
ta
nk
li
qu
i
d
le
ve
l
pr
oces
s
aff
e
ct
ed
by
su
dde
n
changes
of
pa
ram
et
ers
and
input
distu
rb
a
nc
e.
The
n,
a
refe
ren
ce
m
od
el
is
de
ve
lop
e
d
base
d
on
c
on
t
ro
l
t
he
ory
,
a
nd
a
PI
regulat
or
is
de
sig
ne
d
on
M
ATL
AB
to
co
ntr
ol
the
re
fer
e
nce
m
od
el
.
The
de
sign
e
d
P
I
c
on
t
ro
ll
er
is
a
no
ve
l
idea
to
a
dd
sta
bili
ty
to
the
outp
ut
sig
na
ls
of
the
ref
e
ren
ce
m
od
el
,
hen
ce
m
aking
the
a
da
ptive
al
gorith
m
m
or
e
ro
bust
.
Anothe
r
no
ve
lt
y
of
this
research
is
the
incl
us
io
n
of
th
ree
ne
w
a
da
ptive
gains
in
the
final
st
ru
c
ture
of
the
MR
AP
I
DC
t
o
m
ake
th
e
cl
os
e
d
-
loop
syst
e
m
ro
bust
wh
e
n
la
r
ge
va
riat
ion
s
of
pa
r
a
m
et
ers
and
s
udde
n
e
xtern
al
disturba
nces
occur
sim
ultaneousl
y.
The
co
ntr
oller
keeps
the
per
c
entage
of
over
sh
oot
of
the
cl
os
e
d
-
l
oo
p
syst
e
m
below
10%
and
it
s
recove
ry
tim
e
to
s
udde
n
var
i
at
ion
s
of
pa
ra
m
et
ers
is
lowe
r
tha
n
5
s
eco
nds,
w
hich
is
a
huge
a
dvanta
ge
com
par
ed
to
the
ot
he
r
adap
ti
ve
co
ntr
ollers
re
viewe
d.
Anothe
r
a
dvanta
ge
of
thi
s
ada
ptive
al
gorithm
is
that
real
-
ti
m
e
tun
ing
of
the
three
ada
pt
ive
gai
ns
ca
n
be
done
to
im
pro
ve
the
perf
or
m
ance
of
th
e
cl
os
ed
-
lo
op
syst
e
m
.
Fu
rthe
rm
or
e,
the
perf
or
m
ance
of
the
MR
AP
I
DC
is
co
m
par
ed
to
a
cl
assic
PI
co
nt
ro
ll
er
de
sig
ne
d
on
MAT
LA
B
for
the
li
near
iz
e
d
m
od
el
of
the
two
-
ta
nk
syst
em
.
The
adap
ti
ve
c
on
tr
ol
al
gorithm
s,
PI
c
ontr
ol
al
gorith
m
and
the m
od
el
s of t
he
cl
ose
d
-
lo
op
syst
e
m
s ar
e sim
ula
te
d
in M
A
TLAB/Si
m
ulink
.
The
outl
ine
of
this
pap
e
r
is
as
fo
ll
ows:
the
m
od
el
li
ng
and
si
m
ulati
on
of
the
process
is
pro
po
se
d
i
n
sect
ion
2.
Sect
ion
3
disc
us
se
s
the
de
sig
n
of
a
MR
AP
ID
C
base
d
on
m
od
el
ref
ere
nce
th
eor
y.
T
he
sim
ulati
on
resu
lt
s a
re sh
own
in
s
ect
ion 4
. S
ect
io
n 5 dra
ws
th
e c
on
cl
usi
on
.
2.
MO
DELIN
G
AND SIM
UL
ATIO
N
O
F T
HE TWO
-
TA
NK SY
STE
M
In
this
pa
per,
a
two
-
ta
nk
li
quid
le
vel
syst
em
is
sel
ect
ed
as
a
plant
to
be
c
on
tr
olled,
be
cause
it
is
a
nonlinear
in
her
e
ntly
un
sta
ble
syst
e
m
.
The
syst
e
m
is
m
ade
of
tw
o
-
ta
nk
m
ou
nted
ab
ove
a
reservo
ir
,
wh
ic
h
has
the
f
un
ct
i
on
of
a
stora
ge
el
e
m
ent
fo
r
li
quid.
T
he
syst
em
has
an
ind
e
pende
nt
pu
m
p
to
pu
m
p
li
qu
id
fr
om
the
reserv
oir
to
the
ta
nk
s.
T
he
two
-
ta
nk
s
a
re
interact
ing,
w
hi
ch
m
eans
that
the
li
qu
id
m
ov
es
fr
om
on
e
ta
nk
to
the o
the
r.
Whe
n
tw
o
ta
nks ar
e
stat
e d
epe
nd
e
nt,
the inte
racti
on
of
li
qu
i
d
be
tween th
e ta
nks ex
hi
bits a no
nlinea
r
beh
a
viou
r
[
15]
. T
he
sim
plified
blo
c
k diag
ra
m
o
f
the
pro
ce
ss is s
how
n
in
Figure
1.
Figure
1. Bl
oc
k diag
ram
o
f
a
two
-
ta
nk li
quid
level p
r
ocess
The param
et
ers
of the
tw
o
-
ta
nk li
qu
i
d
le
vel
syst
e
m
are
the
fo
ll
owin
g:
ℎ
1
= le
vel of li
qui
d
in
tan
k 1 in
ℎ
2
= le
vel of li
qui
d
in
tan
k 2 in
1
=
cr
os
s secti
on
al
area
of
ta
nk
1
in
2
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
8
8
-
8708
Desig
n of
a model ref
erenc
e
adapti
ve PI
D
c
ontrol
…
(
Y
ohan
Darcy Mf
oum
boulou
)
303
2
= cr
os
s secti
on
al
area
of
ta
nk
2
in
2
1
= cr
os
s secti
on
al
area
of
t
he o
utlet
p
ipe
in
ta
nk 1 in
2
2
= cr
os
s secti
on
al
area
of
t
he o
utlet
p
ipe
in
ta
nk 2 in
2
= flow
r
at
e of l
iqu
id
into
tan
k 1
2
/
= flow
rate o
f
l
iqu
id
out
of tan
k 2
2
/
1
= v
al
ve
r
at
io
of
ou
tl
et
p
i
pe of
tank 1
2
= v
al
ve
r
at
io
of the
outl
et
p
ip
e of tan
k 2
g
= gravit
at
ion
al
force
k
=
pu
m
p
gai
n
u
(
t)
=
input
vo
lt
a
ge
to
the
pum
p
The
no
nlinear
equ
at
io
ns
of
the
two
-
ta
nk
li
qu
id
syst
em
m
od
el
can
be
de
rive
d
by
ap
plyi
ng
the Ber
noulli
’s
law
of co
ns
e
rvat
ion
of m
ass [
15
]
:
(1)
The n
on
li
nea
r dynam
ic
eq
uations d
erive
d fro
m
ta
nk
1
a
re:
(2)
(in
put flo
w
)
(3
)
The
dynam
ic
eq
uatio
ns de
rive
d from
tank
2 a
re:
(4
)
(5)
At
eq
uili
br
ium
for
a
c
on
ti
nu
ous
li
quid
le
vel
set
-
point,
t
he
de
rivati
ve
of
t
he
li
qu
id
le
vels
in
the
ta
nks
m
us
t
be
zero
(
).
In
the
sce
na
rio
w
he
n:
;
t
he
syst
e
m
is
s
ta
te
deco
uple
d.
Ther
e
f
or
e,
to
sat
isfy
the
c
onditi
ons
of
t
he
sim
ulatio
n
of
the
li
quid
le
ve
l
syst
e
m
:
,
t
he
le
vel
of
li
qui
d
in
ta
nk
1
m
us
t be b
ig
ger than
this
of ta
nk 2.
The
sta
te
s
pace
r
e
pr
ese
ntati
on of the
no
nline
ar s
yst
em
is the foll
ow
i
ng
:
(6
)
(7)
The
n:
)
(
0
)
(
2
)
(
2
)
(
2
)
(
)
(
1
2
2
2
2
1
2
1
1
1
1
1
1
2
1
t
u
A
k
t
gh
A
a
t
h
g
A
a
t
h
g
A
a
dt
t
dh
dt
t
dh
(8
)
o
u
t
in
Q
Q
dt
dh
A
)
(
2
)
(
1
1
1
1
1
t
gh
a
Q
dt
t
dh
A
in
in
Q
t
u
)
(
)
(
2
)
(
[
1
)
(
1
1
1
1
1
t
gh
a
t
ku
A
dt
t
dh
)
(
2
)
(
2
)
(
2
2
2
1
1
1
2
2
t
gh
a
t
gh
a
dt
t
dh
A
)
(
2
)
(
2
)
(
2
2
2
2
1
2
1
1
2
t
gh
A
a
t
gh
A
a
dt
t
dh
0
2
.
1
.
h
h
2
1
h
h
2
1
h
h
)
(
2
)
(
2
1
)
(
)
(
1
)
(
)
(
2
2
2
1
1
1
2
1
1
1
1
2
1
t
gh
a
t
gh
a
A
t
h
a
t
ku
A
dt
t
dh
dt
t
dh
T
T
h
h
h
h
C
y
2
1
2
1
1
0
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
11
, No
.
1,
Febr
uar
y
2021
:
300
-
318
304
The param
et
ers
of the
tw
o
-
ta
nk pr
ocess
a
re
giv
e
n
in
Ta
ble
1.
Table
1.
Value
s of the
p
a
ram
e
te
rs
of t
he pr
oc
ess
Para
m
eters
Valu
es
75
1
.5
1
.53
1
5
1
9
5
0
.68
2
0
0
4
3
981
3
0
.00
2
4
The
ne
xt
sect
ion
is
to
der
i
ve
a
li
near
iz
ed
m
od
el
of
the
nonl
inear
ta
nk
pro
cess
to
find
a
n
accurate
eq
uiva
le
nce
of both
m
od
el
s.
2.1.
Li
neariz
ati
on o
f the
nonl
inear m
od
el
of the
two
-
t
ank pr
ocess
The
li
near
iz
at
ion
of
the
tw
o
-
ta
nk
li
quid
syst
e
m
is
per
f
orm
ed
around
it
s
op
e
rati
ng
po
ints,
an
d
to
achieve
that,
only
the
l
inear
te
rm
s
of
the
T
ay
lor
series
ex
pansi
on
of
t
he
nonlinear
m
od
el
are
co
ns
id
ered.
In
(
7) an
d (8) a
re c
on
si
der
e
d
t
o
li
nea
rize t
he nonli
nea
r
m
odel
o
f
the syste
m
.
Let
d
efi
ne
t
he st
at
e v
aria
bles
of the syst
em
:
ℎ
1
=
1
=
le
vel of li
quid
in tan
k 1.
ℎ
2
=
2
=
le
vel of li
quid
in tan
k 2.
By
su
bs
ti
tuti
ng
ℎ
1
an
d
ℎ
2
, by
1
and
2
in (7) an
d (
8)
,
it
is obtai
ned
:
[
.
1
.
2
]
=
[
−
1
1
1
√
2
1
1
1
2
√
2
1
−
2
2
2
√
2
2
]
+
[
1
0
]
(9
)
=
[
1
2
]
=
[
0
1
]
[
1
2
]
(10)
The (9
)
a
nd (1
0) can
b
e
expre
ssed by the
sta
nd
a
r
d nonli
nea
r
m
od
el
:
.
=
(
)
+
(
)
=
[
1
(
)
2
(
)
]
+
[
1
(
)
2
(
)
]
(11)
=
(
)
(12)
w
he
re:
(
)
,
(
)
and
(
)
are the
no
nlinea
r vect
or fu
nctions o
f
t
he
sta
te
v
ect
or.
The
li
near
iz
at
ion
of
the
nonlinea
r
m
od
e
l
is
per
f
or
m
ed
acco
rd
i
ng
to
Tay
lor
se
ri
es
m
et
ho
d.
The
li
near
iz
e
d
m
od
el
is
d
e
riv
ed
base
d
of
t
he
no
nlinear
f
unct
ions
1
,
2
,
1
an
d
2
.
F
or
t
he
case
of
t
he
t
wo
-
ta
nk
syst
e
m
, th
e non
li
nea
r fun
ct
ion
s a
re:
1
=
−
1
1
1
√
2
1
,
1
=
1
(13)
2
=
1
1
2
√
2
1
−
2
2
2
√
2
2
,
2
=
0
(14)
The deri
vative
s of the
f
ir
st f
unct
ion
1
accor
di
ng to
t
he
tw
o s
ta
te
s
1
,
2
and the
con
t
ro
l i
nput
:
1
(
1
)
=
1
[
−
1
1
1
√
2
1
]
T
T
h
h
h
h
C
y
2
1
2
1
1
0
)
(
,
2
2
1
cm
A
A
)
(
,
2
1
2
cm
a
a
1
2
)
s
ec
(
2
cm
g
k
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
8
8
-
8708
Desig
n of
a model ref
erenc
e
adapti
ve PI
D
c
ontrol
…
(
Y
ohan
Darcy Mf
oum
boulou
)
305
1
(
1
)
=
−
1
1
1
√
2
1
(15)
The
n:
2
(
1
)
=
2
[
−
1
1
1
√
2
1
]
2
(
1
)
=
0
(16)
The deri
vative
accor
ding t
o
is:
(
1
)
=
0
(17)
Der
i
vatives
of
the sec
ond fun
ct
ion
2
acco
rd
i
ng to
the t
wo sta
te
s ar
e:
1
(
2
)
=
1
[
1
1
2
√
2
1
]
1
(
2
)
=
1
1
2
√
2
1
(18)
2
(
2
)
=
2
[
1
1
2
√
2
1
−
2
2
2
√
2
2
]
2
(
2
)
=
−
2
2
2
√
2
2
(19)
The
te
rm
s o
f
th
e co
ntr
ol m
at
ri
x
B ca
n
al
s
o be
foun
d
wit
h
the
sam
e p
ro
ce
dur
e:
Der
i
vative
of
1
per the t
wo stat
es:
1
(
1
)
=
1
[
1
]
1
(
1
)
=
0
(20)
Fo
r
2
:
2
(
1
)
=
2
[
1
]
2
(
1
)
=
0
(21)
Der
i
vative
of
2
accor
ding t
o
th
e two st
at
es:
1
(
2
)
=
0
(22)
2
(
2
)
=
0
(23)
Af
te
r
al
l
the
li
near
iz
e
d
ex
pre
ssion
s
of
the
s
yst
e
m
are
done
,
the
li
nea
rized
sta
te
sp
ace
r
epr
ese
ntati
on
of the t
wo
-
ta
nk
pro
ces
s is:
[
.
1
.
2
]
=
[
−
1
1
1
√
2
1
0
1
1
2
√
2
1
(
−
2
2
2
√
2
2
)
]
[
1
2
]
+
[
1
0
]
(24)
=
[
0
1
]
[
1
2
]
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
11
, No
.
1,
Febr
uar
y
2021
:
300
-
318
306
w
he
re
the
coe
f
fici
ents
of
the
m
at
rices
def
in
ed
in
(24
)
are
cal
culat
ed
for
the
equ
il
ib
riu
m
values
of
th
e
sta
te
var
ia
bles
giv
e
n
by
(
6)
an
d
(
7).
At
eq
uili
bri
um
fo
r
co
ntin
uous
li
quid
le
vel
set
-
point,
t
he
de
rivati
ve
m
us
t
be
zero
(
ℎ
1
̇
=
ℎ
2
̇
=
0
).
In
t
he
scena
rio
w
he
n:
ℎ
1
=
ℎ
2
,
the
syst
e
m
is
sta
te
deco
uple
d.
T
her
e
f
or
e
,
to
sat
isfy
the
co
nd
it
io
ns
of
the
sim
ulati
on
of
t
he
li
qu
i
d
le
vel
syst
em
:
ℎ
1
>
ℎ
2
.
The
m
od
el
de
fine
d
by
(
6)
a
nd
(7)
will
be
us
e
d
to
d
e
velo
p
the
ad
a
ptive
con
t
ro
l al
gorithm
.
3.
DESIG
N OF
A MO
DEL
-
R
EFE
REN
CE
ADAPTI
VE P
ID
-
CONTR
O
LL
ER
The
Ma
ssach
use
tt
s
In
sti
tute
of
Tech
no
l
og
y
(
MIT)
r
ule
is
a
gr
a
dient
r
ule.
I
t
was
der
ive
d
at
MIT
in
it
s
instru
m
entat
ion
la
borato
ry,
hen
ce
it
s
nam
e.
The
MIT
r
ule
is
the
or
iginal
ap
proac
h
to
m
od
el
re
fer
e
nce
adap
ti
ve
c
on
t
r
ol
(MR
AC)
[
15,
16]
.
T
o
give
a
represe
ntati
on
of
the
M
IT
ru
le
,
le
t
co
ns
i
de
r
as
a
n
ad
ju
sta
ble
par
am
et
er
of
a
co
ntr
oller.
T
he
desire
d
cl
ose
d
-
l
oop
res
po
ns
e
of
the
ou
t
pu
t
is
;
and
t
he
e
rror
betw
een
the
outp
ut
of
the
cl
os
e
d
-
l
oop
syst
em
an
d
the
outp
ut
of
t
he
re
frer
ence
m
od
el
is
ε.
I
n
the
pa
pe
r,
the
desire
d
out
pu
t
is
pro
pos
ed
to
be
deter
m
ined
by
a
re
fer
e
nce
m
od
el
ou
t
pu
t.
T
o
de
fine
the
MIT
r
ule,
le
t con
si
der
t
he
foll
ow
i
ng loss
fun
ct
io
n [17]:
(
)
=
1
2
2
(25)
It
is
necessary
to
determ
ine
at
ever
y
m
o
m
ent
of
tim
e
the
par
am
et
ers
of
the
con
tr
oller
in
su
c
h
a
way
that
the
functi
on
(
)
is m
i
nim
iz
ed.
The
MIT
f
undam
ental
app
r
oa
ch
co
ns
ist
s
of
adjustin
g
the
par
am
et
ers
of
the
cl
os
e
d
-
l
oop
syst
e
m
su
ch
that
the
los
s
f
un
ct
io
n
descr
i
bed
in
(25
)
is
m
ini
m
iz
ed.
To
m
ini
m
iz
e
the
functi
on
(
)
,
a
re
al
ist
ic
appro
ac
h
wou
ld
b
e
to
c
ha
ng
e
the
pa
ram
et
ers
of t
he
sys
tem
in
the
dire
ct
ion
of the
ne
gative
gr
a
dient
of
:
=
−
=
−
(
)
(26)
w
he
re:
is
an
adap
ta
ti
on
ga
in;
and
(
)
is
the
sen
sit
ivit
y
der
ivati
ve
f
unct
ion
of
the
syst
e
m
towar
ds
it
s
ti
m
e
-
var
yi
ng
par
am
et
ers.
re
pr
es
e
nts in
this case
,
the tim
e v
aryi
ng p
a
ram
et
ers
of the c
ontrolle
r.
In
(
26)
is
the
MIT
ru
le
.
The
sensiti
vity
der
ivati
ve
ex
pr
e
ss
es
how
the
a
dju
sta
ble
pa
ram
e
te
rs
influ
e
nc
e
the
erro
r.
I
n
gen
e
ral,
it
is
assum
ed
that
the
p
a
ram
et
er
changes
a
re
slow
e
r
tha
n
the
ot
her
var
ia
bles
of
the
syst
e
m
.
Hen
ce,
the
se
ns
it
ivit
y
der
ivati
ve
(
)
can
be
e
valuat
ed
by
ass
um
ing
that
the
a
dju
s
ta
ble
par
am
et
er
is
c
on
sta
nt.
T
o
s
umm
arize,
the
f
ollo
wing
ste
ps
ca
n
be
use
d
t
o
desig
n
an
a
da
pt
ive
c
ontr
oller
based
on
the MIT
rule:
Def
i
ne
the
co
e
ff
ic
ie
nts
of the
trans
fer
f
un
ct
io
n of a
plant
with
unknow
n par
a
m
et
ers.
Choose a
r
e
fere
nce m
od
el
.
Choose a
contr
ol alg
or
it
hm
to
achieve
per
fec
t
m
od
el
trac
king.
Def
i
ne
the
er
ror of
t
he
cl
ose
d
-
loop syst
em
.
Der
i
ve
the
exp
ressio
ns
of the
con
t
ro
l
par
am
et
ers.
Apply t
he ne
ga
ti
ve
gradie
nt of
to f
i
nd the
up
dating
pa
ram
eter
s.
3.1.
De
termina
tio
n of the
tr
an
sf
er fu
nctio
n c
oe
ff
ic
ie
nt
s
for
t
he li
neariz
ed
mod
el
The
sta
te
s
pace
r
e
pr
ese
ntati
on of the
tan
k pro
cess is re
prese
nted
a
s
giv
e
n
i
n (24)
:
[
.
1
.
2
]
=
[
−
1
1
1
√
2
1
0
1
1
2
√
2
1
(
−
2
2
2
√
2
2
)
]
[
1
2
]
+
[
1
0
]
=
[
0
1
]
[
1
2
]
To
m
ake m
at
h
e
m
at
ic
al
calc
ul
at
ion
s m
uch
si
m
pler,
(
24)
is re
wr
it
te
n
as:
[
̇
1
̇
2
]
=
[
11
0
21
22
]
[
1
2
]
+
[
1
0
]
=
[
0
1
]
[
1
2
]
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
8
8
-
8708
Desig
n of
a model ref
erenc
e
adapti
ve PI
D
c
ontrol
…
(
Y
ohan
Darcy Mf
oum
boulou
)
307
w
he
re:
11
=
−
1
1
1
√
2
1
;
21
=
1
1
2
√
2
1
;
22
=
−
2
2
2
√
2
2
The
m
at
rices o
f
th
e
stat
e sp
ac
e m
od
el
can be
r
e
pr
ese
nted
as
:
=
[
11
0
21
22
]
;
=
[
1
0
]
;
=
[
0
1
]
; and
=
0
.
To
tra
nsfo
rm
the
sta
te
s
pac
e
m
od
el
of
th
e
li
near
iz
ed
s
yst
e
m
to
trans
fer
f
un
ct
io
n,
the
fo
ll
owin
g
form
ula is app
l
ie
d
[
18]
:
(
)
=
(
)
(
)
=
(
−
)
−
1
+
w
he
re
I
is t
he
i
den
ti
ty
m
at
rix,
an
d:
=
[
0
0
]
The
cal
c
ulati
on
of the tra
nsfe
r
f
unct
ion o
f
t
he
li
near
iz
ed
m
od
el
is
the
fo
ll
ow
i
ng
:
(
)
=
[
0
1
]
×
(
[
0
0
]
−
[
11
0
21
22
]
)
−
1
×
[
1
0
]
(
)
=
[
0
1
]
×
(
[
(
−
11
)
0
−
21
(
−
22
)
]
)
−
1
×
[
1
0
]
The
n:
(
)
=
[
0
1
]
×
[
[
(
−
11
)
0
−
21
(
−
22
]
|
(
−
11
)
0
−
21
(
−
22
)
|
]
×
[
1
0
]
(
)
=
[
0
1
]
×
[
[
(
−
22
)
0
21
(
−
11
)
]
[
2
+
(
−
11
−
22
)
+
11
22
]
]
×
[
1
0
]
(
)
=
[
0
1
]
×
[
[
(
−
22
)
×
(
1
)
21
(
1
)
]
[
2
+
(
−
11
−
22
)
+
11
22
]
]
(
)
=
[
(
)
×
(
−
22
)
×
(
1
)
+
(
1
)
×
21
(
1
)
[
2
+
(
−
11
−
22
)
+
11
22
]
]
The final
repre
sentat
ion o
f
t
he
tran
s
fer f
unct
ion
of the
li
nea
rized m
od
el
is:
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
11
, No
.
1,
Febr
uar
y
2021
:
300
-
318
308
(
)
=
21
(
1
)
2
+
(
−
11
−
22
)
+
11
22
The
desi
gn
of
the
m
od
e
l
refe
ren
ce
a
dap
ti
ve
PI
D
co
ntr
oller
to
sta
bili
ze
the
dev
el
op
e
d
li
near
iz
ed
m
od
el
is
descr
i
bed in t
he
n
e
xt secti
on.
3.2.
P
roced
ure
to
design
t
he
mo
del
-
referen
ce
adapt
i
ve
PID
-
cont
rolle
r
(MRA
P
ID
C)
f
or
th
e
li
neariz
e
d
mod
el
of the t
w
o
-
t
ank li
qui
d level s
ystem
To
m
ak
e
the
der
i
vation
of
the
ada
ptive
P
ID
c
ontrolle
r
m
uch
si
m
pler
m
at
he
m
at
ic
a
lly,
the
final
expressi
on of t
he
tra
nsfer
f
unct
ion
of the li
ne
arized m
od
el
is sim
plifie
d
as
:
(
)
=
3
0
2
+
1
+
2
w
he
re:
(
)
=
(
)
;
0
=
1
;
1
=
(
−
11
−
22
)
;
2
=
11
22
3
=
21
(
1
)
The param
et
ers
of the
n
e
w
t
r
ansf
e
r funct
io
n are
def
i
ned as:
1
=
1
1
,
2
=
2
2
and
3
=
3
3
.
1
,
2
and
3
are the
vary
ing pa
ram
eter
s; an
d
1
,
2
and
3
are t
he fixe
d param
et
ers
of t
he pr
ocess
.
3.2.1.
Desi
gn
of the desi
re
d linear re
ferenc
e mo
del
The
li
near
iz
e
d
m
od
el
of
the
t
ank
syst
em
is
of
a
sec
ond
or
der.
The
refo
re,
the
li
near
re
fe
ren
ce
m
od
el
can
be desig
ne
d
as
a ty
pical
s
econd
orde
r
tra
ns
fe
r funct
io
n as f
ollows:
Fr
om
the
desig
n
s
pecifica
ti
on
s,
the
v
al
ues of t
he do
m
inant
po
le
s
can
b
e
obtai
ned as
fo
ll
ow
s:
(
)
=
2
2
+
2
+
2
=
(
)
(
)
(27)
w
he
re:
(
)
is
the
t
ran
s
fer
f
unct
io
n
of
the
re
fer
e
nce
m
od
el
,
is
the outp
ut o
f
th
e
re
fer
e
nce
m
od
el
a
nd
is
it
s input.
The
n
the
outp
ut
o
f
the
ref
e
re
nc
e m
od
el
is:
(
)
=
2
2
+
2
+
2
(
)
(28)
To
gua
ran
te
e
t
he
sta
bili
ty
of
the
cl
os
e
d
-
lo
op
syst
em
,
the
ref
ere
nce
m
od
el
m
us
t
m
ee
t
the
fo
ll
owin
g
desig
n
c
har
act
erist
ic
s:
Perce
ntage of
Ov
e
rs
hoot
(P
O
): 8
%
Sett
li
ng
ti
m
e
: 2
sec
onds
T
im
e d
el
ay
: 0
seco
nd
Stea
dy stat
e er
ror: 0
Fr
om
the
desig
n
s
pecifica
ti
on
s,
the
v
al
ues of t
he do
m
inant
po
le
s
can
b
e
obtai
ned as
fo
ll
ow
s
[1
9,
20]
:
The dam
pin
g r
at
io for
a
pe
rce
ntage o
f ov
e
rs
hoot
(P
O
) o
f 8
% is:
=
√
(
10
0
%
)
2
2
+
(
10
0
%
)
2
=
√
(
0
.
08
)
2
2
+
(
0
.
08
)
2
=
0
.
63
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
8
8
-
8708
Desig
n of
a model ref
erenc
e
adapti
ve PI
D
c
ontrol
…
(
Y
ohan
Darcy Mf
oum
boulou
)
309
The p
hase a
ng
l
e of the
dom
inant poles is:
=
−
1
(
)
=
−
1
(
0
.
63
)
=
50
.
9
5
∘
The val
ues
of t
he real
and im
a
gin
a
ry poles a
r
e cal
cula
te
d
as
fo
ll
ow:
(
)
=
−
4
=
−
4
2
=
−
2
(
)
=
(
)
×
(
)
=
−
2
(
50
.
9
5
∘
)
(
)
=
−
2
.
46
The val
ue
of t
he
f
irst
dom
inant p
ole is:
1
=
(
)
+
(
)
1
=
−
2
−
2
.
46
T
h
e
n
a
t
u
r
a
l
f
r
e
q
u
e
n
c
y
o
f
t
h
e
r
e
f
e
r
e
n
c
e
m
o
d
e
l
i
s
f
o
u
n
d
b
a
s
e
d
o
n
t
h
e
p
o
l
y
n
o
m
i
a
l
o
f
t
h
e
d
o
m
i
n
a
n
t
p
o
l
e
s
:
(
)
=
(
+
2
+
2
.
46
)
(
+
2
−
2
.
46
)
(
)
=
2
+
4
+
10
.
0516
=
√
10
.
0516
=
3
.
17
/
The desire
d
cl
os
e
d
-
lo
op tra
nsfer
f
un
ct
io
n of
the r
e
fer
e
nce
m
od
el
is:
(
)
=
10
.
0516
2
+
4
+
10
.
0516
To
e
ns
ure
sta
bi
li
t
y
and
rob
ust
ness
of
the
cl
os
e
d
-
lo
op
syst
e
m
wh
e
n
la
r
ge
var
ia
ti
ons
of
par
am
et
ers
occur,
a
pr
oport
ion
al
i
ntegral
(PI)
c
ontr
oller
is
desig
ned
on
MATL
AB
for
the
re
fer
e
nc
e
m
od
el
R(s
)
us
in
g
the
pro
gr
am
m
i
ng
c
omm
and
li
ne
(
,
)
[21,
22
]
.
The
pa
ram
et
ers
an
d
are
t
unne
d
to
fin
d
t
heir
op
ti
m
al
values.
This
ad
diti
on
al
P
I
has
al
so
the
abili
ty
to
gu
ara
nte
e
sta
bili
ty
wh
en
sud
den
e
xt
ern
al
disturba
nces a
f
fecti
ng
the t
wo
-
ta
nk
process
oc
cur. T
he para
m
et
ers
of the
P
I
c
on
tr
oller a
re
:
=
0
.
001
an
d
=
4
The
tra
nsfer
fu
nction o
f
t
he
P
I
c
on
tr
oller is
the
fo
ll
owin
g:
(
)
=
+
=
0
.
001
+
4
3.2.2.
Sele
ctio
n of the c
ontr
ol a
l
go
ri
th
m
t
o achieve
per
f
ect m
od
el
tra
c
king
To
achie
ve
pe
rf
ect
m
od
el
trackin
g,
the
fol
lowing
pro
po
rtion
al
inte
gr
al
der
ivati
ve
(PID
)
co
ntr
ol
al
gorithm
is sele
ct
ed
f
or
t
he p
ro
ces
s as:
(
)
=
(
)
+
∫
(
)
−
̇
(
)
(29)
w
he
re:
is
the
pro
portion
al
ga
in;
is
the
i
ntegr
al
gain;
is
the
de
rivati
ve
gain;
is
the
plant
outp
ut;
and
(
)
=
(
)
−
(
)
, w
it
h
(
)
as t
he
in
put o
f
t
he refe
ren
ce
m
od
el
.
The rep
rese
nta
ti
on
of the
PID
contr
oller in
L
aplace
dom
ai
n
is:
(
)
=
(
)
+
1
(
)
−
(
)
(30)
Evaluation Warning : The document was created with Spire.PDF for Python.