Internati
o
nal
Journal of P
o
wer Elect
roni
cs an
d
Drive
S
y
ste
m
(I
JPE
D
S)
Vol
.
4
,
No
. 2,
J
une
2
0
1
4
,
pp
. 24
1~
25
5
I
S
SN
: 208
8-8
6
9
4
2
41
Jo
urn
a
l
h
o
me
pa
ge
: h
ttp
://iaesjo
u
r
na
l.com/
o
n
lin
e/ind
e
x.ph
p
/
IJPEDS
A
Neural Network Based MPPT
T
echnique Contr
o
ller
for
Photovol
t
aic Pumping Sys
t
em
Mohammed Yaichi*,
Mohammed-Karim
Fellah**, Abdelkrim Mam
meri*
* Photovoltaic Pumping Team, R
e
search
Unit in
Renewable
Ener
gies in
the
Sahar
a
n Medium UR
ER/MS-Adrar,
CDER
** Inte
llig
ent
Co
ntrol
and
Ele
c
tri
cal
Power S
y
st
e
m
s Laborator
y,
Djilla
li
Li
abes U
n
iversit
y
of Sidi-
B
el-Abbes
Article Info
A
B
STRAC
T
Article histo
r
y:
Received
Ja
n 27, 2014
Rev
i
sed
Mar
20
, 20
14
Accepte
d Apr 8, 2014
The ar
ticle pro
poses a novel
method us
ing the artificial n
e
ural network
(ANN) for the improvement of the perfo
rmances of a photovoltaic s
y
stem
composed of a photovoltaic (P
V) array
,
an
inv
e
rter
, a motor as
y
n
chronous
and a centr
ifug
al pump. For t
h
is ty
pe
of s
y
s
t
em, differen
t
optimization
strateg
i
es hav
e
been p
r
oposed
to
improve
th
e over
of th
e
PV sy
stem
effic
i
enc
y
,
i
.
e. t
h
e P
V
generator
is
forced to operat
e
at its
m
a
xi
m
u
m power
point “MPPT”, generally
,
b
y
the
inser
tion
of DC/DC boost converter
between
the pho
tovoltaic a
rray
and the inver
t
er
.
In this work we propose an
approach
, wher
e optim
iz
ation
i
s
reali
zed with
out need
addin
g
a DC/DC
converter to
the
chain
,
using f
i
eld-or
iented con
t
r
o
l through
the m
onitoring of
the vol
tag
e
-fed
i
nverter
frequ
enc
y
.
Th
e m
o
tor
is
also ensured
in
a
ll insol
a
tion
conditions
. A multil
ay
er f
eed fo
rward perceptio
n t
y
p
e
NN is proposed for
MPPT control, and the back-pro
pagation
algor
ithm is used for training
. Th
e
performances of the drive with ANN-
based MPPT are
excellent. The
maximum powe
r
point (MPP) ca
n be easil
y
o
b
tained to frequency-controlled
drive.
Keyword:
PV Pum
p
ing
S
y
stem
Fi
el
d-
ori
e
nt
e
d
cont
rol
MPPT
Neu
r
al Netw
or
k
Perform
a
nces
Copyright ©
201
4 Institut
e
o
f
Ad
vanced
Engin
eer
ing and S
c
i
e
nce.
All rights re
se
rve
d
.
Co
rresp
ond
i
ng
Autho
r
:
M
oham
m
ed Y
a
i
c
hi
,
Ph
ot
o
vol
t
a
i
c
P
u
m
p
i
ng Team
, Ph
ot
o
vol
t
a
i
c
C
o
n
v
e
r
si
o
n
Di
vi
si
on
,
R
e
search
U
n
i
t
i
n
R
e
ne
wa
bl
e
Ener
gi
es i
n
t
h
e
Saha
ra
n M
e
di
um
UR
ER
/
M
S-A
d
ra
r,
C
D
ER
,
Al
ge
ri
a,
Em
a
il: ya
ich
i
_
m
o
h
a
mmed
@
yah
o
o
.
fr
1.
INTRODUCTION
Th
e Sah
a
ran
med
i
u
m
b
y
i
t
s
arid
n
a
t
u
re
an
d
t
h
e av
ailab
ility o
f
so
lar i
mmen
s
e can retu
rn
th
e
appl
i
cat
i
o
n o
f
t
h
e wat
e
r
pum
pi
n
g
vi
a
ph
ot
o
vol
t
a
i
c
(P
V)
so
lar pu
m
p
in
g, lik
e a v
e
ry imp
o
rtan
t an
d
the
m
o
st
attractiv
e op
eratio
n
in th
e agricu
ltu
ral an
d ind
u
s
t
r
ial do
m
a
i
n
s.
During
th
e d
a
y
,
th
e sp
eed
o
f
t
h
e m
o
to
r asy
n
ch
ron
o
u
s
d
e
p
e
nd
s
o
n
th
e tem
p
eratu
r
e and
t
h
e qu
an
tity o
f
th
e so
lar rad
i
atio
n
th
at is fallen
o
n
th
e
p
hoto
v
o
ltaic p
a
n
e
l
s
to
ex
tract th
e
m
a
x
i
m
u
m p
o
w
er. Th
is latt
er, is
obt
ai
ne
d
by
t
h
e pr
o
p
er a
d
j
u
st
m
e
nt
of t
h
e i
n
vert
er
fre
q
u
enc
y
(by increa
se
or
decrea
sing) instead of
the MPPT
circuit (m
axim
um
powe
r
point tracker), inducing a t
o
tal
im
provem
ent of the efficien
cy of the system
without
need a
ddi
ng a
DC/DC converter to the
ch
ai
n. O
n
t
h
e ot
her
han
d
, t
h
e
fl
o
w
daily and the efficiency
of the
m
o
tor-p
um
p for a total hea
d
(HM
T
) de
pe
n
d
o
n
the s
p
ee
d
(related
to
t
h
e stato
r
frequ
e
ncy) if we con
s
id
ered
t
h
at
t
h
e num
ber of st
a
g
es i
s
fi
xed (i
.e St
a
nda
r
d
C
e
nt
ri
f
u
gal
Pum
p
, SC
P) [
1
]
,
[2]
.
F
o
r exam
pl
e, for
a gi
ven
spee
d,
t
h
e pum
p fu
nct
i
o
ns
at
an ope
rat
i
o
n p
o
i
n
t
.
If
i
r
radi
a
n
ce decrea
ses, t
h
at
i
n
v
o
l
v
es
a
re
duct
i
o
n spe
e
d;
t
h
e
sy
st
em
operat
i
ng
poi
nt
i
s
det
e
rm
i
n
ed by
t
h
e i
n
t
e
rsect
i
on
poi
nt
of t
h
e (
Q
, H
)
cha
r
act
eri
s
t
i
c
s of t
h
e
m
o
t
o
r-
pum
p.
The t
e
m
p
erat
ure and t
h
e
rad
i
at
i
on have
ra
nd
om
va
ri
at
i
on nat
u
re. B
y
appl
y
i
n
g
t
h
e t
echni
que
of
MPPT, the e
f
ficiency of the
syste
m
rises whateve
r
is
th
e
so
lar
rad
i
ation v
a
lu
e an
d th
e te
m
p
eratu
r
e
of th
e
envi
ro
nm
ent
.
Thi
s
t
e
c
hni
que
was
a
ppl
i
e
d
b
y
di
f
f
ere
n
t
m
e
ans [3]
.
I
n
t
h
i
s
st
udy
, we exec
u
t
e t
h
i
s
t
echni
qu
e wi
t
h
th
e use
o
f
artificial n
e
ural
networ
k (ANN). Th
e ap
p
licatio
n of
ANN is
rece
nt
l
y
gr
o
w
i
ng i
n
ph
ot
o
v
o
l
t
a
i
c
sy
st
em
. A feedf
o
r
w
a
r
d
A
NN i
m
pl
em
ent
s
no
nl
i
n
ear i
n
p
u
t
-
o
u
t
p
ut
m
a
ppi
ng
. A
ba
ck p
r
o
p
a
g
at
i
o
n t
y
pe
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-86
94
I
J
PED
S
Vo
l. 4
,
No
. 2
,
Jun
e
2
014
:
24
1
–
25
5
24
2
feed
fo
r
w
ar
d
A
NN i
s
t
r
ai
ne
d
wi
t
h
t
h
e
dat
a
b
a
ses ge
nerat
e
d
by
a
pr
oced
u
r
e t
h
at
eval
uat
e
d
on a
fi
el
d
-
o
r
i
e
nt
ed
cont
rol
l
e
d i
n
d
u
ct
i
on m
o
t
o
r
d
r
i
v
e, i
t
ex
pl
ai
n
s
i
n
det
a
i
l
i
n
§
3
. T
h
e bac
k
pr
opa
gat
i
o
n t
r
ai
n
i
ng al
g
o
ri
t
h
m
need
s
o
n
l
y inpu
ts and
th
e d
e
sired ou
tpu
t
to
ad
ap
t t
h
e
weigh
t
[4
].
2.
SOLAR P
U
MPING
S
Y
STE
M
S
Th
e
p
h
o
t
ovo
ltaic syste
m
is c
o
n
s
titu
ted
o
f
a self-p
ilo
ted
asyn
chrono
u
s
mo
tor op
eratin
g
a cen
trifug
al
lo
ad
.
Th
e
un
it is fed
b
y
so
lar
cells th
rou
g
h
an
inv
e
rt
er. Pum
p
in
g
with
ou
t
in
term
ed
iate p
o
w
er storag
e en
ab
led
us t
o
have a si
m
p
l
e
r phot
ov
o
l
t
a
i
c
sy
st
em
,
m
o
re
rel
i
a
bl
e;
m
a
i
n
t
e
nan
ce-
fee
i
s
l
e
ss expe
nsi
v
e t
h
a
n
a sy
st
em
wi
t
h
b
a
tter
y
[5
].
Th
e system
to
b
e
inv
e
stig
at
ed
is an
immersed
centri
f
ugal m
o
tor-pum
p
SP5A7 (nom
inal speed
2
860
rp
m
f
o
r an
indu
str
i
al fr
eq
u
e
n
c
y eq
u
a
lizes w
ith 50H
z)
.
Fi
gu
re
1 s
h
ow
s t
h
e
bl
oc
k
di
a
g
ram
of a
cl
os
ed l
o
o
p
fi
el
d-
o
r
i
e
nt
ed
co
nt
r
o
l
l
ed (
F
OC
) i
n
d
u
ct
i
o
n
m
o
t
o
r
d
r
i
v
e i
n
corp
oratin
g th
e pro
posed ANN
-b
ased
M
PPT co
n
t
ro
ller. Th
e co
m
m
an
d vo
ltag
e
,
i
s
ge
ne
rat
e
d
from
the fre
quency or s
p
ee
d
command
(irra
diance
G a
n
d t
e
m
p
erature T
a
depe
n
d
ent
s
).
Fi
gu
re
1.
Fi
el
d
-
o
r
i
e
nt
ed
co
nt
r
o
l
o
f
i
n
d
u
ct
i
o
n
m
o
t
o
r sh
owi
n
g
ne
ural
net
w
or
k
base
d i
m
pl
em
ent
a
t
i
o
n
2.
1.
Ph
ot
ov
ol
t
a
i
c
arr
a
y m
o
d
e
l
Som
e
authors
have
propose
d m
o
re
sophisticated m
odels that prese
n
t bet
t
er accuracy a
n
d serve
for
di
ffe
re
nt
pu
rp
o
s
es. The basi
c
equat
i
o
n f
r
om
t
h
e t
h
eory
o
f
s
e
m
i
cond
uct
o
rs
t
h
at
m
a
t
h
em
ati
cal
l
y
descri
be
s t
h
e
I–V c
h
aracteristic of the
PV
array,
equ
a
tion (1
) d
e
scr
i
b
e
s
th
e sing
le-d
iode m
o
d
e
l pr
esen
ted in
Figu
r
e
2
[
6
],
[7]
.
Fi
gu
re
2.
Si
n
g
l
e
-di
ode
m
odel
of
t
h
e t
h
eo
ret
i
cal
PV cel
l
a
n
d
equi
val
e
nt
ci
rc
ui
t
o
f
a
pract
i
c
al
PV
de
vi
ce
includi
ng the
s
e
ries and
pa
rallel resistances
(1)
1
∆
(2)
e
xp
1
(3)
(4)
Whe
n
:
(5)
p
h
I
D
I
D
s
h
I
s
s
h
p
N
R
N
s
s
p
N
R
N
V
I
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
PED
S
I
S
SN
:
208
8-8
6
9
4
A N
e
ur
al
N
e
t
w
ork B
a
se
d M
P
PT Tec
h
ni
q
u
e
C
ont
r
o
l
l
e
r f
o
r
Ph
ot
ov
ol
t
a
i
c
P
u
m
p
i
n
g
Syst
e
m
(
M
oh
a
mme
d
Y
a
i
c
hi
)
24
3
(6)
W
h
er
e
and
are th
e
p
h
o
t
ovo
ltaic (PV) and
satu
ration
curren
t
s, resp
ectiv
el
y, of th
e array
an
d
cells conne
cted i
n
series
pe
r m
o
dule.
is
the e
q
uivalent
series
resistance
of the
array and
is t
h
e
equi
val
e
nt
pa
r
a
l
l
e
l
resi
st
ance. M
o
d
u
l
e
s co
n
n
ect
ed i
n
paral
l
el
i
n
crease t
h
e curr
ent
an
d
m
odul
es con
n
e
c
t
e
d i
n
seri
es pr
o
v
i
d
e
g
r
eat
er
out
pu
t
v
o
l
t
a
ges.
and
a
r
e t
h
e
num
ber of m
o
dules connected
in se
ries a
nd in
p
a
rallel resp
ectiv
ely o
f
th
e en
t
i
re array.
e
is the electron c
h
a
r
ge
1.6
∙
1
0
C
,
i
s
t
h
e B
o
l
t
zm
ann co
nst
a
nt
1.3
8
∙
1
0
J/K
,
Ke
lv
i
n
is the tem
p
erature
of the
p-
n
j
unct
i
o
n,
a
n
d
is th
e
d
i
od
e id
eality co
n
s
tan
t
.
(7)
W
h
er
e
is th
e
b
a
ndg
ap en
ergy of
t
h
e sem
i
c
o
ndu
ctor
≅
1
.12
for
th
e po
lycrystallin
e
Si
at
25
and
is the
refe
rence
tem
p
erature
.
(8)
The
n
(9)
∆
(10)
The cell te
m
p
erature ca
n
be
determ
ined from
a
m
bient te
m
p
erature and with the
he
lp of s
o
me
standa
rd
test in
fo
rm
ation:
(11)
2.
2. I
n
ve
rter
Mo
del
Fi
gu
re 3 s
h
o
w
s a seri
es-c
on
nect
i
o
n o
f
si
ngl
e p
h
as
e in
v
e
rters. In
t
h
is asy
m
m
e
trical
m
u
lt
ilev
e
l
topology, each H-bri
dge m
u
st be fe
d
by an indivi
dual
DC-voltage and is used to i
n
crease the m
u
ltilevel
co
nv
er
ter
p
e
rf
or
m
a
n
ces, w
ithou
t add
i
ng
an
y co
m
p
lex
ity in
the po
wer circu
i
t
[8
], [9
].
d
1
K
'
d
1
K
d
2
K
'
d
2
K
a
2
d
i
d
3
K
'
d
3
K
d
4
K
'
d
4
K
o
d
1
v
d
2
v
1
d
i
d
1
U
d
2
U
Fi
gu
re
3.
St
r
u
c
t
ure asy
m
m
e
t
r
ical
5,
7
o
r
9 l
e
vel
casca
ded
i
n
vert
er
7
5
1
.
9
,
2
, at stan
d
a
rd test con
d
ition
s
(STC)
Each
co
up
le of switch
e
s
,
,
∈
2
1
,2
∙∙
is cont
rol
l
e
d
b
y
a cou
p
l
e
of
swi
t
c
hi
n
g
fu
nct
i
o
ns
,
su
ch th
at:
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-86
94
I
J
PED
S
Vo
l. 4
,
No
. 2
,
Jun
e
2
014
:
24
1
–
25
5
24
4
,
∈
0,1
1
(12)
Th
e
conv
ersi
on
o
f
th
e switch
co
mm
u
t
at
io
ns in to
a
v
o
l
t
a
ge i
s
de
scri
b
e
d
by
a
c
o
n
v
er
s
i
on
f
u
n
c
t
i
o
n
,
su
ch
t
h
at:
⇒
1,0,1
(13)
The
o
u
t
p
ut
v
o
l
t
age o
f
eac
h ce
l
l
i
s
gi
ve
n
by
:
∙
,0
,
(14)
Wi
t
h
th
e i
n
pu
t
vo
ltag
e
o
f
each
cell.
Th
e
ou
tpu
t
vo
l
t
ag
e of
a ph
ase is g
i
v
e
n b
y
:
(15)
Wi
t
h
:
s
,
sa
,
b
o
r
c
;
d
,
d
1
,
2
o
r
3
: re
present t
h
e
num
ber of the
phas
e (leg).
2
1
1
12
1
1
1
2
(16)
is th
e array i
n
p
u
t
.
To i
m
prove t
h
e out
p
u
t
v
o
l
t
a
ge f
o
r s
u
ch i
n
vert
er
s, m
a
ny
di
ffe
re
nt
m
odu
l
a
t
i
on st
rat
e
gi
e
s
have
bee
n
devel
ope
d.
Am
ong these strategies, the SVM (Space Vect
or Modulation). This techni
que provi
de the nearest
swi
t
c
hi
n
g
vect
ors se
q
u
ence t
o
t
h
e re
fere
nce
vect
or t
h
at
i
s
depe
n
d
s t
h
e m
o
d
u
l
a
t
i
on i
nde
x
r
wi
t
h
out
i
n
vol
vi
n
g
t
r
i
g
o
n
o
m
e
t
r
i
c
funct
i
o
ns
an
d
p
r
o
v
i
d
e t
h
e a
ddi
t
i
onal
ad
va
nt
ages
of
su
pe
ri
or
harm
oni
c
qual
i
t
y
. The ge
ne
ra
l
i
zed
alg
o
rith
m
b
e
in
g
u
s
ed
to
d
e
termin
e, for th
e hex
a
gon
al stru
ctu
r
e, th
e ex
act p
o
s
ition
of th
e v
ecto
r
o
f
referen
c
e
(det
ect
i
o
n
of
n
earest
t
h
ree
ve
ct
ors a
n
d
dut
y
cy
cl
es com
put
at
i
on)
was
de
v
e
l
ope
d a
n
d
st
u
d
i
e
d i
n
det
a
i
l
i
n
[9]
.
2.
3. M
o
t
o
r As
ynchr
o
n
o
us
Mo
del
The m
a
t
h
em
ati
cal
dy
nam
i
c
m
odel
of t
h
e
asy
n
ch
ro
n
o
u
s
m
o
t
o
r i
s
descr
i
bed i
n
[1
0]
, [
11]
by
t
h
e
fo
llowing
equ
a
tio
n
:
(17)
Wi
t
h
:
0
0
;
00
0
0
00
;
;
1
;
;
,
:
axes
c
o
r
r
es
po
ndi
ng
t
o
t
h
e a
s
y
n
ch
ro
n
ous
ref
e
rence
a
x
es
i
n
Par
k
m
odel
.
,
,
,
are
d-
axis
stat
or
current, rotor flux and q-a
x
is
stator
c
u
r
r
ent,
rot
o
r
flu
x
res
p
ectiv
ely
.
,
,
,
and
a
r
e: stator
and
r
o
t
o
r m
a
i
n
i
n
duct
a
nces,
r
e
si
st
ances a
n
d
m
u
t
u
al
i
nduct
a
nce
res
p
ect
i
v
el
y
.
σ
di
s
p
ersi
on
f
act
or,
and
are th
e angu
lar sp
eed
o
f
th
e
ro
tatin
g
m
a
gnet
i
c and electric
fields
respecti
v
ely.
The i
n
d
u
ct
i
o
n
m
o
t
o
r de
vel
o
p
s
an
el
ect
rom
a
gnet
i
c
t
o
r
que
expresse
d as
follows:
(
1
8)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
PED
S
I
S
SN
:
208
8-8
6
9
4
A N
e
ur
al
N
e
t
w
ork B
a
se
d M
P
PT Tec
h
ni
q
u
e
C
ont
r
o
l
l
e
r f
o
r
Ph
ot
ov
ol
t
a
i
c
P
u
m
p
i
n
g
Syst
e
m
(
M
oh
a
mme
d
Y
a
i
c
hi
)
24
5
i
s
t
h
e num
ber of
p
o
l
e
pai
r
s.
2.
4. Cen
t
ri
fu
g
a
l
Pum
p
Mo
d
e
l
The va
ri
at
i
on
of t
h
e
p
u
m
p
’s spee
d can
gi
ve
us n
u
m
e
rous c
h
arts Q-H. T
h
e
use of a centri
f
ugal pum
p
n
eeds a
preli
m
in
ary stud
y
o
f
th
e m
o
st i
m
p
o
r
tan
t
ch
arts
th
at
ch
aracterize it, wh
ere efficien
cy will
b
e
o
p
t
i
m
u
m
with
th
e to
tal
h
ead
and
th
e sp
eed
env
i
sag
e
d
b
y
co
n
t
ro
l t
h
e pu
m
p
ed
water qu
an
tity to
a d
e
sirab
l
e h
e
ad
.
In
ad
d
ition
,
t
h
ey
are related
to di
m
e
n
s
io
n
s
, k
i
nd
s and
sp
eed of th
e pu
m
p
.
Th
e ch
art o
f
the water q
u
a
n
tity-h
ead
Q-H (Fig
ure 4
)
e
x
p
l
ai
n
s
th
e d
i
fferent v
a
riatio
n
s
in
th
e h
ead
o
f
p
u
m
p
i
n
g
, accord
i
n
g
to
water
q
u
a
n
tity wh
ich fo
rm
s b
e
n
t
charts. Th
e cod
e
th
en
pro
ceed
s
to
th
e
g
e
n
e
rati
o
n
of
the entire c
u
rve. Succe
ssivel
y
,
the progra
m generates a
series of cha
r
acteristics for
a
fi
xe
d ra
n
g
e
of s
h
a
f
t
ro
tation
a
l sp
eed
s
(g
en
erally 3
5
–56
Hz), related
to
th
e sam
e
p
u
m
p
.
It all
is
d
o
n
e
referri
n
g
to
th
e a
ffi
n
ity laws
a
t
v
a
r
i
ab
le
sp
ee
d
.
T
h
e a
ffi
ni
t
y
l
a
ws a
r
e
val
i
d
onl
y
un
de
r c
o
n
d
i
t
i
ons
o
f
c
o
n
s
t
a
nt
e
ffi
cien
cy [
1
],
[1
2
]-[
13
].
Ω
Ω
⁄
;
Ω
Ω
⁄
;
Ω
Ω
⁄
at
(19)
Whe
r
e
,
,
,
Ω
and
are the
actual vol
um
etric
fl
o
w
rat
e
,
hea
d
,
t
h
e
pum
p s
h
aft
i
n
pu
t
powe
r, rotational s
p
ee
d
res
p
ectively and t
h
e pum
p
e
ffi
ciency, whereas
,
,
,
Ω
and
are the
sam
e
variables
refe
rre
d to the
desi
gn
spee
d.
Once t
h
e cha
r
acteristic and power
dem
a
nd c
u
rves a
r
e
de
fi
ned
,
t
h
e
pum
p e
ffi
cie
n
cy m
a
y be
calcu
lated
as
fo
llo
ws:
(20
)
Whe
r
e,
∙
∙
(21)
is th
e
po
wer
ou
tpu
t
in
term
s o
f
pu
m
p
ed
water [
k
W
]
(i.e.
h
y
d
ra
ulic o
u
tp
ut p
o
we
r)
,
∙
:
9.81
∙
th
e con
s
tan
t
o
f
grav
ity,
1
0
∙
water vo
lu
m
i
c
m
a
ss th
e co
nstan
t
an
d
is th
e l
o
ad constan
t
.
Fi
gu
re 4.
Characteristic
Usi
n
g t
h
e rel
a
t
i
ons
(1
9) a
n
d t
h
e p
u
m
p
dat
a
sheet
, a set
of c
u
r
v
es
gi
vi
n
g
t
h
e
fl
ow
v
e
r
s
us th
e h
e
ad
and
p
a
ram
e
terized
b
y
sp
eed
can
b
e
th
en
ob
tained
,
b
u
t
th
is
p
r
o
cedure is larg
e and
in
ord
e
r to
u
s
e th
ese
resu
lts
du
ri
n
g
cal
cul
a
t
i
ons t
h
i
s
set
o
f
poi
nt
s sh
oul
d b
e
fi
t
t
e
d t
o
obt
ai
n an al
ge
brai
c
equat
i
o
n, t
h
i
s
i
s
do
ne by
t
h
e u
s
e of
a t
w
o-
va
ri
abl
e
s t
h
i
r
d
or
der
p
o
l
y
nom
i
a
l
fun
c
t
i
on i
n
o
r
der
t
o
achi
e
ve t
h
e
best
p
o
ssi
bl
e r
e
gres
si
o
n
coe
ffi
cient
[1
3]
:
,
∙
∙
∙
∙
∙
∙
∙
∙
∙
∙
∙
∙
(
2
2)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-86
94
I
J
PED
S
Vo
l. 4
,
No
. 2
,
Jun
e
2
014
:
24
1
–
25
5
24
6
Fo
r th
e
d
e
term
in
atio
n
of th
e pu
m
p
o
p
e
rating
p
o
i
n
t
it is req
u
i
red to
kn
ow
b
o
th
th
e
pu
m
p
and
p
i
p
e
lin
e
characte
r
istics. The
pipi
ng syste
m
d
eals with
th
e t
o
tal h
e
ad
th
at m
u
st b
e
ov
erco
m
e
b
y
th
e pu
m
p
. Th
e Q-H
charact
e
r
i
s
t
i
c
of t
h
e pi
pe net
w
or
k i
s
gi
ve
n as
a fun
c
t
i
on
of t
h
e ge
o
d
et
i
c
head an
d hea
d
l
o
sses (as f
u
nct
i
on
o
f
t
h
e fl
o
w
-
r
at
e)
[1
4]
. Th
us
, i
t
sho
u
l
d
at
l
east
equal
t
h
e
hea
d
co
rres
p
on
di
n
g
t
h
e fl
ow c
o
m
put
ed by
t
h
e
pum
p
f
l
ow
-h
ead equatio
n
.
I
t
co
m
e
s
:
(23)
The c
o
nst
a
nt
relates to
th
e
b
e
ad
loss cau
s
ed
b
y
fl
u
i
d
friction
.
: Geo
d
e
tic
head
.
2.
5. Vec
t
ori
a
l
Com
m
a
nd Ap
pro
a
ch
The
vect
o
r
co
nt
r
o
l
i
s
base
d
on t
h
e fi
el
d-
or
i
e
nt
ed c
ont
r
o
l
(FOC
) m
e
t
hod
. Fo
r t
h
e
re
gul
at
i
on
of t
h
e
m
a
in
varia
b
les (cu
rre
nt, flu
x
, spee
d)
t
o
thei
r
refe
rence
val
u
e
s
, re
g
u
lators
w
e
re
used
o
f
t
h
e
ty
pe PI
[
10]
,
[
1
1]
.
So
we h
a
v
e
as in
terest to
k
e
ep
on
ly th
e
axis flux com
p
onent, and
th
at mean
s to
o
b
lig
e
th
e
axis
fl
u
x
c
o
m
pone
n
t
t
o
be
zer
o.
0
and
(24)
Th
e
flux
an
d torqu
e
will b
e
red
u
c
ed
resp
ectiv
ely th
en to
:
(25)
I
(26)
There
f
ore t
h
e
and
ax
is statoric v
o
ltages eq
uatio
n
s
will b
e
:
(2
7)
Th
e co
m
p
en
satio
n h
a
s as
o
b
jectiv
e to
u
n
c
ou
p
l
e th
e two
and
ax
is
vo
ltag
e
s and cu
rrents.
Und
e
r
su
ch
co
nd
itio
ns, th
e system
b
eco
m
e
s lin
ear lik
e in
case
o
f
DC m
o
to
r.
Thus we
h
a
v
e
:
(28)
Whe
r
e,
(29)
(30)
Are th
e co
m
p
en
sation
v
o
ltag
e
s.
Th
e tran
sfer fun
c
tio
ns of th
e
p
l
an
t fo
r t
h
e co
n
t
ro
lle
rs
of t
h
e vect
o
r-c
ont
ro
l
l
e
d i
n
d
u
ct
i
on
m
o
t
o
r dri
v
e
s
can be deri
ved
as
sh
o
w
n
i
n
[
1
0]
,
[
1
1]
.
S
p
eed,
and
cu
rren
t con
t
ro
llers:
(31)
Whe
r
e
is the
Laplace
ope
rat
o
r.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
PED
S
I
S
SN
:
208
8-8
6
9
4
A N
e
ur
al
N
e
t
w
ork B
a
se
d M
P
PT Tec
h
ni
q
u
e
C
ont
r
o
l
l
e
r f
o
r
Ph
ot
ov
ol
t
a
i
c
P
u
m
p
i
n
g
Syst
e
m
(
M
oh
a
mme
d
Y
a
i
c
hi
)
24
7
B
a
sed o
n
t
h
os
e equat
i
o
ns a
n
d aft
e
r s
o
m
e
al
geb
r
a (Im
posi
t
i
on o
f
t
h
e
pol
es
m
e
t
hod
) we
obt
ai
n
PI
param
e
ters for speed a
n
d current (Table
1), were
J
is th
e ro
tor in
ertia m
o
men
t
,
is th
e visco
s
ity co
efficien
t
o
f
the in
du
cti
o
n
m
o
to
r. After
is ob
tain
ed
upo
n
calcu
latio
n
of th
e
tran
sfer fun
c
tion
of a clo
s
ed lo
op
,
param
e
ters
and
are i
d
e
n
t
i
f
i
e
d
fr
om
im
posi
t
i
on
o
f
t
h
e
p
o
l
e
s m
e
t
hod.
A
w
i
se ch
oi
ce t
h
e
and
was
m
a
de by
c
h
an
g
i
ng t
h
e
val
u
e
o
f
.
Table 1. PI
Re
gulator
C
o
effic
i
ents
Coef
f
i
cients
Cu
rren
t
Speed
ω
3.
LOCATI
O
N
OF MA
X
I
MU
M
P
O
WER
P
O
INT
At
t
h
e m
a
xi
m
u
m
power
p
o
i
n
t
we
have:
∙
0⇒
0
(32)
W
i
t
h
t
h
e
c
u
r
r
e
n
t
de
scri
be
d
by
eq
uat
i
o
n
(
0
1),
t
h
e v
o
l
t
a
ge
m
a
y
be e
x
p
r
esse
d
as:
∆
(33)
Back
su
b
s
titu
ti
o
n
of th
e
p
a
rtial d
e
riv
a
ti
v
e
of
with
resp
ect to
an
d
usi
n
g
_
fo
r
and
_
fo
r
gi
ve
s:
∆
1
_
∆
_
_
0
(3
4)
Wi
t
h
_
and
_
t
h
e
vol
t
a
ge
an
d
cu
rre
nt
at
m
a
xim
u
m
powe
r
.
Nu
m
e
rical
m
e
t
h
od
is app
lied
t
o
so
lv
e fo
r
_
u
s
in
g an in
itial guess g
i
v
e
n b
y
:
,
1
∆
or
,
∆
(35)
We
obt
ai
n
a
rel
a
t
i
on l
i
n
ea
r
wh
i
c
h can
be
e
x
p
r
essed
by
:
_
0
.9325
0
.0142
(36)
Once
_
i
s
fo
un
d,
_
m
a
y
be calcul
a
t
e
d
usi
n
g e
quat
i
o
n
(3
3
)
an
d t
h
us t
h
e cu
rr
ent
an
d
vol
t
a
ge
at
t
h
e m
a
xim
u
m
power
p
o
i
n
t
i
s
det
e
rm
i
n
ed a
s
a c
onse
q
uenc
e t
h
e m
a
xim
u
m
power
.
_
0.94
1
∆
.
∆
.
.
∆
1
(37)
At o
p
tim
al regim
e
, fo
r calc
u
lation t
h
e s
p
ee
d
refe
re
n
c
e, t
h
e assessm
en
t o
f
pow
er of
asyn
chr
ono
us
m
o
t
o
r i
s
gi
ve
n
by
:
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-86
94
I
J
PED
S
Vo
l. 4
,
No
. 2
,
Jun
e
2
014
:
24
1
–
25
5
24
8
_
3
_
_
(38)
The
rel
a
t
i
ons
g
ove
r
n
i
n
g t
h
e
o
p
erat
i
o
n
of
t
h
e
m
o
t
o
r-p
um
p [1
5]
.
(39)
∙
∙
(40)
≅
∙
(41)
(4
2)
By su
b
s
titu
ting equ
a
tio
ns
and
in
to Eq
u
a
tion (4
1).
∙
(43)
∙
(44)
Th
us t
h
e
rel
a
t
i
ons
(
4
1) t
h
e
n
(
3
8
)
bec
o
m
e
s:
∙
∙
(45)
_
∙
(46)
By r
ear
r
a
ng
ing th
e Equ
a
tio
n (4
5)
g
i
v
e
s:
0
(47)
Wi
t
h
,
,
0.94
1
∆
.
∆
.
.
∆
1
For eac
h
value
of ra
diation and t
h
e tem
p
erature, i.e. t
h
e va
lue of
. B
y
sol
v
i
n
g t
h
e E
quat
i
on (
4
7)
,
i
t
s
ro
ot
s,
gi
ve
t
h
e s
p
eed
of
i
n
stru
ctio
n co
rrespo
n
d
i
ng
t
o
t
h
e
op
ti
m
a
l o
p
e
r
a
tio
n of
t
h
e
gen
e
r
a
t
o
r PV
.
In
ad
d
ition
,
t
h
is sp
eed
d
e
p
e
n
d
s
o
n
t
h
e rad
i
ation
and
t
h
e am
b
i
en
t tem
p
eratu
r
e. Resu
lts ex
am
p
l
es are su
mmarized
in Figure
7 (re
d
stars).
4.
NEU
RAL
NE
TWOR
K BA
S
E
D
MPPT
Th
e
A
N
N
th
eo
r
y
, i
n
g
e
n
e
r
a
l, h
a
s
b
e
en
well-
d
i
scussed
i
n
th
e liter
a
tu
r
e
[1
6
]-[1
8
]
, an
d
a n
u
m
b
e
r
o
f
aut
h
ors
[
3
]
,
[
4
]
,
[1
9]
ha
ve
d
e
scri
be
d i
t
s
op
erat
i
o
n
i
n
t
h
e
PV sy
st
em
and f
o
r s
p
eed
co
nt
r
o
l
o
f
a
n
i
n
d
u
ct
i
o
n
m
o
tor. Ne
ural
network a
r
chi
t
ecture is s
p
ecified thro
u
g
h
fi
ndi
ng
t
h
e a
p
p
r
op
ri
at
e
so
l
u
tion
fo
r th
e
no
n-lin
ear
and c
o
m
p
l
e
x sy
st
em
s or t
h
e
ran
d
o
m
vari
ab
l
e
ones. Am
on
g i
t
s
t
y
pes, t
h
e
m
o
re wi
des
p
read
, im
port
a
n
t
and
usef
ul
back
p
r
opa
gat
i
o
n
net
w
o
r
k
.
T
h
e
fu
n
c
t
i
on a
n
d
res
u
l
t
s of a
n
a
r
t
i
f
i
c
i
a
l
neu
r
al
net
w
or
k a
r
e
det
e
rm
i
n
ed
by
i
t
s
archi
t
ect
ure
t
h
at
ha
s
di
ffe
r
e
nt
ki
nd
s. T
h
e
sim
p
l
e
r arc
h
itecture c
o
ntains
three layers
as
shown i
n
Figure
6.
The input layer receives t
h
e
extern data. T
h
e second la
yer, hi
dde
n layer, contains
se
veral hi
dden
neurons
whic
h
receive
data from
the input laye
r a
n
d
send them
to the thi
r
d layer,
out
put layer. T
h
is later
responds
t
o
the system
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
PED
S
I
S
SN
:
208
8-8
6
9
4
A N
e
ur
al
N
e
t
w
ork B
a
se
d M
P
PT Tec
h
ni
q
u
e
C
ont
r
o
l
l
e
r f
o
r
Ph
ot
ov
ol
t
a
i
c
P
u
m
p
i
n
g
Syst
e
m
(
M
oh
a
mme
d
Y
a
i
c
hi
)
24
9
We ca
n co
ncl
ude
u
n
l
i
m
i
t
e
d neu
r
al
n
e
t
w
o
r
k a
r
c
h
i
t
ect
ure
s
. The
m
o
re several
hi
d
d
e
n
l
a
y
e
rs an
d
neurons in eac
h layer are added; the
m
o
re com
p
lex they
becom
e
. The network is fu
lly connected
whic
h a bias
sig
n
a
l is co
upled
to all
n
e
uro
n
s
in th
e
h
i
dd
en layer t
h
rou
gh a
weigh
t
. Th
e
realizatio
n
o
f
th
e b
a
ck
p
r
op
ag
ation
n
e
twork
is b
a
sed
o
n
two
m
a
in
po
in
ts: lear
n
i
ng
an
d kno
w
l
ed
g
e
. Th
is r
e
sear
ch w
a
s app
lied
by th
e
use o
f
si
gm
oi
d funct
i
on as an
act
i
v
at
i
on fu
n
c
t
i
on i
n
or
de
r t
o
cal
cul
a
t
e
t
h
e hi
dde
n l
a
y
e
r out
put
an
d t
h
e
l
i
n
ear
fun
c
tion
to calcu
late th
e
ou
tpu
t
.
In
our st
udy
of the effect of
ne
t
w
o
r
k a
r
c
h
i
t
ect
ure
on
l
earni
ng
, we
ha
ve c
h
ose
n
arc
h
itecture with
one
hi
d
d
en l
a
y
e
r
.
We ha
ve c
h
an
ged t
h
e n
u
m
b
er o
f
ne
ur
o
n
s i
n
hi
dde
n l
a
y
e
r
;
t
h
i
s
archi
t
ect
ure i
s
s
u
m
m
ar
i
zed i
n
Tabl
e
2. B
y
5
neu
r
ons
i
n
hi
d
d
en
l
a
y
e
r, t
h
e
r
e
gres
si
o
n
c
o
ef
f
i
ci
ent
R
-
sq
ua
re
(R
2
)
i
n
cr
eases
to
0.999
9.
Tabl
e 2.
E
f
f
ect
o
f
num
ber of
hi
d
d
en
l
a
y
e
rs o
n
l
ear
ni
n
g
Nu
m
b
er
of neur
ons in hidden lay
e
r
R
2
3
0.
9984
5
0.
9999
8
0.
9991
9
0.
9989
11
0.
9991
Fi
gu
re 5 s
h
ow
s t
h
e pr
o
pose
d
AN
N t
o
p
o
l
o
g
y
,
i
t
use
m
u
l
t
ilay
e
r perce
p
t
r
o
n
t
y
pe net
w
or
k co
nsi
s
t
s
o
f
an i
n
p
u
t
l
a
y
e
r,
one hi
dde
n l
a
y
e
r and a
n
o
u
t
put
l
a
y
e
r wi
t
h
si
gm
oi
dal
ty
pe t
r
ansfe
r
f
unc
t
i
on. T
h
e com
posi
t
e
net
w
or
k use
s
t
w
o
neu
r
ons at
t
h
e i
n
p
u
t
,
o
n
e f
o
r t
h
e
radi
at
i
o
n an
d t
h
e ot
her
for t
h
e am
bi
ent
t
e
m
p
erat
ure
,
whi
c
h
i
s
norm
a
l
i
zed and t
h
en
pul
se
neu
r
o
n
s i
n
t
h
e hi
d
d
en l
a
y
e
r
.
The
out
put
l
a
y
e
r has o
n
l
y
one
ne
ur
on
fo
r t
h
e
cont
rol
DC/AC
fre
q
u
ency
.
Th
e pro
c
edu
r
e d
e
scrib
e
d
in
th
e prev
i
o
u
s
sectio
n
will b
e
utilized
to
g
e
n
e
rate train
i
n
g
d
a
t
a
fo
r
ANN-
base
d M
P
P
T
.
Fi
gu
re
5.
Ne
ur
al
net
w
or
k t
o
p
o
l
o
gy
(
2
-
5
-1
)
f
o
r
spee
d
ge
ner
a
t
e
d
The
outputs are calculated
by
th
e fo
llowing
eq
u
a
tion
:
∑
∙
with
:
1
∙
∙
5
(48)
(49)
∑
∙
(50)
The u
pdat
e
of
is done
accordi
n
g to t
h
e
rule
of
delta:
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-86
94
I
J
PED
S
Vo
l. 4
,
No
. 2
,
Jun
e
2
014
:
24
1
–
25
5
25
0
Δ
and
Δ
(51)
Whe
r
e:
,
Δ
and
Δ
(52)
Wi
t
h
is th
e step
o
f
train
i
ng
and
,
∙
,
(53)
∙
,
∙
∙
(54)
Algo
rith
m
architectu
r
e of t
h
e
b
ack pro
p
a
g
a
tio
n im
p
lies th
e fo
llowing
step
s as sh
own
in Fi
g
u
re
6
.
The t
e
st
s resul
t
s of A
N
N
-
base
d M
PPT at
t
h
e fi
ve ope
rat
i
n
g
case (pr
o
p
o
se
d l
earni
ng e
x
a
m
pl
es) were
fo
u
nd t
h
at
t
h
e
out
p
u
t
spee
d
resp
o
n
se o
f
t
h
e
m
o
t
o
r-
pum
p
with
an
alytic
al resu
lts is h
i
g
h
e
r th
an
th
e in
pu
ts
radi
at
i
o
n an
d a
m
bi
ent
t
e
m
p
erat
ure as s
h
ow
n
i
n
Fi
g
u
re
7
(a
)-
(e)
whi
c
h s
h
ows
t
h
e
out
put
of
A
NN
at
fi
v
e
case
and s
h
ows t
h
e cor
r
es
po
n
d
i
n
g dat
a
wi
t
h
t
h
e a
n
al
y
t
i
cal
resul
t
s
. Al
so
, t
h
e M
PPT m
e
t
hod w
a
s val
i
d
at
ed
fo
r t
h
e
case
37
an
d co
mp
a
r
ed w
ith th
e
d
a
ta
,
th
e
y
w
e
r
e
fo
und
to
correlate v
e
ry well (Test o
f
g
e
n
e
ralizatio
n
i
n
Fig
u
re
8
)
. Th
is m
ean
s th
at th
e n
e
twork can
be op
er
ated
at an
y cond
itio
n
s
cli
m
a
tic.
Fi
gu
re 6.
A
r
chi
t
ect
ure of
t
h
e back propagation
1:
1
;∙∙∙
∙∙∙∙∙
∙;
1
;∙
∙∙∙;
0.004
,
0
.01,
1
,
:
2:
1
;∙∙∙∙
∙∙∙∙∙;
And the example
gi
ven one by
one
To initi
a
li
z
e
the wei
ghts o
f
the
ne
twork
by
random value
s
0
;
25
2
;
31
5
Begin
Ye
s
No
∆
.
5
1
1
2
End
,
,
,
,
cal
c
ulate the
output
Of
the
syn
a
pt
ic
weights
ane
w
0
,
1
.
5
3
,
.
5
4
1
Evaluation Warning : The document was created with Spire.PDF for Python.