TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 12, No. 9, September
2014, pp. 67
7
1
~ 677
9
DOI: 10.115
9
1
/telkomni
ka.
v
12i9.509
0
6771
Re
cei
v
ed Ma
y 15, 201
4; Revi
sed
Jul
y
5, 2014; Accept
ed Jul
y
20, 2
014
Error Analysis of ON-OFF and ANN Controllers Based
on Evapotranspiration
Susilo Adi Wid
y
anto*
1
, Achmad Widod
o
2
, Achm
ad
Hiday
a
tno
3
, Su
w
o
ko
4
1,2,
4
Departmen
t of Mechanica
l
Engin
eeri
ng,
Univers
i
t
y
of
Di
pon
eg
oro
(UN
D
IP)
3
Departme
n
t of Electrical En
gi
neer
ing,
Un
iver
sit
y
of Dip
on
eg
oro (UNDIP)
Jl. Prof Sudarto, SH,
T
e
mbal
ang, Semar
a
n
g
, Ph/F
ax: +
62 024
74
600
59 e
x
t 1/62 0
2
4
746
005
9 e
x
t 2
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: susilo
70@
ya
hoo.com
*
1
, a
w
i
d
@u
ndi
p.ac.id
2
, hid
y
atno
@u
n
d
ip.ac.i
d
3
,
su
w
o
k
o
@gm
a
il.com
4
A
b
st
r
a
ct
T
he ch
an
gin
g
of cl
i
m
ate
d
i
rectly
affects on
d
e
creas
i
ng agric
ultur
a
l
pro
ductio
n
. T
herefore,
application of
autom
ated
system
in
agric
ult
u
re activities is
potent
ial issue which m
u
st be considered. This
pap
er prese
n
t
s
error analys
is of ON-OFF and ANN
co
ntroll
ers w
h
ich
are app
lie
d to the auto
m
at
e
d
irrigation system
. Controllin
g of
irrigation
sy
stem
used a
c
a
lculated
Penm
an-Monteith
ev
apotrans
piration
and
a refere
n
c
e of soi
l
moi
s
ture as the c
o
mpar
ed
i
n
p
u
t. Input para
m
eters of the e
v
apotra
nspir
a
ti
o
n
inclu
d
e
d
te
mp
erature,
heat r
adi
ati
on,
atmo
spher
e pr
essu
re an
d w
i
nd
s
pee
d. T
he
per
forma
n
ce
of s
u
ch
control
l
ers w
e
re eval
uate
d
an
d compar
ed ba
sed on err
o
r
of both control
l
er
s. T
he
simul
a
ti
on resu
lt of ON
-
OF
F
controller
w
a
s used as t
he refer
ence
o
f
contro
ller
dev
elo
p
m
ent b
a
se
d on AT
meg
a
8 microco
n
troll
e
r.
T
he simul
a
tion
results show
that the error
of
the ON-OF
F
controller
can be
adj
usted by settin
g
th
e
sampli
ng
time of
the dea
d
z
o
ne discreti
z
at
io
n.
Error
of th
e
ON-OF
F
controller
w
i
th sa
mplin
g ti
me of
0.0
5
secon
d
is eq
ua
l to the error of ANN contro
ller;
thes
e are 1
4
.3
% of the refere
nce of sign
al a
m
p
litu
de.
Ke
y
w
ords
:
ON-OFF controller, ANN controller,
irrigation system
,
evapotr
anspiration
Copy
right
©
2014 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
Un
controllabl
e exploitation
of natural re
so
u
r
ces di
re
ctly affects on natural de
creasi
n
g
quality and
al
so
ch
ang
es t
he
climate te
nds to b
e
ext
r
eme
conditio
n
. In othe
r h
a
nd, the
need
of
food increa
ses propo
rtio
nally with populatio
n gr
owin
g, while
agriculture
area
s extre
m
ely
decrea
s
e
s
ca
use
d
by expansio
n of housing are
a
.
The
r
efore, application of
automated syste
m
in
agri
c
ultu
re a
c
tivities is the potential issu
e whi
c
h mu
st be co
nsi
dere
d
.
Variou
s auto
m
ated agri
c
u
l
ture system
s
hav
e
be
en
develop
ed i
n
clud
ed
we
ed
co
ntrol
system to
re
duce ne
gative impa
ct to t
he envi
r
onm
ent due
to ex
ce
ssive
use
of herbici
de
s and
Pesticid
e [1,
2], automate
d
irrigation
system [3
-5],
agri
c
ultu
ral robot
[6,
7] a
s
well as
sm
art
agric
ulture [8, 9].
In automation of irrigati
on system, th
e in
itial step can be in
stalle
d a water val
v
e
controlle
r in
irrig
a
tion
system ba
se
d o
n
time fu
ncti
on. However, this
co
ntrol
l
er
ca
n
cau
s
e
difficulty in a
c
hievin
g the
optimal g
r
o
w
i
ng conditi
o
n
. The next te
chn
o
logy i
s
a
pplication of
soil
humidity cont
rol
system. A
c
tually, this
controle
r
a
ppli
c
ation
wa
s
o
n
ly su
cces o
n
water full l
and
con
d
ition. Th
e high
co
st in
inve
stment a
nd the difficul
t
y of mainten
ance of such
sen
s
o
r
s
be
co
me
con
s
id
eratio
n
s
of the re
asons
why the
s
e sen
s
o
r
s we
re not wi
dely
applied. In the last d
e
ca
de,
appli
c
ation of
soil hu
midity sen
s
o
r
s
sp
re
ads
widely
a
n
d
it wa
s co
rre
l
ated to the reasonin
g
of low
co
st in inve
st
ment an
d ma
intenan
ce.
Currently, the
goal of
re
sea
r
ch
a
c
tivities
in this
are
a
a
r
e
aimed pa
rticu
l
arly to redu
ce water
con
s
umpti
on ba
se
d on sm
art irrigation
cont
roller sy
stem [3-
5].
With the
sa
me rea
s
on
s,
Smart
co
ntroller
syste
m
ba
sed
on
water con
s
e
r
vation in
irrig
a
tion are
a
was d
e
velo
ped. This me
thod use
d
rai
n
water a
s
a
supple
m
ent of the irrigati
on
system, so
th
at
the saving
water con
s
u
m
ption
can
b
e
in
creased
a
r
oun
d of
67%
[4]. Application
of automated
irrig
a
tion
system clo
s
ely correlate
s
to the topo
gra
p
h
i
c conditio
n
a
nalysi
s
an
d the
water
re
sou
r
ce
s. Re
sea
r
ch of sma
r
t irrigati
on
appli
c
abl
e for ag
riculture sand
area
wa
s al
so
develop
ed in
[10]. The dev
elope
d co
ntro
ller sy
st
em b
a
se
d on eva
potran
s
pi
ratio
n
co
ndition a
n
d
level of water consum
ption was lim
ited by the avail
ability of wa
ter resources. The key of the
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 9, September 20
14: 67
71 – 677
9
6772
method i
s
scheduli
ng irrig
a
tion process base
d
on th
e cal
c
ulatio
n of water b
a
la
nce of the pl
ant
[10].
1.1. Penman Montei
th Ev
apotr
a
nspir
a
tion
Theo
retically, wate
r
con
s
u
m
ption of
ag
ricultu
r
e
area
is
pro
portio
nal to the
water lo
st
cau
s
e
d
by
grou
nd
evap
oration
an
d
plant
s tra
n
s
piration
on
this a
r
e
a
. This called
as
evapotra
nspiration.
Evapo
transpiration con
d
ition re
f
e
rs to the
referen
c
e
eva
potran
s
pi
ratio
n
(ETo). In th
e
appli
c
ation,
ETo is
ra
rely
mea
s
ured,
but it is
com
m
only u
s
ed i
n
mathem
atical
equatio
n su
ch as Pe
nma
n
-Mo
n
teith [11], in which the input p
a
rameters in
clu
de tempe
r
atu
r
e,
radiatio
n, atm
o
sp
here p
r
e
s
sure an
d wi
n
d
sp
eed.
Pra
c
tically, dete
r
mination of E
T
0 is
depi
cte
d
in
Figure 1.
Figure 1. Sch
e
matic Di
ag
ram of Measurement
of We
ather Data to Cal
c
ulate
Ref
e
ren
c
e
Evapotran
s
pi
ration (ET
0
)
Penman Mo
n
t
eith equation
is described i
n
Equation (1
) [12, 13].
0.404
/
1
0.34
(1)
Whe
r
e:
ET
0
= refe
ren
c
e e
v
apotran
s
pi
ra
tion [mm/day],
∆
.
.
∗
.
.
exp
= 2.718
3 (ba
s
e of natu
r
al logarith
m
),
T
mean
= mean d
a
ily air tempe
r
atu
r
e at 2 m heig
h
t [°C],
= net ra
diatio
n at the cro
p
surfa
c
e [M
Jm
-2
/day],
G
= soil h
eat flux density [MJm
-2
/day],
=0.000
665P
= win
d
sp
eed
at 2 m height [ms
-1
],
= satu
ration v
apor p
r
e
s
sure [kPa],
=
ac
tual vapor press
u
re [kPa],
=
(T
)=
saturation vapor p
r
e
s
sure d
e
ficit [kPa],
P
= atmo
sphe
ri
c pre
s
su
re [kPa],
λ
= latent heat
of vaporizatio
n, 2.45 [MJ/kg],
Cp
= sp
ecifi
c
he
at at consta
nt pressu
re, 1.0
13x10
-3
[MJ
/
(k
g°C)],
ε
= ratio mol
e
cular weight of
water vapo
ur/dry air =0.6
2
2
.
Based
on th
e Penma
n
-M
onteith eq
uat
ion revi
sed
by the FAO
(19
99) [1
2], the process of
evapotra
nspiration wa
s si
mulated and
the
result
wa
s u
s
e
d
a
s
refe
ren
c
e
in
put of O
N
-O
FF
controlle
r in a
u
tomated irrig
a
tion system.
Several
su
ccessful attemp
ts for im
provi
ng the Pe
nm
an-M
onteith
equatio
n pa
rameters
are repo
rted
in the following
study. Harrison
(1
963) p
r
e
s
e
n
ted that the latent heat of
vapori
z
ation i
s
given a
s
a l
i
near fun
c
tion
with
air tem
peratu
r
e [14],
while a
correlation eq
uati
o
n
for
the
slo
pe of
satu
ration
vapor pressu
re cu
rve
wa
s repo
rted
by Murray
[15]. Several empi
rical
equatio
ns for cal
c
ul
ating t
he
satu
ration
vapo
r p
r
e
s
sure
in te
rm
s
of air tempe
r
ature
were
al
so
prop
osed [12
], [16-17]. Althoug
h these
equatio
ns
h
a
v
e different a
l
gorithm
s, clo
s
e results a
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Erro
r Analysi
s
of ON-OFF
and ANN Co
ntro
lle
rs Ba
se
d on… (Su
s
il
o Adi Widyan
to)
6773
obtaine
d fro
m
them. It i
s
n
o
ted th
at mo
st of
th
ese
correlati
o
ns a
r
e
re
stricted
within
the
temperature
range fro
m
0 to 50°
C.
1.2. Artificial
Neural Net
w
ork (ANN)
Control syste
m
ba
sed on ANN ha
s
be
e
n
used
i
n
ma
ny fields
su
ch
as m
onito
rin
g
syste
m
on building damage index [18],
increasing
power sy
stem stabilit
y [19, 20], adaptive control
of
spa
c
e
ro
bots
[21, 22], etc.
Princi
pally, artificial
ne
ural
netwo
r
k (A
NN) is a m
e
tho
d
for produ
ci
ng
the output si
gnal from va
riou
s input p
a
ram
e
ters
in
which its correlation i
s
determi
ned b
y
activation fun
c
tion. The ANN meth
od is depi
cted in
Figure 2. Each input sign
al (a) is given th
e
weig
ht functi
on (w). The
multiplicatio
n
betwee
n
in
put paramete
r
and the
weight functio
n
is
summ
ed an
d
simulated in
action fun
c
tio
n
to deter
min
e
the output level F(a,w). If there co
nsi
s
ts
of n inp
u
t pa
rameters
(it al
so
contai
ning
n weight
func
tions
)
, the output func
tion is
desc
ribed
in
Equation (2) [23]:
∑
∗
(
2
)
Σ
in
g
a
i
I
nput
fu
nc
ti
o
n
A
c
ti
v
a
ti
o
n
f
unc
t
i
on
ou
t
put
a
1
= g
(
in
)
I
nput
Li
n
k
s
W
j,
i
a
j
O
u
t
put
Li
n
k
s
Figure 2. ANN Method [11
]
In automated
irrig
a
tion
system, ANN
co
ntrolle
r u
s
e
s
comp
uted ev
apotra
nspiration data
as inp
u
t para
m
eter an
d th
e refe
ren
c
e
soil moi
s
tu
re
as refere
nce sign
al. The
resulted wei
ght
function
was
use
d
to
build
actual
control
l
er
syst
em
in
whi
c
h th
e
out
put is u
s
e
d
to
co
ntrol
a
servo
valve.
2. Rese
arch
Metho
d
This re
se
arch was pe
rfo
r
med
by m
odelin
g evap
otran
s
pi
ration
ba
sed
on
revise
d
Penman
-Mot
eith equation.
The input p
a
ram
e
ters
in
clud
ed tempe
r
ature,
atmosphere pressu
re,
radiatio
n
and
win
d
sp
eed.
Cal
c
ul
ated
evapotra
nspiration
rep
r
e
s
e
n
ts the
a
c
tu
al soil m
o
ist
u
re
affected by weather
condition whi
c
h
will be
adjusted t
o
the reference soil moisture.
The outp
u
t data of evapotranspi
ratio
n
was u
s
ed a
s
in
put of ON-OF
F
controlle
r a
nd ANN
controlle
r. Th
en, the bot
h
perfo
rma
n
ce
s were comp
ared. T
he
block dia
g
ra
m
s
de
scri
be b
o
th
controlle
rs a
r
e depi
cted in
Figure 3.
(a)
(b)
Figur
e 3. (a)
ON
-OFF
cont
rolle
r, (b) AN
N co
ntrolle
r
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 9, September 20
14: 67
71 – 677
9
6774
2.1. Input Paramete
r
Cal
c
ulation
o
f
evapotran
s
piration
condi
tions
wa
s pe
rforme
d
i
n
cycle ba
se of 24
h
o
u
r
s
usin
g the
we
ather data
in
put. The
s
e
d
a
ta in
clud
e:
a
i
r temp
erat
ure, ra
diation,
wind
spee
d a
nd ai
r
pre
s
sure whi
c
h
we
re a
s
su
med a
s
sinu
siodal
sign
al. Referrin
g to t
he data,
simu
lating data i
n
put
is de
scribe
d in the followin
g
se
ction.
Heat
radiatio
n is
a majo
r factor th
at
deter
mi
ne
s the rate of e
v
apotran
s
pi
ra
tion. Heat
radiatio
n is a
compo
nent on plants e
n
e
rgy bala
n
ce
with resp
ect
to the net r
adiation. In fact,
infrared ra
dia
t
ion is also
a comp
one
nt in the net
radiation. Ho
wever, the b
a
lan
c
e is al
ways
negative o
r
zero
so
that it ca
n be
negl
ected. In
thi
s
simul
a
tion, radiation
wa
s assum
ed a
s
a
sinu
sio
dal si
g
nal with am
pl
itude of 2 MJ/m
2
in range
of 112 MJ/m
2
. The frequ
en
cy is 2
π
/24 o
r
0.2168 rad/h
our de
rived from 24 hou
rs cycle.
Tempe
r
atu
r
e
and humi
d
ity are pa
ram
e
te
rs that
affect
on the dro
u
g
h
t and the atmosp
he
re
drying ca
pabi
lity.
While,
va
por pressu
re deficit
(VP
D
)
is a
meteo
r
ol
ogical vari
abl
e that i
s
u
s
e
d
to
measure the
atmosphe
re
drying
ca
p
ability.
VPD sho
w
s the
vapor
pressure differen
c
e
(co
n
centratio
n
of
wate
r va
por) b
e
twe
e
n
plant
s a
nd
a
i
r-d
ried
moi
s
t
u
re. In
the
m
odelin
g, the
air
pre
s
sure i
s
a
s
sumed
a
s
si
nusi
odal
si
gn
al with
amplit
ude of
5 kPa
and
the co
n
s
tant
of 95 kPa.
Freq
uen
cy is
2
π
/24 or 0.21
68 rad/h
o
u
r
.
Tempe
r
atu
r
e
affects
on
alt
e
ration
of th
e
ET
correlati
on VPD an
d
also
adve
c
tio
n
. Wh
en
all other fa
cto
r
s a
r
e e
qual,
ET on warm
con
d
ition
s
tends to
be la
rg
er than th
e pl
ant tempe
r
at
ure.
ET increa
se
s wa
rm veg
e
tation be
ca
use le
ss en
ergy is requ
ired to evap
orate the
water.
Tempe
r
atu
r
e
also imp
a
ct
s the relative e
ffect
iveness
of the radiant
energy and
wind affect
s on
evaporating water. Radi
a
n
t
energy
is more
effe
ctively utilized fo
r ET whe
n
tempe
r
atures
are
high. In
co
ntrast, wi
nd
ha
s more imp
a
ct
on ET
when
tempe
r
ature
s
a
r
e l
o
w. In
this
simulatio
n
,
temperature
i
s
a
si
nu
soid
al si
gnal
with
amplitud
e
5
o
C in
24
ho
urs
cycle,
so
t
he freque
ncy
is
2
π
/24 or 0.21
68 rad/h
o
u
r
. The tempe
r
at
ure rang
e (offset) i
s
arroun
d of 30
o
C.
The
wind
h
a
s
two m
a
jor rol
e
s; fi
rstly
,
it tran
spo
r
t
s
h
eat that
build
s u
p
o
n
adja
c
e
n
t
surfa
c
e
s
such as dry de
sert or asphalt
to v
egetation which acce
lerate
s evapo
ration (a p
r
o
c
ess
referred
to
a
s
a
d
vectio
n).
Secondly,
wind
se
rve
s
to a
c
cele
rat
e
evap
oration by
enha
ncing
turbule
n
t tran
sfer
of wate
r
vapor from m
o
ist veget
atio
n to the d
r
y a
t
mosph
e
re. In this
ca
se, t
h
e
wind
co
nsta
ntly repla
c
e
s
th
e moi
s
t air lo
cated
with
in
a
nd ju
st above
the plant
ca
n
opy with d
r
y
air
from ab
ove.
Wind
sp
eed
wa
s a
s
sume
d as a
sinu
so
i
dal curve
wit
h
amplitu
de o
f
1 km/h i
n
ra
nge
of 3.5 km/h.
2.2. Referen
ce
P
a
rameter
Output pa
ra
meter of the
evapotra
nspiration ca
lcul
ation re
pre
s
e
n
ts a
c
tual soil
moistu
re
influen
ced by
weathe
r parameters. Soil moisture
co
ndition
s sho
u
l
d be arrang
e
d
to approp
ri
ate
the specific
soil moisture
whi
c
h
is determined by the cultivati
on
plant type by adjustin
g
wa
ter
irrig
a
tion.
In re
al
con
d
itions,
be
side
s the type
of
plant, soil m
o
isture
i
s
al
so
influen
ced
by
age
of
plant an
d soil type. In this modelin
g, the
referen
c
e
so
il moistu
re
was a
s
sume
d
as a
Gau
s
sia
n
sinu
soi
dal si
g
nal. Refe
ren
c
e soil m
o
istu
re is a
s
sume
d
in a ra
nge
of 35% with am
plitude of 15
%,
while the fre
q
uen
cy is 2
π
/2
4 followin
g
the 24 hou
r cycle (Figu
r
e 6
)
.
2.3. Design
of Electro
nic
dev
i
ce of ON-OFF Co
ntr
o
ller
Assu
med inp
u
t signal
s we
re gen
erate
d
by el
ectroni
c circuit which used Atme
ga 8 as
sinu
sio
dal si
g
nal gen
erator. Due to the
sinu
sio
dal si
g
nals h
a
ve the
same frequ
e
n
cy, the ci
rcuit
only use
d
a
sign
al gen
era
t
or and
ea
ch
amplit
ude
a
nd con
s
tants wa
s adju
s
te
d by ope
ratio
nal
amplifier
L
M
358.
The co
n
f
iguration of
circuit
is
depi
cted in Fi
gure 4. The u
s
e
of 1 byte DAC
(083
2)
sh
ows that the a
c
cu
racy
of
output
sign
al is
19.6
m
V. Before
o
perate
d
, ea
ch
input si
gnal i
s
adju
s
ted the
op-a
m
p gai
n to approp
riate
the am
plitud
e of the input data ch
ara
c
te
ristic.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Erro
r Analysi
s
of ON-OFF
and ANN Co
ntro
lle
rs Ba
se
d on… (Su
s
il
o Adi Widyan
to)
6775
Figure 4. Electroni
c Archit
ecture of Voltage Si
nu
siod
al Signal Assumed a
s
Inpu
t Parameters
3. Results a
nd Analy
s
is
Block dia
g
ra
m of the in
pu
t para
m
eters
m
odel
and P
enman
-Mo
n
teith evapot
r
a
n
spi
r
atio
n
usin
g Simulin
k is illu
strate
d in Figu
re 5
.
A simulated
evapotra
nspi
ration
sign
al is a si
nu
soid
al
sign
al whe
r
e
its freq
uen
cy
is
same
a
s
th
e freq
uen
cy
of refe
ren
c
e
sign
al (Gau
ssian
sin
u
soid
al).
The
amplitud
e of eva
potra
npiratio
n
o
u
tp
ut sig
nal i
s
l
o
wer than
the
referen
c
e
sig
nal (Figu
r
e
6). It
implies th
e d
e
sig
ned
ON-OFF controlle
r a
s
well a
s
ANN
cont
roll
er mu
st have
the amplifi
c
a
t
ion
function.
Figure 5. Block
Ddia
gra
m
of Penman-M
onteith
Evapotran
s
pi
ration
Figure 6. Co
mpari
s
o
n
bet
wee
n
Refe
re
nce
Signal and
Calcul
ated Evapotran
s
pi
ratio
n
Signal
3.1. ON-OFF
Con
t
roller
In processin
g
the simulate
d evapotra
nspira
tion
signa
l, ON-OFF
controlle
r use
s
dead
zon
e
ci
rcuit and memo
ry integratio
n as fee
d
b
a
ck sig
nal
on the outp
u
t of calculated
evapotra
nspiration (Fi
g
u
r
e
3a). O
u
tput
si
gnal i
s
a fo
rm
of ON-OF
F
configurat
ion with
vari
ation of
pulse wi
dth [
24]. Wh
en
si
gnal i
s
hi
gh,
sole
noid valv
e will
be
ope
ned
so th
at the irrigatio
n
water
will be di
scharged to
enhance the
soil
humidity.
Reference signal i
s
approximat
ed by t
r
iangul
ar
sign
al. Actual
ly, the increa
se
of soil m
o
i
s
ture
cau
s
ed
by disch
a
rgi
n
g water is det
e
rmin
ed
by the
soil type, lev
e
l of de
nsity
and g
r
ai
n pa
rticle
size. In
the
control sy
stem, the
soil
ch
ara
c
teri
sti
c
i
s
integrate
d
into a con
s
tant
whi
c
h is u
s
e
d
to co
rrect the
slope of the i
n
crea
se of so
il moisture.
The o
c
curre
n
ce
of erro
r (the
ratio b
e
twee
n maxi
mum e
rro
r a
nd the
ampli
t
ude of
referen
c
e si
g
nal) of O
N
-O
FF cont
rolle
r can b
e
set b
y
adjusting
sampling time
of dead
zon
e
. By
assumin
g
th
e
co
nsta
nt of
soil
ch
aracte
ristic i
s
1
and
adju
s
ting
sa
mpling
time
of dea
dzone
are
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 9, September 20
14: 67
71 – 677
9
6776
0.5 and 0.00
5 se
con
d
, the cont
rolled
sign
al erro
rs
are in rang
e of 0-10 o
r
28
.6% and 0-5
or
14.3% (Fi
gure 7),
re
spe
c
t
i
vely [25]. Howeve
r, adj
u
s
ting a
very
small th
e
sampling
time
o
f
dead
zo
ne ca
n affect on the performa
n
ce of co
ntrol system caused by
the passivity
of
mech
ani
cal system, su
ch the sol
enoid v
a
lve.
Figure 7. Performa
nce of ON-OFF Controller (refe
r
en
ce si
gnal i
s
bl
ue, cont
rolled
signal i
s
red,
and O
N
-OFF
con
d
ition of solenoi
d valve is bla
c
k) with
Dea
d
Zone S
a
mpling Tim
e
of 0.005
se
con
d
[
25]
3.2. Hard
w
a
r
e
Performan
ce of O
N
-OF
F
Controller
Based
on
the
simul
a
tion
result, the
O
N
-OFF
controll
er
wa
s m
ade
by u
s
ing
AT
mega
8
microcontroll
er. Th
e p
e
rfo
r
man
c
e
is sh
own
by di
spl
a
ying the
inp
u
t and
the
ou
tput sig
nal
s u
s
in
g
oscilo
ssope.
The referen
c
e sig
nal a
n
d
the calc
ul
ated evap
otran
s
piration
(Eto
) of the
sign
al
pro
c
e
ssi
ng is sho
w
n in Fig
u
re 8a
and 8
b
, respe
c
tively. The neede
d amplificatio
n function of t
h
e
Eto signal to approp
riate the refe
ren
c
e
sign
al is
aro
u
nd of 3 times. It equals wit
h
the simulati
on
result in Figure 6.
(a)
(b)
Figure 8. Oscilloscop
e Display of (a) Re
feren
c
ce sig
n
a
l as Ga
ussia
n
sinu
sio
dal, (b) Eto si
gnal
Princi
pally, O
N
-OFF
controller i
s
de
sig
ned to tra
n
sf
orm the Eto
sign
al into pu
lse
width
modulatio
n si
gnal (P
WM)
whi
c
h is u
s
e
d
to activate a sole
noid valv
e. Acco
rdin
g to the modell
e
d
system in Fi
gure 3
a
, ON-OFF co
ntroll
er wa
s bu
ilt
by inputting prog
ram
m
ing
function into
the
microcontroll
er m
e
mo
ry (embed
ded
al
gorithm
) a
s
t
he
sign
al p
r
o
c
e
ssi
ng. PWM output
sig
nal
whi
c
h
i
s
prod
uce
d
by adju
s
ting time sa
mpling of
dea
dzo
ne of 10 ms is depi
cte
d
in Figure 9
a
. By
this P
W
M
sig
nal, the
refe
rence
sign
al i
s
a
p
p
r
oximat
ed by
trian
g
u
l
ar
sig
nal
whi
c
h i
s
sho
w
n
in
Figure 9b.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Erro
r Analysi
s
of ON-OFF
and ANN Co
ntro
lle
rs Ba
se
d on… (Su
s
il
o Adi Widyan
to)
6777
Dea
d
zo
ne sampling time
of 10 ms
(a)
Dea
d
zo
ne sampling time
of 10 ms
(b)
Figure 9. (a)
Pulse wi
dth modulatio
n si
gnal to activa
te solen
o
id valve, (b). App
r
oximation of
the
referen
c
e si
g
nal with trian
gular
sign
al a
s
co
ntrolle
d si
gnal of ON-O
FF cont
rolle
r
3.3.
AN
N Co
ntr
o
ller
Acco
rdi
ng to
the block diagra
m
in Figur
e 3
b
, the input signal
s of neural netwo
rk
controlle
r are
the cal
c
ulate
d
evapotran
s
piration
and t
he refe
ren
c
e
sign
al. Base
d
on these inp
u
ts
sign
al config
uration, neu
ral
net
work u
s
ed
feed
forward m
e
thod
. Theo
reticall
y, feed-forwa
r
d
neural net
wo
rks
can
app
roximate a
n
y nonlin
ear
f
unctio
n
, and
thus the
b
a
ckpropa
gati
on
algorith
m
s a
r
e popul
ar for
training fee
d
-f
orward neu
ral
networks [26
]
.
The wei
ght
factor obta
i
ned from training p
r
o
c
e
dure d
epe
n
ded on the
network
architectu
re.
As the mod
e
l
1, the desi
gn of neu
ral
netwo
rk
architecture is
d
e
scrib
ed a
s
t
h
e
followin
g
: 1 input layer with
15 neuron
s and 2 hidd
en
layers
with 1
0
and 5 neu
rons a
nd 1 out
put
layer. In Matlab pro
c
e
d
u
r
e
,
the network
co
n
s
tru
c
tion i
s
expre
s
sed
as the follo
wi
ng:
jarnet
=ne
w
ff(m
i
nm
ax(P),[15 10 5 1],{'t
an
sig' 'tan
sig' 'ta
n
sig' '
purelin'
}
);
From the trai
ning p
r
o
c
edu
re, the ANN contro
lled
dat
a is cl
ose to the referen
c
e
signal
(Figu
r
e 1
0a).
The e
rro
r o
c
curs
wh
en
sig
nal directio
n cha
nge
s extremely (up
-
d
o
w
n o
r
do
wn
-u
p),
while o
n
the
continu
o
u
s
alternatio
n error ten
d
s to
be low. Error co
ntroll
ed
signal of ANN
controlle
r ach
i
eves in ra
ng
e of 0-3 or 14
.3%. It
equals with the cont
rolled
sign
al error of ON-O
FF
controlle
r
with setting d
e
a
d
zo
ne
sam
p
li
ng time
of 0.
005
se
co
nd
(Figure 7
)
. By modifying th
e
netwo
rk a
r
chi
t
ecture, th
e e
rro
r
can
be
redu
ced
to
3.
9% (Fig
ure 1
0b). Erro
rs o
c
curs in
rang
e of
0-0.82
which
wa
s
dist
ribu
ted in l
o
catio
n
s
wh
ere
th
e si
gnal
dire
ction
ch
ang
e
s
. Th
e mo
dified
netwo
rk a
r
chi
t
ecture i
s
describ
ed a
s
the followi
ng: 1 input layer, 2 hidde
n layers with 96 and 1
neuron
s an
d
1 output laye
r. Base
d on li
terature revie
w
, the erro
r l
e
vel ca
n also
be re
du
ced
by
usin
g ca
sca
de co
rrelatio
n artificial n
eural
netwo
rk (CANN) m
odel with e
m
bedd
ed Ka
lman
learni
ng rule [26].
(a)
(b)
Figure 10. Error of ANN
Co
ntrolled Sig
n
a
l
Resu
lted by Modifying Net
w
or
k Ar
chite
c
ture, (a
) 1
input layer wit
h
15 neu
ron
s
and 2 hid
den
layers
with 10
and 5 neu
ro
ns an
d 1 outp
u
t layer, (b) 1
input layer, 2 hidde
n layers with 96 and
1 neuron
s an
d 1 output layer
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 9, September 20
14: 67
71 – 677
9
6778
3.3. Weight
Function o
f
ANN
Applica
b
le to
the indep
end
ent co
ntrol
system
, the wei
ght functio
n
should b
e
de
cl
ared i
n
an empi
rical formul
ation or
a con
s
tant
s if the wei
ght fu
nction i
s
not continue. The
weig
ht functio
n
is then
u
s
ed
in the o
p
e
r
ating sy
stem
algo
rith
m in
stalled i
n
mi
cro
c
o
n
troll
e
r
memory
as t
he
embed
ded co
ntrol
sy
stem.
To
cal
c
ulate
the ANN
wei
ght fun
c
tion,
activation fu
n
c
tion i
s
nee
d
ed, in thi
s
ca
se th
e
choi
ce a
c
tivation functio
n
a
r
e pu
relin a
n
d
tansig.
In Matlab proce
dure, the
cal
c
ulation of wei
ght
function
of A
N
N with
net
work
archite
c
ture of
1 in
p
u
t layer, 2
h
i
dden
layers with
96 a
n
d
1
neuron
s and
1 output layer is expre
s
sed
as the followi
ng:
Y(i) = p
u
reli
n(tansig
(inp
ut(:,i)*weight_l
a
y
er1 +
weig
ht_
b
ias1) * weig
ht_layer2
+
weig
ht_bia
s
2
)
4. Conclusio
n
Erro
r of ON-OFF co
ntroll
ed sig
nal is
cau
s
e
d
by approxim
ating
referen
c
e si
g
nal usi
n
g
triangul
ar
sig
nal. The erro
r level of ON-OFF co
ntrolle
r can b
e
adju
s
ted by settin
g
sampli
ng time
of dead
zon
e
. By adjusti
ng the de
ad
zone
sa
mpli
ng time of 0
.
005 seco
nd,
error i
s
14.
3%.
Adjusting th
e
dead
zon
e
sampli
ng tim
e
must
con
s
i
der
cha
r
a
c
teristic of the
celenoi
d valve to
prevent the delayed resp
ond of the mach
ani
cal system
. It is diferent to the ANN
controller, the
error l
e
vel i
s
affected
by d
e
termini
ng m
e
thod
and
ne
twork
archite
c
ture.
The
u
s
e of fee
d
fo
rward
ANN
i
n
cl
ude
d
1
i
nput
l
a
yer with
1
5
n
e
u
ron
s
and
2 hidde
n
laye
rs
with 10 and
5
ne
uro
n
s
an
d
1
output layer
a
nd 1 in
put lay
e
r, 2 hid
den l
a
yers
with 9
6
and 1
neu
ro
ns a
nd 1
out
put layer, e
r
rors
are 14.3% a
n
d
3.9%, resp
ectively.
Ackn
o
w
l
e
dg
ements
This
work was support
ed by Hibah Penelitian of KKP3N from
the Mi
nistry of Agri
culture
Republi
c
of I
ndonesi
a
under
Contract
no 105/
PL.220/I.1/3/2014.K, March
10, 2014. The outhors
thanks to Mr. Vighorm
e
s a
nd Rom
adh
o
n
for their hel
p in prep
ari
n
g
the experime
n
ts.
Referen
ces
[1]
Sampur
no
RM, Seminar
KB
, Suhar
noto Y
.
W
eed C
ontr
o
l D
e
cisi
on S
upp
ort S
y
stem
Based
o
n
Precisio
n Agric
u
lture Ap
pro
a
c
h
.
T
e
lkomnik
a
. 201
4; 12(2): 47
5-48
4.
[2]
Luck JD, Z
and
ona
di RS, Luc
k BD, Shearer SA. R
educi
ng
Pesticid
e Over-Applic
atio
n
w
i
t
h
Map-Bas
e
d
Automatic Bo
o
m
Section
Con
t
rol on A
g
ricu
ltural S
p
ra
yers.
Transactions of the ASABE.
201
0;
53(
3)
:
685-
690.
[3]
Card
enas
LB,
Dukes
MK, Miller G
L
.
Se
nsor-Bas
ed C
ontrol of
Irrig
a
tion in
B
e
r
m
uda
Grass
.
Procee
din
g
s
o
f
the
200
5 Am
erica
n
Soc
i
et
y of Agr
i
cult
ural
Eng
i
n
eers A
n
nua
l Inter
natio
nal
Meeti
n
g
.
T
a
mpa, F
l
orida
.
2005.
[4]
Dukes MD. W
a
ter Cons
ervat
i
on Pote
ntia
l o
f
Landsc
ape Ir
rigati
on Smart Contro
llers.
T
r
ansacti
on o
f
the ASABE.
20
12; 55(2): 4
–25
.
[5] Mecham
BQA.
Practical Gui
d
e to Usin
g Soi
l
Moisture
Se
ns
ors to Contro
l Lan
dsca
pe Irri
gatio
n.
Wo
rld
W
i
de W
eb. 20
05, http://
w
w
w
.
nc
w
c
d.or
g/ims/ims_inf
o
/practi
1d.pdf
.
[6]
T
illett ND, Hag
ue, Marc
hant J
A
. A Ro
botic
S
y
stem for
Pl
a
n
t Scal
e H
u
sb
andr
y.
J
ourn
a
l of
Agric
u
ltura
l
Engi
neer
in
g R
e
searc
h
. 199
8; 69: 169-1
78.
[7]
Johns
on
DA, N
a
ffin DJ, Pu
hal
l
a
JS, Sanc
hes
J,
W
e
llin
gton
C
K
. Devel
opm
e
n
t and
Implem
entatio
n of a
T
eam of Robotic
T
r
actor for Auto
mon
ous Pe
at Moss Harve
r
sting.
Journ
a
l of F
i
eld Rob
o
ti
cs
. 2009; 26:
549-
672.
[8]
King RE, Sigri
m
is N, Comp
utation
a
l Intell
i
genc
e in Cro
p
Productio
n
, Guest Eds.
Computers a
n
d
Electron
ics in
Agricult
ure
. 20
01; 31(1).
[9] Murase
H.
Artifici
al Inte
lli
ge
n
c
e in A
g
ricu
ltur
e, Guest Eds.
Co
mp
uters an
d Electro
n
ics i
n
Agric
u
lture
.
200
0; 29(1-
2).
[10]
Micha
e
l D, Ma
r
y
l, She
dd, D
a
vis SL.
Smar
t Irrigation C
o
ntroll
ers: O
peration of Eva
p
o
t
ranspir
a
tion-
Based C
ontrol
l
e
rs
. Universit
y
of F
l
orida. Re
p
o
rt number: AE
446. 20
09.
[11] Bro
w
n
PW
.
Convertin
g
Refer
ence Eva
potra
nspir
a
tion i
n
to T
u
rf W
a
ter Use
. T
u
rf
Irrigation
Mana
geme
n
t
Series N
o
2. Arizon
a Coo
per
ative E
x
tensi
o
n
.
2000.
[12]
Alle
n RG, Per
e
ira L
S
, Ra
es D.
Smith M. Cr
op
ev
a
potrans
pir
a
tion. Gu
id
eli
n
es for c
o
mputi
ng cr
op
w
a
te
r
requ
ireme
n
ts, Rome: F
A
O. 1998.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
Erro
r Analysi
s
of ON-OFF
and ANN Co
ntro
lle
rs Ba
se
d on… (Su
s
il
o Adi Widyan
to)
6779
[13]
Zotarelli
L, Du
kes
MD, Rom
e
ro CC,
Mi
gli
a
ccio
KW
, Mor
gan
KT
. Step
b
y
Step
C
a
lcu
l
atio
n of
th
e
Penma
n
-Mont
eith Ev
ap
otran
s
pirati
on (F
AO-
56 M
e
tho
d
).
U
n
iversit
y
of F
l
o
r
ida.
Rep
o
rt n
u
m
ber: AE4
59.
201
3.
[14]
Harriso
n L
P
. F
und
ament
al’s c
once
p
ts an
d
d
e
finiti
ons r
e
lati
ng to
hum
idit
y.
In W
e
xler
A
Ed
i
t
o
r
Hu
mi
di
t
y
and mo
isture, 3, Rein
hol
d Pu
blish
i
n
g
Co., Ne
w
Y
o
rk, 196
3.
[15]
Murra
y
,
F
W
.
On the comput
ation of satur
a
tion va
por pr
essure.
J. Appl. Meteor
. 196
7; 6: 203-2
04.
[16]
W
a
rd, Andre
w
D,. Environme
n
tal H
y
dr
olo
g
y
.
Second E
d
itio
n, Le
w
i
s Pu
blis
hers, 200
4.
[17]
Ya
w
s
CL, Yan
g
HC. T
o
estimate vapor pr
es
sure eas
il
y.
Hy
drocar
bon Pro
c
essin
g
. 198
9; 68(1
0
): 65.
[18]
Mardi
y
o
no, Su
r
y
a
n
ita R, Ad
n
an A. Monit
o
ri
ng S
y
stem o
n
Predicti
on of B
u
ild
in
g Dam
a
g
e
Inde
x
usin
g
Neur
al-N
et
w
o
r
k
.
T
e
lkomnik
a
.
2012; 1
0
(01):
155-
164.
[19]
Harikris
hn
a D, Srikanth NV.
D
y
n
a
mic Sta
b
ilit
y En
ha
nceme
n
t
of Po
w
e
r S
y
stems Usin
g Ne
ural-N
et
w
o
r
k
Contro
lle
d Static-Comp
ens
ato
r
.
Te
lko
m
nika
.
201
2; 10(0
1
): 9-16.
[20]
Gunad
in IC, Abdil
l
a
h
M, So
e
p
rija
nto A, Pe
nan
gsa
ng O. Stead
y-St
ate
Stabil
i
t
y
Asses
s
ment Usin
g
Neur
al Net
w
o
r
k Based on N
e
t
w
ork Eq
uiv
a
le
nt.
T
e
lkomnika
.
2011; 9(3): 41
1-42
2.
[21]
W
enhu
i Z
,
Yamin F
,
Xiao
pi
n
g
Ye. Ad
aptiv
e Ne
ural
Net
w
ork Ro
bust C
o
ntrol for Sp
ace
Rob
o
t
w
i
t
h
Uncertaint
y
.
Telkom
n
i
ka
. 201
3; 11(3): 51
3-5
20.
[22]
Shuh
ua Z
,
Xia
opi
ng Y,
Xia
o
m
ing J, W
enh
ui Z
.
Adaptiv
e
Contro
l of Spa
c
e Ro
bot Man
i
pul
ators
w
i
t
h
T
a
sk Space Base on Ne
ural N
e
t
w
o
r
k.
Te
lkomn
i
ka
. 20
14; 12(
2): 349-3
56.
[23]
Krogh A. W
hat are artifici
al ne
ural n
e
t
w
orks?.
Nature Biotec
hno
log
y
. 2
008;
26(2):19
5
-1
97
.
[24]
Umair SM, Us
man R. A
u
to
mation
of Irri
g
a
tion S
y
st
em
Using
ANN
ba
sed C
ontro
ller.
Internati
ona
l
Journ
a
l of Elec
trical & Co
mp
u
t
er Sciences IJ
ECS-IJENS
. 2010; 10(
02): 41
-47.
[25]
W
i
d
y
anto SA, W
i
dod
o A, Hid
a
y
atno A.
Si
mulati
on of Auto
mate
d Irrigati
o
n ON-OF
F
Co
ntroll
er Base
d
on Ev
ap
otrans
pirati
on A
n
a
l
y
s
is
. Proce
edi
n
g
of
20
14 IA
ES Internati
o
n
a
l C
onfer
enc
e
on
Electric
al
Engi
neer
in
g, Computer Sci
e
n
c
e and Inform
a
t
ics (EECCS 2014). Yog
y
a
k
ar
ta, 2014.
[26]
Diama
ntop
ou
lo
u MJ, Georg
i
ou PE,
Pa
pa
michai
l DM. P
e
rformanc
e Ev
alu
a
tion
of Artificial
Ne
ura
l
Net
w
orks
in Es
timating
Refer
e
nce Eva
potra
n
s
pirati
on
w
i
t
h
Minima
l Mete
o
r
olo
g
ica
l
Dat
a
.
Globa
l NEST
Journ
a
l
. 20
11; 13(1): 18-
27.
Evaluation Warning : The document was created with Spire.PDF for Python.