TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 16, No. 3, Dece
mbe
r
2
015, pp. 531
~ 538
DOI: 10.115
9
1
/telkomni
ka.
v
16i3.900
3
531
Re
cei
v
ed
Jul
y
24, 201
5; Revi
sed O
c
tob
e
r 25, 201
5; Acce
pted No
vem
ber 1
5
, 2015
A Controller Design Researh Based on the Cloud Model
Feng Jie*, Jingfan
g Wa
n
g
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
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: matlab_
b
y
sj
@12
6
.com
A
b
st
r
a
ct
A novel c
ontro
l
structure mod
e
l is pr
opos
ed
base
d
on c
l
ou
d
mo
del for th
e first time
．
T
he s
t
ructure
mo
de
l is a
no
nlin
ear
mode
l i
n
natur
e, an
d i
t
can
be c
o
mp
osed
of a gro
u
p
of unc
ertain
reaso
n
in
g rul
e
s
easily
.
No
nIin
e
a
r ma
pp
ing c
h
aracteristics of
cloud
mo
de
l
i
s
analys
ed i
n
this sudy, an
d the des
ig
n method
of the intell
ige
n
t controll
er is pr
ese
n
ted b
a
s
ed on the struc
t
ure mo
de
l,
an
d some si
mul
a
tion exa
m
ples
are
show
ed
．
Ke
y
w
ords
:
clo
ud mod
e
l, clo
u
d
control
l
er, un
certainty r
eas
o
n
in
g, non
lin
ear
map
p
i
ng, inte
ll
ige
n
t control
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
Both in artificial intelligence, data mining,
intelligent control, or in other fields, the
prima
r
y p
r
obl
em is ho
w t
o
express
kn
owle
dge,
n
a
tural l
ang
uag
e is
cle
a
rly t
he b
e
st
way
of
expre
ssi
on. Prevalen
ce b
e
twee
n quali
t
ative c
once
p
ts and qu
a
n
titative data are uncertai
n
ty,
esp
e
ci
ally ra
ndomn
e
ss a
n
d
fuzzine
s
s [1-4]. Clo
ud
model i
s
a u
n
ce
rtainty tra
n
sformation
model
betwe
en q
ual
itative con
c
e
p
t and it
s qu
antitative data, whi
c
h
are
expre
s
sed i
n
natural
lang
u
age
values, it i
s
mainly refle
c
t
i
ng the fu
zzin
ess an
d
rand
omne
ss of thi
ngs in the
ob
jective worl
d
or
the co
ncept
of human
kn
owle
dge,
an
d the two
are fully integrat
ed, the two co
nstituting
s is
mappe
d to each othe
r be
tween qu
alita
t
ive and quan
titati
ve [5, 6].
In the proce
s
s of spatial d
a
ta
mining a
nd
kno
w
le
dge
di
scovery, the
clou
d mo
de
l cal
c
ulatio
ns are
usi
ng b
o
th qua
ntitative
analysi
s
and
pro
c
e
ssi
ng of
data, but also it is full
atte
ntion to the role of qualitative thinking a
n
d
descri
p
tion.
Clou
d mod
e
l has
been
use
d
to the
exca
vation of the spa
c
e g
ene
ra
lized
kno
w
le
d
ge
and asso
ciati
on rule
s, the expres
sio
n
of knowl
edg
e
discove
ry, continuo
us da
ta discretization,
the unce
r
tai
n
ty query of spatial dat
aba
se and
uncertainty reasonin
g
, interp
retation a
n
d
recognitio
n
of
rem
o
te
sen
s
ing, an
d oth
e
r
field
[7-1
1].
Clou
d mo
del
is a
pplie
d to f
u
zzy reliability
analysi
s
of ai
rcraft navigati
on sy
st
em
s a
nd its fault d
e
t
ection; it ca
n
effectively improve
relia
bility
analysi
s
of
ai
rcraft navigati
on sy
stem [1
2]. The
cl
ou
d
model i
s
int
r
o
duced into
th
e co
ncept of t
h
e
tree, it is facilitated further to
enrich the basi
c
connot
ation of
the concept tree, abstract probl
em
is solved in
fuzzy attrib
ute domain,
the pr
omoti
on of con
c
e
p
ts are a
c
hi
eved, kno
w
l
edge
extraction
is
made
and
predictio
n rule i
s
do
ne [1
3
]. In the field
of intelligent
co
ntrol, the
clo
u
d
model is
appl
ied to the co
ntrolle
r de
sig
n
method, it
is a useful at
tempt [14, 15], the control
of
the hu
man
e
x
perien
c
e
is
conve
r
t to th
e lan
gua
ge
control
rul
e
b
y
the la
ngua
ge of
atom
s
and
clou
d mod
e
l, a co
ntrolle
r
was d
e
si
gne
d
for the in
ve
rted pe
ndulu
m
system, a
nd i
t
is for effe
ctive
control [16]. Based o
n
the one
-dim
ensi
onal a
n
d
two-dim
e
n
s
i
onal cl
oud
model, a hy
brid
pea
ce
kee
p
in
g rule
re
ason
ing metho
d
i
s
p
r
opo
se
d [
17], it simplifi
e
s the
co
mpl
e
x cont
rol rul
e
s
.
Acco
rdi
ng to
the cha
r
a
c
teristi
cs
of electro
-
hyd
r
auli
c
se
rvo vari
able pitch system, a two-
dimen
s
ion
a
l cloud mod
e
l controlle
r is d
e
s
ign
ed [18, 1
9
].
Although th
e
clou
d mo
del
ha
s be
en u
s
ed
in in
telli
g
ent controll
er de
sign,
som
e
goo
d
results
have
been
mad
e
in
these control
attempts, b
u
t this i
s
o
n
ly the be
ginni
ng,
the literature
in
this area is st
ill limited, so
there i
s
still a lot of
i
s
sues
worth further
st
udy, such as
cloud-based
sub
s
et of the languag
e model and the
unce
r
tainty
inferen
c
e rule
s, their desi
g
ns impa
ct the
control inp
u
t-output no
nlin
ear m
appi
ng
s, ho
w de
sig
n
s a
r
e
simpli
fied in cl
oud
model
cont
ro
ller
stru
cture. Fo
r this stu
d
y, tw
o a
s
p
e
ct
s
were
studie
d
,
and th
e rel
e
vant sim
u
lat
i
on results
a
r
e
given.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 16, No. 3, Dece
mb
er 201
5 : 531 – 538
532
2. Cloud
Mod
e
l
Let U
be a
common
set
U = {x}, it is calle
d on
th
e field. T i
s
a su
bset of
U on
the
langu
age, C
T
(x) is a m
appi
ng from
U to the clo
s
ed i
n
terval [0,1], for any elem
ent
x
U, there is
a sta
b
le ten
d
ency
ran
dom
numb
e
r
C
T
(x
), C
T
(x
) di
stri
bution
of U i
s
call
ed the
cl
oud m
odel
[5
]. In
particula
r, let R1(E1,E2
) b
e
norm
a
lly di
stribute
d
ra
n
dom fun
c
tion
, which E1 i
s
the expe
ct
ed
value, E2 is standard devia
tion,
by satisfying the form
ula:
)
,
(
1
En
Ex
R
x
i
(
1
)
)
,
(
1
He
En
R
P
i
(
2
)
2
)
(
2
1
i
i
P
Ex
x
i
e
(
3
)
The
cloud
m
odel i
s
con
s
tituted by the
data
)
,
2
,
1
)(
,
(
i
u
x
drop
i
i
, it is call
ed a o
ne-
dimen
s
ion
a
l
norm
a
l cl
oud
model [7], it is referred to
as the
one
-di
m
ensi
onal
no
rmal
clou
d, d
a
ta
)
,
(
i
i
u
x
on the com
p
osition of the cloud m
o
d
e
l is
calle
d a one-dimen
s
ion
a
l clou
d dropl
ets.
Whe
r
e Ex, En, and He a
r
e three imp
o
r
tant figur
e
s
feature on th
e comp
ositio
n of the clou
d
model, they a
r
e called the
expecte
d value, entro
py and hyper
entropy, whic
h is
denote
d
by (Ex,
En, He).
Figure 1. One
-
dime
nsi
onal
Normal cl
oud
model
Figure 1 is a
visual form of five one-di
mensi
onal n
o
r
mal cl
oud m
odel. Clea
rly, a point
map to its di
stributio
n val
ue field is o
n
e
-ma
n
y rela
ti
onship, it is not a clea
r di
stributio
n curve
.
Clou
d model
overall sh
ap
e is the most importan
t, distrib
u
tion chara
c
te
risti
c
of cloud d
r
o
p
let
reflect
s
the fu
zzi
ne
ss a
nd random
ne
ss o
f
the languag
e con
c
e
p
t.
The gen
eral
comp
ositio
n of the cloud
model in
fere
nce rule
s co
ntains the two parts of
the rule
ante
c
ed
ent (con
di
tions)
and th
e rule
after pi
ece
s
(rul
es
knowl
edge
). T
he kno
w
n
clo
ud
model i
s
(Ex, En, He
), if th
ere
are
pa
rticular i
nput x
= x0 co
ndition,
it is
kno
w
n
a
s
cloud
mod
e
l
with X
co
nditi
on, it i
s
referred to
as
CG
x
; if a
specifi
c
condition i
s
x
= C
T
(x
)
=
C
T
(x
0
)
,
i
t
is
c
a
ll
e
d
clou
d model
with Y conditi
on [8, 9], it is denote
d
CG
y
.
X condition
cl
oud mod
e
l:
)
,
(
1
He
En
R
P
i
(
4
)
2
)
(
2
1
i
i
P
Ex
x
i
e
(
5
)
Y condition
cl
oud mod
e
l:
)
,
(
1
He
En
R
P
i
(6)
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Controll
er
De
sign
Re
se
arh Ba
sed o
n
the Cloud M
odel (F
eng
Ji
e)
533
i
i
P
Ey
y
)
ln(
2
(
7
)
Whe
r
e: in
the
formul
a (4),
(5), the Ex, E
n
, He
is
the
cloud m
odel
di
gital features
(Ex, En, He
)
o
f
the rule ante
c
ed
ent; the Ey, En, He cloud model di
g
i
tal features
after rule
(Ey, En, He) in the
formula (
6
), (
7
).
Figure 2 sh
o
w
s the
un
cert
ainty rea
s
oni
ng of t
he one
-dime
n
si
onal
clou
d model
rules, it is
con
s
id
erin
g o
n
ly one infe
rence rul
e
. When
a
parti
cu
lar inp
u
t x re
peatedly
stimulate CG
x
, CG
x
rand
omly ge
nerate
s
a set of
i
value. These values
reflect t
he inte
nsity of the correspon
ding
activation of
qualitative rules,
whi
c
h
grou
p
i
in turn
stimulates CG
y
, these
quantitativel
y
gene
rate a set of random
clou
d dro
p
let
s
drop (y
i
,
i
). In this gro
up
of rando
m cl
oud droplet
s,
the pro
c
e
s
sin
g
method
s
can be
used
backward cl
o
ud [10, 11], t
he weighte
d
averag
e met
hod
can al
so b
e
u
s
ed [17], the corre
s
p
ondin
g
quant
itative
output value y is finally obtained.
Figure 2. Single Rul
e
Rea
s
oni
ng of one
-dime
n
si
onal
Clou
d Model
3.
The Nonline
a
r Mapping
of the
CIOUD MO
Del
Con
s
id
erin
g
the re
asonin
g
process
a
nd rea
s
oni
ng
stru
ctu
r
e of
the on
e-dim
ensi
onal
clou
d m
odel:
if A
i
then
B
i
, i=1
~
N.
Th
e
d
o
main
of
clou
d mo
del A
i
is U[-1,1],
the domain
of clo
u
d
model coll
ect
i
on B
i
is V[-1,1], cloud mod
e
l set
is defin
ed as follo
ws, N=5.
A
1
=(-1, 0.3, 0.01)
，
A
2
=(-0.5, 0.3, 0.01)
A
3
=(0, 0.3, 0.01)
，
A
4
=(0.5, 0.3, 0.01)
A5=(1, 0.3, 0.01)
，
B1=
(
-1, 0.3, 0.01)
B2=(-0.5, 0.3,
0.01)
，
B3=(0, 0.3, 0.01)
B4=(0.5, 0.3, 0.01)
，
B
5
=(1,
0.3,
0.01)
(8)
Acco
rdi
ng to the above co
ndition
s clou
d X,
Y condit
i
ons
cloud a
n
d
quantitative output
algorith
m
of the wei
ghted
averag
e, an input x is
kn
own, it is bel
ong to U, the co
rre
sp
ond
ing
output y can
be cal
c
ul
ate
d
, y is belo
ng to t
he V part. Wh
en x
is any chan
ge in the
U, the
corre
s
p
ondin
g
output y i
s
the curve
1
in Fig
u
re
3.
Wh
erei
n the
ab
sci
ssa
re
pre
s
ent
s x, the
ordin
a
te represe
n
ts y. Ob
viously, this i
s
cl
os
e to a
clea
r ma
ppi
ng line
s
, it is simil
a
r to t
he
prop
ortio
nal control relation
ship.
In addition to the one-dime
nsio
nal cl
oud
model
ca
n a
c
hieve lin
ear
mappin
g
s, it can al
so
easily to
achi
eve nonli
nea
r relatio
n
ship,
the clo
ud m
o
del set (8)
wa
s revise
d to
(9), taki
ng
N
= 4
,
nonlin
ear rel
a
tionship i
s
t
he
curve
2 i
n
Figu
re
3,
appa
rently, it is
simila
r to
the n
onlin
e
a
r
s
a
turation charac
teris
t
ics
.
A1=(-1, 0.3, 0
.
01)
，
A2=(0, 0.3, 0.01)
A3=(0.5, 0.3, 0.01)
，
A4=(1, 0.3, 0.01)
B1=(-1, 0.3, 0
.
01)
，
B2=(1, 0.3, 0.01)
B3=(1, 0.3, 0.01)
，
B
4
=(1,
0.3,
0.01)
(9)
The gra
ph 3 in Figura 3 is
for clou
d mo
del of
a formula (10
)
, N = 5, the cloud
model of
inference
rules is nonlinear rela
tion
ship.
Maintain con
s
tant N
= 5, the clo
ud mo
d
e
l wa
s revi
se
d
to (11), the di
fferent nonlin
ear ma
ppin
g
curve
4 i
s
obt
ained, it is sh
own in Fig
u
re
3.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 16, No. 3, Dece
mb
er 201
5 : 531 – 538
534
A
1
=(1, 0.2, 0.01)
，
A
2
=(0.33, 0.2, 0.01)
A
3
=(-0.33, 0.2, 0.01), A
4
=(-0.6
7, 0.2, 0.01)
A
5
=(-1, 0.2, 0.01)
，
B
1
=(-1, 0.3, 0.01)
B
2
=(-0.5, 0.3, 0.01)
，
B
3
=(0, 0.3, 0.01)
B
4
=(0.5, 0.3, 0.01)
，
B
5
=(1,
0.3,
0.01)
(10
)
A
1
=(-1, 0.2, 0.01)
，
A
2
=(-0.33, 0.2, 0.01)
A
3
=(0.33, 0.2,
0.01)
，
A
4
=(0
.
67, 0.2, 0.01)
A
5
=(1, 0.3, 0.01)
，
B
1
=(-1, 0.3, 0.01)
B
2
=(-0.5, 0.3, 0.01)
，
B
3
=(0, 0.3, 0.01)
B
4
=(0.5, 0.3, 0.01)
，
B
5
=(1,
0.3,
0.01)
(11
)
Figure 3. Non
linear M
appin
g
of the Clou
d Model
By this one
-dimen
sion
al
clou
d mod
e
l
input
an
d ou
tput mappin
g
,
the linear
mappin
g
function
can
be a
c
hi
eved
either, th
e
certain
nonli
n
e
a
r m
appi
ng f
unctio
n
can
a
l
so
be
achieved,
whi
c
h
only m
odify the defi
n
ition of a
li
mited set
of
clou
d mo
del,
or
modify several
pie
c
e
s
of
rea
s
oni
ng rules in the rul
e
base. Acco
rdi
ngly t
he desi
gn, one-dime
nsio
nal cl
oud
model control
l
er
has
strong fl
exibility, the linear
controll
er can be de
sign
ed,
o
r
no
nlinea
r controller de
sign
f
o
r
different
cont
rolled
obj
ect
s
, in additio
n
,
whe
n
con
s
id
ering
onli
ne
adju
s
tment ru
les, the
dyna
mic
swit
chin
g co
n
t
rol can b
e
achieved bet
we
en linea
r co
ntrol and n
onlin
ear control.
4. Design of the Cloud M
odel Con
t
rol
l
er
Esse
ntially, input-o
utput
control
i
s
re
ali
z
ed
by the
controlle
r, it i
s
actu
ally a
mappin
g
relation
shi
p
, i.e., it is the mappin
g
from
the devi
a
tion input o
f
the control
output. A one-
dimen
s
ion
a
l cloud mod
e
l mappin
g
is co
n
s
ide
r
ed, it
is shown in Figure 4, the map con
s
i
s
ts of two
parts: the m
u
lti inferen
c
e
rules of on
e-dim
e
n
s
iona
l cloud mo
d
e
l se
ction a
nd the weig
hted
averag
e processing
se
ctio
n. Mult
i-pa
rt rule
rea
s
oni
ng a
c
tually con
s
i
s
ts of
more th
an o
ne-
dimen
s
ion
a
l
singl
e rule-ba
s
ed
re
asonin
g
structu
r
e, t
he nu
mbe
r
o
f
rule
s in th
e
rule
ba
se i
s
N,
whe
r
e th
e in
put x stim
ulat
es
different
p
a
rts of the
rul
e
ante
c
e
dent
CGA
1
~
CG
AN,
different
values
are
ge
nerate
d
, then
it is p
r
o
c
e
s
sed by po
st
pi
ece
s
rule
s C
G
U1 ~ CG
U
N
r
u
le
s,
re
su
lt
ing
in a larg
e nu
mber of
clou
d
dropl
ets d
r
o
p
(yNj
k,
Nj
). After the
weight
ed average
cl
oud d
r
opl
ets
are tre
a
tmen
t, and ultimately the quantitative out
put value y is obtained, it is the input
x
corre
s
p
ondin
g
value.
Set deviation e
∈
E = [-1,1
], the deviation integ
r
al value ei
∈
EI = [-l, 1], the d
e
viation
cha
nge
rate
ec
∈
EC
= [-1,
1], quantitative inp
u
t e, ei
and
ec
is respectively p
r
o
c
e
s
sed
by ea
ch
one-dime
nsio
nal clo
ud m
odel map
p
in
g, the final
output of the
three contro
l comp
onent
s is
sep
a
rately
wi
th uP, uI and
uD. In e
s
se
nce, in
th
e p
h
ysical sen
s
e
control, the
control of the
s
e
three
co
mpo
nents are
si
milar to
conv
entional
PID
controlle
rs wi
th the fun
c
tio
n
s
of the th
ree
control comp
onent
s, but a
l
so it ha
s a d
i
fferent natu
r
e. Set uP
∈
UP =
[-1,1], uI
∈
UI = [-1,1]
and
uD
∈
UD
= [-1,
1]. Without loss of ge
ne
rali
ty, here-10
0
%
is rep
r
e
s
en
ted by -1, + 1
00% is indi
ca
ted
by +1. For simplicity, e to uP mappin
g
is ca
lle
d the P-type cl
oud mod
e
l controlle
r; ei to uI
mappin
g
is
said for I-type
clou
d mod
e
l controlle
r; ec mappin
g
to uD is
sai
d
for D-clou
d mo
del
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Controll
er
De
sign
Re
se
arh Ba
sed o
n
the Cloud M
odel (F
eng
Ji
e)
535
controlle
r. P-type, I-type and D-type
clou
d mo
d
e
l
controller
a
r
e e
s
sentiall
y belong to
th
e
categ
o
ry of o
ne-di
men
s
ion
a
l clo
ud mo
d
e
l, the di
ffere
nce i
s
that th
e meani
ng of
input pa
ram
e
ters
differ: In P-type clou
d mod
e
l controller, the dev
iation i
s
dire
ctly made as the the
mapping inp
u
t
of one
-dime
n
s
ion
a
l cl
oud
model; in I
-
type cl
oud
mod
e
l co
ntroll
er,
bias i
n
teg
r
al
value is mad
e
as
one-dime
nsio
nal
clou
d m
o
del m
appin
g
input; in
t
he
D-type
cl
oud
model
control
l
er, the
chan
ge
rate of the deviation .is made a
s
one
-di
m
ensi
onal
clo
ud model m
a
pping in
put.
In the domain
of E,
EI and
EC, their clo
u
d
model
s we
re defined:
E1 = (-1,0.3,0
.
01),
E2 = (-0.5,0.3,0.01)
E3 =
(0,0.3,0.
01), E
4
=
(0.5,0.3,0.01)
E5 =
(1,0.3,0.
01)
(12)
EI1 =
(-1,0.3,
0
.01), EI2 =
(-0.5,0.3,0.01)
EI3 =
(0,0.3,0.01), EI4 =
(0.5,0.3,0.01)
EI5 =
(1,0.3,0.01)
(13)
EC1 = (-1,0.3
,
0.01),
EC2 = (-0.5,0.3,0.01)
EC3 =
(0,0.3,0.01),
EC4 =
(0.5,0.3,0.01)
EC5
=
(1,0.3,0.01)
(14)
In the domain
of UP, UI and UD, the sa
me clou
d mo
dels a
r
e defin
ed:
UP1 =
UI1 =
UD1 = (-1,0.3
,
0.01),
UP2 =
UI2 =
UD
2
= (-0.5,0.3,0.01)
UP3 =
UI3 =
UD3 =
(0,0.3,0.01),
UP4 =
UI4 =
UD4 =
(0.5,0.3,0.01)
UP5
=
UI5
=
UD5
=
(1,0.3,
0
.01)
(15
)
Figure 4. The
Mapping Pro
c
e
s
sor of On
e-Di
men
s
ion
a
l Clou
d Mod
e
l
P-type clou
d model control
rules:
If E=E1, then
UP=UP1
If E=E2, then
UP=UP2
If E=E3, then
UP=UP3
If E=E4, then
UP=UP4
If
E=E5,
then
UP=UP5
(16
)
Similarly, I-type and
D-typ
e
clou
d m
ode
l control
rule
s are as follo
ws:
If EI =EI1, then UI=UI1
If EI =EI2, then UI=UI2
If EI =EI3, then UI=UI3
If EI =EI4, then UI=UI4
If EI =EI5, then U
I
=U
I
5
(17
)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 16, No. 3, Dece
mb
er 201
5 : 531 – 538
536
If EC=EC1, then UD=UD1
If EC=EC2, then UD=UD2
If EC=EC3, then UD=UD3
If EC=EC4, then UD=UD4
If
EC=EC5,
then
UD=UD5
(18
)
5.
Simulation Results a
nd Analy
s
is
First, the
clo
ud mo
del
co
ntrolle
r an
d convent
ional PID controller
ar
e
co
nsid
ere
d
in the
same
cont
ro
lled obje
c
t, their control
performan
ces a
r
e re
se
arched in di
fferent time lag
situation
s
. Co
ntrolled o
b
je
ct is in formula
(19).
Plant 0:
43
13
22
2
)
(
2
3
s
s
s
s
s
G
(
1
9
)
Clou
d mo
del
co
ntrolle
r
consi
s
ts of t
h
ree
sepa
rat
e
on
e-di
men
s
ion
a
l
cloud
model
controlle
rs: P
-
type
clou
d m
odel
co
ntrolle
r, I-ty
pe
clo
u
d
mod
e
l
contro
ller
and
D-type cl
oud
mo
de
l
controlle
r, which
implem
ent P (pro
po
rtional
)
contro
l, I (Integ
ral
)
control an
d
D
(differe
ntial)
control fun
c
ti
on, the
co
rre
s
po
ndin
g
ma
gnificatio
n
is
85,30
and
65
, respe
c
tively, and th
en A
s
the
sum
of the
three
control
comp
one
nts
are
as th
e
control
amo
u
n
t
of the
cont
rolled
obj
ect.
Conve
n
tional
PID three pa
rameters were
also taken a
s
K
P
= 85, K
I
= 30, K
D
=
65.
Figure 5 i
s
re
sults
of cl
oud
model
co
ntrolle
r
and PID cont
rolle
rs, t
hey are on t
h
e sa
me
controlled
ob
ject Plant 0,
cont
rol p
e
rf
orma
nce
results a
r
e u
n
d
e
r diffe
rent
delay conditi
ons.
Thereare th
e four
curve
s
in Fi
gure 5
(
a
)
: dotte
d
line,
da
she
d
, soli
d an
d d
o
tted
lines,
whi
c
h
are
respe
c
tively the control
effects
of clo
ud
model
c
ont
rol
l
er un
de
r del
ay 0s, 0.02
9s, 0.05s a
nd 0
.
1s
ca
se. The
r
e
are the t
w
o
curve
s
in
Fig
u
re 5
(
b
)
: dotted an
d soli
d
lines,
which
are
re
spe
c
tively
control effect
s for the PID controll
er in
the ti
me lag 0s an
d 0.029
s ca
se, and if
the stagnati
ons
rea
c
h to 0.0
5
s or 0.1
s
, PID control is diverg
en
ce,
the system is not cont
rol
l
able. Whe
n
the
system d
e
lay
s
are 0s, th
e cont
rol ch
ara
c
teri
stics
curve
s
of
clo
ud mod
e
l co
ntrolle
r and
PID
controlle
r are
almost id
enti
c
al; when th
e
time l
ag rea
c
he
s 0.0
2
9
s
, conve
n
tional
PID cont
rol i
s
critical oscillation state, an enlarged view i
s
s
een in Figure 5(b); wh
en delay
s
reach
0.05s,
conve
n
tional
PID cont
rolle
r ha
s stalle
d, and P
+ I + D-type clo
ud mod
e
l co
ntrolle
r ha
s
not
spread
whe
n
the time lag reache
s 0.1s,
and the outp
u
t
control i
s
be
tween 0.8
-
1.1
5
.
(a)
Clou
d Mo
del Co
ntrol Effect
(b) PID
Contr
o
l Effect
Figure 5. Con
t
rol perfo
rma
n
ce
comp
ari
s
on with different delays
The cases a
r
e con
s
id
ered
whe
n
the con
t
rolled
o
b
ject
cha
nge
s, thre
e different co
ntrolled
objec
ts
are set as
follows
:
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
A Controll
er
De
sign
Re
se
arh Ba
sed o
n
the Cloud M
odel (F
eng
Ji
e)
537
Plant 1:
4
.
0
2
.
1
3
.
2
)
(
2
3
s
s
s
s
G
(
2
0
)
Plant 2:
3
2
1
)
(
2
s
s
s
G
(21)
Plant 3:
52
37
12
2
)
(
2
3
s
s
s
s
s
G
(
2
2
)
Whe
n
the controlle
d obj
ect is chan
g
ed,
the para
m
eters and
stru
cture of the two
controlle
rs
are kept
con
s
tant, the
syst
em Dela
ys are set to 0
s
.
Figure 6
(
a) shows th
e con
t
rol
effect of clou
d model
controller,
wherein
the object
co
ntrol curve
3 (solid lin
e) i
s
magnified to
10
2
times. Fig
u
re
6(b
)
sho
w
s th
e control effe
ct of
the
PID
controlle
r, wh
erein
the
obje
c
t control
curve
3 (solid li
ne) i
s
ma
gnified t
o
10
6
time
s.
The
simulatio
n
re
sult
s are
sho
w
n i
n
Fig
u
re
5 an
d 6, i
t
is
clea
r that rob
u
st clo
ud mo
del co
ntrolle
r
is better than
conve
n
tional
PID controller.
(a)
Clou
d Mo
del Co
ntrol Effect
(b) PID
Contr
o
l Effect
Figure 6. Con
t
rol perfo
rma
n
ce
comp
ari
s
on of the different controll
ed obje
c
t
6 Conclu
sion
The control method of
clo
ud mod
e
l is p
r
opo
se
d in
thi
s
stu
d
y, it is not req
u
ire
d
to give a
pre
c
ise math
ematical m
o
d
e
l of controll
ed obje
c
t,
it is ba
sed
solel
y
on a perso
n's feeli
ngs
a
n
d
logic, hum
an
qualitative control expe
rien
ce
is ex
pre
s
sed in
natural la
ng
uage, the
s
e
are
conve
r
ted
to langu
age
con
t
rol
rule
s by
the clou
d
m
o
del rea
c
tor,
this metho
d
can well re
alized
from q
uantita
t
ive to qualit
ative, and th
en it
real
i
z
e
s
map
p
ing
fro
m
qu
alitative to qu
antitative
controls.
Thi
s
metho
d
i
s
b
a
se
d o
n
th
e
clou
d m
odel
controlle
r d
e
sign, control
strategy i
s
cle
a
r
and intuitive,
rea
s
oni
ng i
s
simple, j
u
st t
he digita
l
cha
r
acte
ri
stic parameters and control rule
s are
slightly modifi
ed, the different cont
rol m
aps
c
an b
e
a
c
hieve
d
. The
simulation
re
sults
sho
w
tha
t
the desi
gn of the cont
rolle
r is su
cce
ssful,
robu
st
is stro
ng, there is g
ood ap
plicati
on value.
Referen
ces
[1
]
L
i
dey
i
,
Gan
w
eny
an
, Li
u
l
u
y
i
n
.
Artificia
l
I
n
telli
ge
nce
an
d Co
gn
itive P
h
ysics
．
10
th
N
a
tion
al A
n
n
u
a
l
Confer
ence Pr
ocee
din
g
s of C
h
in
ese Adv
anc
es in ar
tifici
al i
n
telli
ge
nce. Bei
jing. 2
0
0
3
: 6-1
5
.
[2]
LI DY, Li
u C
Y
. Stud
y
on th
e
univ
e
rsal
it
y of
the n
o
rmal
clo
ud mo
de
l.
Eng
i
ne
erin
g Sci
e
n
c
es
. 20
04;
8
:
28-3
4
.
[3]
LI DY. Uncertai
nt
y
i
n
kno
w
l
e
d
ge repr
ese
n
tati
on.
Engi
ne
erin
g Scienc
es
. 20
00; 10: 73-
79.
[4]
Z
hang Ji
nchu
n
,
HU Gu
y
u
. Ap
plicati
on of un
certaint
y
r
eas
o
n
in
g base
d
on
cloud mo
del i
n
time serie
s
pred
iction.
JOURNAL OF ZHEJIANG UNIVERSITY (SCIENCE).
200
3; 4(5).
[5]
Li
De-
y
i, M
e
n
g
Ha
i-ju
n, Sh
i
Xu
e-me
i.
Mem
bersh
ip c
l
o
uds
an
d c
l
ou
d
ge
nerators.
T
h
e researc
h
a
n
d
Devel
o
p
m
ent o
f
Comp
uters.
1
995; 6: 15-
20.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 16, No. 3, Dece
mb
er 201
5 : 531 – 538
538
[6]
Yang Z
h
i-
xia
o
, F
an Yan-fen
g
. Clou
d
map
p
in
g an
d me
mbershi
p
clo
u
d
of mapp
ing.
Applic
atio
n
Rese
arch of C
o
mputers.
20
1
2
; 29(2).
[7]
Li DY, Shi
XM, W
a
rd P. Soft inference mec
h
a
n
ism bas
ed o
n
clou
d mod
e
ls. Bonn, Germa
n
y
. 19
96: 38-
63.
[8]
Li D
e
yi, H
an
J, Chan
E. K
n
o
w
l
e
d
ge r
epr
esent
ati
on an
d
disc
over
y b
a
sed on
li
ngu
istic
atoms
.
Sing
apor
e. 199
7: 89-97.
[9]
Li D
e
yi,
Di K
a
i
c
han
g, Li
D
e
ren, Sh
i
Xu
em
ei. Min
i
n
g
ass
o
ciati
on r
u
les
w
i
t
h
l
i
n
guistic
clou
d mo
dels
.
Australi
a. 199
8
.
[10]
W
A
NG Guo
y
i
n
, XU C
han
gl
in,
et al. p-ord
e
r
Norma
l
Clo
ud
Mode
l Rec
u
rsi
v
e Defin
i
tio
n
a
nd An
al
ysis
of
Bidirecti
o
n
a
l C
ogn
itive Com
p
uting.
Ch
in
ese
Journ
a
l of Co
mputers.
201
3; 3
6
(11).
[11]
DI Kaich
a
n
g
, LI De
yi, L
i
D
e
ren. C
l
ou
d T
heor
y
an
d Its Appl
icatio
ns i
n
Spati
a
l D
a
t
a
Mini
ng
an
d
Kno
w
l
e
d
ge D
i
s
c
over
y
.
JOURNAL OF IMAGE AND GRAPHICS.
1999; 4(
11): 930-
93
5.
[12]
Z
hang F
e
iz
ho
u
,
F
an Yuezu, Sun
Xia
n
fan
g
. Asse
ssment a
nd Ana
l
ysis of
F
u
zz
y
Rel
i
a
b
il
it
y
Bas
ed o
n
Clou
d
Mod
e
ls i
n
Aircraft Navi
gatio
n S
y
stem.
Acta Automatica SINICA. 2002; 28(1): 1
26-
130.
[13]
YANG Z
hao
h
u
i, LI D
e
yi.
Plan
ar Mo
del
an
d its A
ppl
icatio
n i
n
Pre
d
ictio
n
.
Ch
ine
s
e Jo
urna
l o
f
Co
mp
uters.
19
98; 21(1
1
): 961
-969.
[14]
Li D
e
-
y
i, C
h
e
ng DW
, Sh
i
Xue-mei.
Unc
e
rtaint
y re
aso
n
in
g bas
ed
on
cl
oud m
o
d
e
ls
in
control
l
ers
.
Journ
a
l of Co
mputer Scie
nce
and
Math
e
m
ati
cs w
i
th Applica
t
ion.
19
98; 3: 9
9
-12
3
.
[15]
Zhang Feiz
ho
u, Fan Y
uezu,
et al. Intellige
n
t Control Bas
ed on mem
ber
ship cl
oud g
e
n
e
rator. Act
a
Aeron
autic
a et Astronautic
a SI
NICA. 1999; 2
0
(1): 89-9
2
.
[16]
Z
hang F
e
izh
o
u
,
F
an Yuezu,
et al. Intell
ig
en
t C
ontrol Inv
e
rted Pe
nd
ulum
w
i
t
h
Cl
ou
d Mo
dels. C
ontr
o
l
T
heor
y
& Appl
i
c
ations. 20
00;
17(4): 51
9-5
2
3
.
[17]
Li Z
h
o
ng, Ya
n
g
Yid
ong. C
o
ntroll
abl
e Pitc
h
Prop
ell
e
r C
ontrol B
a
sed
on Qua
litative
Reas
oni
ng
of
Varia
n
t Dimen
sion C
l
ou
d Mo
del. Jour
na
l of Nanj
ing
Un
iv
e
r
sit
y
of Aero
na
utics & Astron
autics. 20
03
;
35(2): 16
2-1
6
7
.
[18]
Li zh
on
g, Gao
Jian. Stu
d
y
on T
w
o-
Dime
n
s
ion
Clo
ud
Mode
l C
ontrol
for Electric-
h
yd
raulic
Ser
v
o
Varia
b
le Pitch.
Journ
a
l of S
y
st
em Simul
a
tion.
2004; 1
6
(5): 1
050-
105
2.
[19]
Z
heng E
n
ran
g
,
Hui Lic
hua
n, et al. Intelli
g
ent
Contr
o
l Al
gorithm of C
l
o
ud mod
e
ls a
n
d
Simul
a
tio
n
Rese
arch of Super
heat
ed Steam T
e
m
perature Co
ntrol. Jo
urna
l of S
y
ste
m
Simulati
on. 200
7; 19(1).
Evaluation Warning : The document was created with Spire.PDF for Python.