Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
Vol.
5, No. 6, Decem
ber
2015, pp. 1304~
1
310
I
S
SN
: 208
8-8
7
0
8
1
304
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
/
IJECE
A Neuro-fuzzy Approach for P
redicting Load Peak Profile
Abdellah
Drai
di, Djamel Labed
Laboratoire de g
e
nie électrique
d
e
Constan
tine
,
D
e
part
em
ent
of
electr
i
cal engineer
ing,
University
of
Co
nstantine 1, Algeria
Article Info
A
B
STRAC
T
Article histo
r
y:
Received Feb 13, 2015
Rev
i
sed
Ju
l 3
,
2
015
Accepte
d
J
u
l 28, 2015
Load for
ecas
t
i
ng
has
m
a
n
y
appl
i
cat
ions
for powe
r
s
y
s
t
em
s
,
in
clu
d
ing ener
g
y
purchasing and
generation, load sw
itching, contract evaluation
,
and
infras
t
ruc
t
ure
d
e
velopm
ent
.
Lo
ad fore
cas
t
i
ng i
s
a com
p
l
e
x
mathem
at
ica
l
process char
acterized b
y
r
a
ndom data
a
nd a multitude of
input v
a
riab
les.To
s
o
lve load
fore
c
a
s
ting,
two diff
e
r
ent
approach
es
are us
ed
, th
e
tra
d
itiona
l a
nd
the int
e
ll
igent o
n
e; int
e
ll
igent s
y
stem
s have pro
v
ed their
effi
cie
n
c
y
in loa
d
forecas
t
i
ng dom
ain. Adapt
i
ve ne
uro-fuzz
y infer
e
nce s
y
s
t
em
s
(ANF
IS
) are a
combination
of two intellig
ent
techniques wher
e
we can
get neur
al n
e
tworks
and fuz
z
y
log
i
c
advan
t
ages sim
u
ltan
e
ousl
y
. In
t
h
is paper
,
we
will for
e
c
a
st
night lo
ad p
eak
of Algeri
an pow
er s
y
st
em
using
m
u
ltivari
a
te
inp
u
t ad
aptiv
e
neuro-fuzzy
in
f
e
rence s
y
stem
(ANFI
S
) introducing the ef
fect of the
temperatur
e
and
ty
p
e
of
th
e da
y
as input v
a
riables.
Keyword:
Loa
d
forecasting
N
e
uro
-
f
u
zzy netw
or
k
Power system
s
Tem
p
erature
Ty
pe of
t
h
e da
y
Copyright ©
201
5 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
:
Ab
del
l
a
h D
r
ai
di
,
Lab
o
rat
o
i
r
e
de
geni
e él
ect
ri
qu
e de C
o
nst
a
nt
i
n
e,
Depa
rtem
ent of elect
rical engineering,
Un
i
v
ersity
of Co
n
s
tan
tin
e 1
,
Alg
e
ria.
Em
a
il: d
r
aid
i
_ab
d
e
llh@u
m
c
.
e
d
u
.d
z
1.
INTRODUCTION
The Alge
rian
econom
y
is linke
d st
ro
ng
ly to
fo
ssil en
erg
y
m
a
rk
et;
as the prices of energy are
fl
uct
u
at
i
n
g t
h
e
necessi
t
y
of
d
e
vel
o
pi
n
g
ne
w
p
o
we
r c
o
ns
u
m
pti
on st
rat
e
gi
es ri
ses,
t
h
i
s
c
oul
d
be ac
hi
ev
ed
by
optim
izing the
basic
power sy
ste
m
opera
ti
ons including: load
flow, ec
onom
ic dispatch a
n
d loa
d
forecas
ting.
Short-te
rm
load forecasting i
s
i
m
portant for pe
rform
i
ng many power
utili
ty functions, including
gene
rat
o
r
u
n
i
t
com
m
itm
ent
,
hy
d
r
o-t
h
erm
a
l
co
or
di
nat
i
o
n,
sh
ort
-
t
e
rm
m
a
i
n
t
e
nan
ce,
f
u
el
al
l
o
cat
i
on,
po
we
r
in
terch
a
ng
e, tran
saction
ev
al
u
a
tio
n, as well as n
e
twor
k a
n
al
y
s
i
s
fu
nct
i
o
ns, sec
u
ri
t
y
an
d l
o
ad
fl
o
w
st
udi
es
,
cont
i
n
ge
ncy
pl
anni
ng
, l
o
a
d
sh
eddi
ng
, a
n
d
l
o
a
d
sec
u
ri
t
y
st
rat
e
gi
es [
1
]
.
A va
riety of
m
e
thods a
n
d ideas ha
ve
bee
n
tried
for loa
d
forecasti
ng s
i
nce m
a
ny decades; va
rying
from
classical to a
r
tificial intelligence
ones
,
t
hos
e m
e
thods
are
discusse
d i
n
the
ne
xt title.
In loa
d
forecas
ting, t
h
e
problema
tic resides in three as
pects
:
1)
First, e
r
ror, tha
t
m
eans, to ha
ve a m
i
nim
u
m
differe
nce between forecasted and
real value
s
;
2)
secon
d
, ex
ecu
t
io
n
tim
e, esp
ecially with
sho
r
t term
LF,
wh
ere
redu
cing
sim
u
latio
n
ti
m
e
is
essen
tial;
3)
Th
ird
,
ex
tern
al
p
a
ram
e
ters affectin
g
LF, su
ch
as
weath
e
r
variab
ility an
d
,
i
n
long
er term
, cli
m
at
e
v
a
riab
ility; th
e
g
r
owth of
p
o
p
u
l
ation
is
one of tho
s
e
p
a
ra
m
e
ters alo
ngsid
e
with
th
e
eco
no
m
i
c an
d so
cial
welfare of t
h
e
p
opu
latio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
A Neu
r
o
-
f
u
zzy
App
r
oa
ch
for
Pred
ictin
g Lo
ad
Pea
k
Pro
file
(Ab
d
e
llah
Dra
i
d
i
)
1
305
2.
CO
NVE
NTI
O
N
A
L A
N
D
ARTIF
ICI
A
L
METHO
D
S
FOR
LOA
D
F
O
REC
A
STI
N
G
2.
1.
Tra
d
itiona
l Approa
ches
2.
1.
1.
Time Series
Methods
These m
e
t
hods
t
r
eat
t
h
e l
o
a
d
pat
t
e
rn a
s
a t
i
m
e seri
es si
gn
al
wi
t
h
k
n
o
w
n
seaso
n
al
,
week
l
y
and
dai
l
y
p
e
ri
o
d
i
cities. Th
ese
p
e
ri
o
d
i
cities g
i
v
e
a
ro
ugh
pred
ictio
n
o
f
th
e lo
ad
at th
e g
i
v
e
n
season
,
d
a
y of th
e
week
and
t
i
m
e
of t
h
e da
y
.
The di
f
f
ere
n
ce bet
w
ee
n t
h
e
pre
d
i
c
t
i
on a
nd the actual load can be c
o
nsidere
d
as a stoc
hastic
p
r
o
cess (r
andom
sig
n
a
l)
. Th
e
tech
n
i
qu
es u
s
ed
f
o
r
th
e an
aly
s
is o
f
th
is
r
a
n
d
o
m
sig
n
a
l ar
e:
1)
Kalm
an
Filters Meth
od
;
2)
Bo
x Jenk
ins M
e
th
od
;
3)
Regressi
on P
r
ocesses;
4)
Sp
ectral Exp
a
nsio
n T
echn
i
qu
e.
[2
,3
,4
]
2.
1.
2.
Regre
ssion
Based me
thods
T
h
e gener
a
l
p
r
o
c
ed
ur
e f
o
r t
h
e r
e
gr
e
ssio
n
ap
p
r
o
a
ch
i
s
:
1)
T
o
s
e
lect
th
e
prope
r a
n
d/
or a
v
ailable
w
eath
e
r
v
a
riable
s;
2)
A
s
s
u
m
e
b
a
sic f
u
nctio
nal ele
m
en
ts;
3)
Find pro
p
e
r
c
o
ef
f
i
cie
n
ts for t
h
e linear com
b
i
n
ation of the
a
s
s
u
m
e
d
bas
i
c
f
u
nctio
nal ele
m
ents
.
Sin
ce tem
p
eratu
r
e is th
e mo
st im
p
o
r
tan
t
in
fo
rm
atio
n
o
f
all weath
e
r
v
a
riab
les, it is u
s
ed
m
o
st
co
mm
o
n
l
y in
t
h
e reg
r
ession
ap
pro
ach. Howev
e
r, ad
d
itio
n
a
l v
a
riab
les su
ch
as
hu
m
i
d
i
ty, win
d
v
e
lo
city an
d
cl
ou
d y
i
el
ds b
e
t
t
e
r resul
t
s
. T
h
e f
u
nct
i
onal
r
e
l
a
t
i
onshi
p bet
w
een l
o
ad
an
d
weat
he
r va
ri
a
b
l
e
s h
o
w
eve
r
i
s
n
o
t
st
at
i
onary
but
depe
n
d
s o
n
s
p
a
t
i
o
t
e
m
poral
el
e
m
ent
s
.
2.
2.
Intellig
ent Sy
stems
An
i
n
tellig
en
t syste
m
can
b
e
d
e
fin
e
d
as a syst
e
m
th
at ex
h
i
b
its in
tellig
en
ce in
cap
t
uring
and
p
r
o
cessi
n
g
i
n
fo
rm
atio
n
.
Practically sp
eak
ing
,
an
in
tellig
en
t system
is t
h
e
o
n
e
, wh
ich em
p
l
o
y
s artificia
l
in
tellig
en
ce tech
n
i
q
u
e
s to fu
lfi
ll so
m
e
o
r
all of its co
m
p
u
t
atio
n
a
l
req
u
i
rem
e
n
t
s.
2.
2.
1.
Artificial Neu
r
al
Ne
tw
or
ks
(A
NN
)
Th
e ANN is cap
ab
le to
p
e
rfo
rm
n
o
n
-lin
ear
m
o
d
e
lin
g
and ad
ap
tation
.
It
d
o
e
s no
t requ
i
r
e fu
n
c
tion
a
l
rel
a
t
i
ons
hi
p
be
t
w
een l
o
a
d
a
n
d weat
her
vari
abl
e
s i
n
ad
va
nce.
The ANN can
learnf
rom
expe
rience
, ge
neralize
fr
om
previ
o
us exam
pl
es t
o
newo
nes, a
b
st
rac
t
s essent
i
a
l
charact
eri
s
t
i
c
s fr
o
m
i
nput
co
nt
ai
ni
n
g
i
rrel
e
vant
dat
a
.
The
ANN
gi
ve
s m
o
re precise
forecast as c
o
m
p
ared to c
onventio
nal tec
h
niques
[2].
2.
2.
2.
R
u
l
e
Ba
s
e
d
Ex
p
e
rt
Sy
s
t
e
m
s
An
ex
pert syste
m
is a co
m
p
u
t
er p
r
og
ram
,
wh
ich
h
a
s th
e ab
ility to
act as
a k
n
o
w
led
g
e
ex
p
e
rt. Th
is
m
eans t
h
i
s
pr
o
g
ram
can
reas
o
n
, e
x
pl
ai
n a
n
d
have
i
t
s
k
n
o
wl
edge
base e
x
pa
nde
d
as
new
i
n
fo
rm
ati
on
bec
o
m
e
s
available to it.
The load-forec
ast
m
odel can
be
built usi
n
g the
knowledg
e about
the
loa
d
forecast dom
ai
n
from
an e
x
pert i
n
t
h
e field. T
h
e
knowle
dge
e
ngi
neer e
x
tract
s t
h
i
s
k
n
o
wl
e
dge
f
r
o
m
t
h
e l
o
ad
f
r
e
que
ncy
d
o
m
a
in.
Thi
s
kn
o
w
l
e
d
g
e i
s
represe
n
t
e
d as
f
act
s and r
u
l
e
s usi
n
g t
h
e fi
rst
pre
d
icate logic
to repre
s
ent the facts and IF-THE
N
pr
o
duct
i
o
n
rul
e
s. S
o
m
e
of t
h
e rul
e
s
d
o
not
chan
ge
o
v
er t
i
m
e
, som
e
chan
ges
very
sl
owl
y
;
whi
l
e
ot
hers
cha
nge
continuously and he
nce t
h
ey
are
u
p
d
a
ted time to
ti
m
e
[5
].
2.
2.
3.
Fuz
z
y
System
s
Fuzzy
set
s
we
re i
n
t
r
od
uce
d
t
o
re
prese
n
t
an
d m
a
ni
pul
at
e dat
a
an
d i
n
f
o
r
m
at
i
on t
h
at
po
ssesses n
o
n
-
statistical u
n
certain
ty. Fu
zzy sets are a
g
e
n
e
ralizatio
n
o
f
con
v
e
n
tion
a
l set th
eory th
at
was
in
trodu
ced
as a n
e
w
way to
represen
t v
a
gu
en
ess i
n
th
e d
a
ta. It i
n
tro
d
u
ces
v
a
gu
en
ess (with
t
h
e aim
o
f
redu
cing
co
m
p
lex
ity) b
y
eli
m
in
atin
g
t
h
e sh
ar
p bou
nd
ary b
e
tw
een
th
e
me
m
b
er
s of
the class fr
o
m
n
o
n
m
e
m
b
er
s [6
],
[7
]
In our pa
per, we
will
propose
a
hybrid method in t
h
e solution
of l
o
a
d
fo
recasting,
whic
h is a
com
b
i
n
at
i
on
of
ne
ural
net
w
or
ks a
n
d
f
u
zzy
l
o
gi
c, t
h
i
s
m
e
t
h
o
d
i
s
cal
l
e
d a
d
a
p
t
i
v
e
neu
r
of
uz
zy
i
n
fere
nce sy
st
em
s
(ANFIS) and
it will b
e
d
i
scussed
later on
the article. Our
pu
rpo
s
e is to
red
u
ce ex
ecu
tio
n ti
m
e
an
d
erro
rs th
us
to ha
ve a
faste
r
and trustworthy forecast c
o
m
p
ari
n
g with ot
her m
e
thods
use
d
in the
field.
3.
AD
APTI
VE
NEU
R
O
-
FUZ
Z
Y
INFERE
N
C
E S
Y
STEM
S (
A
N
F
IS
)
3.
1.
What is
ANFI
S?
Jan
g
et
al
propos
ed A
N
F
I
S
archi
t
ect
u
r
e i
n
19
93 [
8
]
.
The
acrony
m
ANFIS de
ri
ves i
t
s
nam
e
from
adapt
i
v
e
ne
ur
o-
fuzzy
i
n
fere
nce sy
st
em
. Usi
ng a
gi
ve
n i
n
p
u
t
/
o
ut
p
u
t
da
t
a
set
,
AN
FIS
con
s
t
r
uct
s
a
fuzz
y
i
n
fere
nce sy
st
em
(FIS)
w
h
ose m
e
m
b
ershi
p
f
u
nct
i
o
n param
e
t
e
rs are t
une
d (a
dj
u
s
t
e
d)
usi
n
g ei
t
h
er a
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJECE
Vol. 5, No. 6, D
ecem
ber
2015 :
1304 –
1310
1
306
b
ackpr
op
ag
atio
n algo
r
ith
m
alo
n
e
or
i
n
com
b
in
atio
n
w
i
t
h
a least sq
u
a
res typ
e
o
f
m
e
t
h
od
. Th
is ad
just
m
e
n
t
allo
ws y
o
ur
fuzzy syste
m
s to
learn
fro
m
th
e d
a
ta th
ey are mo
d
e
ling
.
[9
]
AN
FIS i
s
an
adapt
i
v
e
net
w
or
k w
h
i
c
h al
l
o
ws t
h
e im
pl
em
ent
a
t
i
on of
neu
r
al
net
w
o
r
k t
o
p
o
l
o
gy
,
to
g
e
th
er
with fu
zzy l
o
g
i
c
[10
]
, [11
]
. An
ANFIS
stud
y co
m
p
iles th
ese two
m
e
t
h
od
s and
u
tilizes th
e
charact
e
r
i
s
t
i
c
s of
bot
h m
e
t
hods.
Al
so,
A
N
F
I
S gat
h
ers
b
o
t
h
t
h
e neu
r
al
net
w
o
r
k a
n
d f
u
zz
y
l
ogi
c, an
d i
s
abl
e
t
o
treat non line
a
r and c
o
m
p
lex problem
s
[12]. ANF
I
S is
a class of adaptive m
u
ltilayer feeding forwa
r
d
net
w
or
ks,
w
h
i
c
h i
s
fu
nct
i
o
nal
l
y
equi
val
e
nt
t
o
a f
u
zzy
i
n
fere
nce sy
st
em
.
3.
2.
ANF
IS Architecture
Acco
r
d
i
n
g t
o
J
a
ng
an
d al
[
8
]
,
[
13]
t
h
e
gl
ob
a
l
st
ruct
u
r
e
of
a
d
apt
i
v
e
n
e
u
r
o
-
fuzzy
sy
st
em
s i
s
sh
ow
n i
n
Fi
gu
re 1:
Figure 1.
ANFIS system
str
u
cture
L
ayer
1
:
E
v
er
y
no
de
i
i
n
th
is layer is an adap
tiv
e
n
o
d
e
wit
h
a
no
d
e
fun
c
tio
n.
(1
)
is th
e m
e
m
b
ersh
ip grad
e
o
f
A
i
and it s
p
ecifies
the
de
gree t
o
whic
h t
h
e
give
n i
n
put
x
(o
r
y
) satifies
th
e qu
an
tifier
A
i
.
or
(
2
)
Whe
r
e
,
,
is th
e
param
e
ter set.
L
ayer
2
:
I
n
th
i
s
layer
th
e
ou
tpu
t
is th
e pr
oduct o
f
all th
e in
co
m
i
n
g
sign
als:
,
1
,2
(3)
Each n
ode
o
u
t
put
re
prese
n
t
s
t
h
e
firing s
t
rength
of a
r
u
le.
L
ayer 3:
The
i
th
no
d
e
calcu
lates th
e ratio
of th
e
i
th
rule'
s
firing st
ren
g
th to t
h
e sum
of all rules'
firin
g
strengths:
,
1
,2.
(4)
Ou
t
p
u
t
s of th
is layer are
called
norm
aliz
ed firing s
t
rengthes
.
L
ayer
4
Eve
r
y
n
ode
i
in
t
h
is l
a
yer is an ad
aptiv
e no
d
e
wit
h
a no
d
e
fun
c
tion
:
(
5
)
{
p
i
, q
i
, r
i
} is the p
a
ram
e
ter set
of th
is
no
d
e
wh
ich
are
referred
to as
c
o
nse
q
uent par
a
me
ters
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
A Neu
r
o
-
f
u
zzy
App
r
oa
ch
for
Pred
ictin
g Lo
ad
Pea
k
Pro
file
(Ab
d
e
llah
Dra
i
d
i
)
1
307
L
ayer 5:
The
no
des
of t
h
i
s
l
a
y
e
r com
put
es t
h
e ove
ral
l
o
u
t
p
ut
as t
h
e s
u
m
m
a
t
i
on of a
l
l
i
n
com
i
ng
si
ngal
s
:
∑
∑
∑
(
6
)
3.
3.
ANFIS Co
mputationa
l Co
mplex
i
ty
Diffe
re
nt layers cha
r
acteristics are
shown in
Table
1:
Table 1.
Layers
cha
r
acteristics
L
a
y
e
r
#
L
-
Ty
pe
# Nodes
# Par
a
m
L
0
input
n
0
L1
v
a
lu
es
(p
•n
)
3
•
(p
•n
)=
|S1
|
L2
Ru
les
p
n
0
L
3
norm
a
lize
p
n
0
L
4
L
i
n.
Funct
p
n
(n
+1
)•p
n
=
|
S
2
|
L
5
su
m
1
0
ANFIS uses two
sets
of
p
a
rameters: S1 and
S2
:
1)
S1
rep
r
esen
ts t
h
e
fu
zzy
p
a
rtitio
n
s
used in
t
h
e ru
les LHS
1
,
,
,
,
,
,…,
,
,
…,
,
,
(7
)
2)
S2
rep
r
esen
ts t
h
e co
efficien
ts
o
f
th
e lin
ear
fun
c
tio
ns in th
e
ru
les RHS
2
,
,…
,
,…,
,
,…,
(8
)
ANFIS uses a t
w
o-pass learni
ng cycle
1)
For
w
a
r
d
pass:
S1 i
s
fi
xe
d a
nd
S2 i
s
c
o
m
put
ed
usi
n
g a L
east
Squa
re
d
Err
o
r
(LSE
) al
go
ri
t
h
m
(O
ff
-line Lea
r
n
i
ng
).
2)
B
ackwa
rd
pa
s
s
:
S2 i
s
fi
xe
d
and
S
1
i
s
c
o
m
put
e
d
usi
n
g a
gra
d
i
e
nt
desce
n
t
al
g
o
ri
t
h
m
(usu
al
l
y
B
ack-
p
r
o
pagat
i
on
) [
1
4]
.
3.
4.
B
a
si
c Fl
ow
Di
agr
am
o
f
Co
mput
ati
o
ns
i
n
A
N
FIS
The
A
N
FI
S E
d
i
t
o
r
GU
I
(M
A
TLAB
)
ap
pl
i
e
s
f
u
zzy
i
n
fe
re
nc
e t
echni
qu
es t
o
dat
a
m
odel
i
n
g
;
basi
c fl
ow
di
ag
ram
of co
m
put
at
i
ons i
n
AN
FIS
i
s
gi
ve
n i
n
Fi
g
u
r
e
2:
Fi
gu
re
2.
B
a
si
c fl
o
w
di
ag
ram
of
com
put
at
i
o
n
s
i
n
AN
FI
S
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJECE
Vol. 5, No. 6, D
ecem
ber
2015 :
1304 –
1310
1
308
4.
OVERVIEW
ON T
H
E AL
GERIAN
POWER SYSTE
M
LOAD PATTERN
The elaboration of the
Alge
rian
load forec
a
sting is becoming incr
easingly difficult because of the
u
n
c
ertain
ties
related
to th
e
facto
r
s u
s
ed in its prep
ar
at
i
o
n,
especi
al
l
y
t
hos
e l
i
nked t
o
cons
um
pt
i
on habi
t
s
ch
ang
i
ng
. In
Alg
e
ria, th
is ch
an
g
e
is du
e to
an
in
creas
i
n
g
sen
s
itiv
ity o
f
th
e co
n
s
u
m
er to
t
h
e te
m
p
eratu
r
e rise
and c
o
n
s
eq
ue
n
t
l
y
it
i
s
refl
ect
ed t
h
r
o
ug
h
out
t
h
e cha
n
ge
s on
annual and dai
l
y load
summe
r curves
. The a
n
nual
consum
ption peak, which hist
orically
has be
en reache
d
in
the winter,
m
o
ve
d to the
sum
m
erin
2009,
when the
summ
er peak
has excee
de
d the winte
r’s
by
5.1%
(Fi
g
ure
3).
Fi
gu
re 3.
Hi
st
o
r
i
cal
ev
ol
ut
i
o
n of
m
a
xim
u
m
dem
a
nd fr
om
20
00
t
o
A
u
g
u
st
2
0
1
1
[
15]
The ave
r
age s
p
ecific consum
ption per low voltage
cust
om
er has increased
to
26
23k
W
h
i
n
20
09
.
If
the ave
r
a
g
e c
o
nsum
ption of
Alge
rian
ho
m
e
s h
a
s in
creased, it is in
teresting
to
no
te t
h
at th
is in
crease is
d
r
i
v
en
m
a
i
n
l
y
by
so
ut
h c
u
st
om
ers w
h
o
re
pre
s
ent
j
u
st
1
0
%
o
f
th
e t
o
tal nu
m
b
er of
lo
w
vo
ltag
e
custo
m
ers.
Thi
s
pec
u
l
i
a
ri
t
y
i
s
expl
ai
ne
d
by
t
h
e m
a
ssive use
o
f
ai
r
con
d
i
t
i
oni
n
g
,
gi
ve
n t
h
e s
p
ec
i
a
l
cl
im
at
e of
sout
h re
gi
o
n
s
whi
c
h i
s
cha
r
a
c
t
e
ri
zed by
hi
g
h
t
e
m
p
erat
ures
du
ri
n
g
se
veral
m
ont
hs
of t
h
e
y
ear (Fi
g
u
r
e 4
)
. T
h
e
avera
g
e c
o
nsu
m
pti
on
of c
u
st
om
ers i
n
t
h
e
n
o
rt
her
n
re
gi
on
i
s
expl
ai
ned
b
y
po
p
u
l
a
t
i
on
d
e
nsi
t
y
. O
v
er
5
2
%
o
f
th
e low
v
o
ltage cu
st
o
m
er is lo
cated
no
rt
h
o
f
th
e co
un
try.
Fi
gu
re
4.
C
o
m
p
ari
s
on
bet
w
ee
n l
o
a
d
c
u
rves
o
f
t
w
o
day
s
wi
t
h
m
a
xim
u
m
dem
a
nd
[1
6]
5.
THE DATA
SET
The
dat
a
used
f
o
r
A
N
F
I
S l
e
a
r
ni
n
g
, c
h
ec
ki
n
g
and
t
e
st
i
n
g
i
s
t
a
ken
f
r
om
:
1)
SO
NEL
GAZ
l
o
ad
cu
rv
e hi
st
ory
dat
a
base
[
17]
whi
c
h c
o
n
t
ai
n dai
l
y
l
o
ad
cur
v
es
wi
t
h
d
a
y
a
n
d
ni
g
h
t
pea
k
s
.
2)
M
a
xi
m
u
m
t
e
m
p
erat
ure
o
f
In Sal
e
h (re
gi
on
o
f
Tam
a
nraset) the
hottest area in Al
geria and
min
i
m
u
m te
mp
er
atu
r
e of
S
e
tif
the coldest area in Alge
ria, this gives
us a
n
ave
r
age temperat
ure of the
whole
co
un
tr
y [
1
8
]
5.
1.
ANF
IS Architecture
We
have
u
s
ed
fo
r t
r
ai
ni
n
g
Load curves of
July 18,
2010 and J
uly 27,
2009
Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE
ISS
N
:
2088-8708
A Neu
r
o
-
f
u
zzy
App
r
oa
ch
for
Pred
i
ctin
g Lo
ad
Pea
k
Pro
file
(Ab
d
e
llah
Dra
i
d
i
)
1
309
Inpu
ts: [6
64x
6]
m
a
trix
, an
exa
m
p
l
e is g
i
v
e
n
in
Tab
l
e 2.
Tabl
e
2.
Sam
p
l
e
of
i
n
put
m
a
tri
x
Day
M
a
x T
e
m
p
M
i
n T
e
m
p
W
o
r
k
ing day
Friday Saturday
Peak Generation
01/03/
201
0
31
6
1
0
0
6087
M
W
13/05/
201
1
33
8
0
1
0
6045
M
W
Fi
gu
re
5 s
h
ows
t
h
e
AN
FIS
st
r
u
ct
u
r
e:
Figure
5. ANFIS a
r
chitecture
1)
Layer 1
:
con
t
ain
s
t
h
e inpu
t m
a
trix
.
2)
Layer 2
:
wh
ich calcu
lates th
e
me
m
b
ersh
ip
v
a
lu
e fo
r prem
ise
p
a
ram
e
ters, h
e
re,
we assign
fo
r
me
m
b
ersh
ip
fun
c
tio
ns for th
e
first two
i
n
pu
ts and
two
m
f’s
fo
r th
e
rest three in
pu
ts.
3)
Layer
3
:
wh
ich calcu
late f
i
r
i
ng
str
e
ng
th
of
the 12
8 ru
les.
4)
Lay
e
r 4:
w
h
ich
n
o
rm
alizes all
firin
g
stre
n
g
ths
.
5)
Layer 5
:
calcu
l
ates th
e
o
v
e
rall su
m
o
f
t
h
e in
co
m
i
n
g
sign
als,
th
e ou
tpu
t
represen
ts th
e lo
ad
p
eak
pre
d
i
c
t
e
d
by
t
h
e m
odel
.
Fig
u
r
e
6
.
Respo
n
s
e
o
f
Ou
tpu
t
f
r
o
m
Mar
c
h
20
10
t
o
Febr
u
a
ry 2
012
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJECE
Vol. 5, No. 6, D
ecem
ber
2015 :
1304 –
1310
1
310
Figure 6 shows ANFIS output (pea
k ge
neration forecaste
d represe
n
ted
by red crosses
)
vers
us input
(real
pea
k
ge
nerat
i
o
n re
pre
s
ent
e
d by
bl
u
e
ci
rcl
e
s).
W
e
can see fro
m
Fi
gure 6 t
h
at
AN
FIS t
r
ai
ni
ng i
s
satisfactory, t
h
at
m
eans outputs are
gene
rally close
t
o
targ
et
s with so
m
e
excep
tio
ns for some p
o
i
n
t
s.
5.
2.
N
e
t
w
ork Test
ing
We
have
u
s
ed
M
a
rsh
2
0
1
2
da
t
a
t
o
t
e
st
t
h
e
ne
ural
net
w
or
k;
t
h
e
resul
t
i
s
s
h
o
w
n
i
n
Fi
g
u
re
7
.
Fig
u
r
e
7
.
N
e
tw
or
k testin
g usin
g
Mar
c
h 2012
d
a
ta
6.
CO
NCL
USI
O
N
We can de
duc
e
from
results shown in Fi
gure 6
a
nd Fi
gure 7 that our forecast using ANFIS was
acceptable. To have m
o
re accurate an
d excellent forecasting
we m
u
st use
m
o
re input dat
a
set to
have a
good
neu
r
al
net
w
or
k
t
r
ai
ni
n
g
.
Our
purpose i
s
to im
ple
m
ent ar
tificial intelligence techni
que
s in
loa
d
forecasting especially
for
Al
ge
ri
an
po
w
e
r g
r
i
d
.
As
w
eat
her a
nd es
peci
al
l
y
t
e
m
p
erat
ure
re
prese
n
t
s
t
h
e m
a
i
n
param
e
t
e
r i
n
fl
uenci
n
g
Alge
rian cons
um
ption, the necessity
of developing a m
odel for pea
k
fo
recasting rises.
The ot
her pa
ra
meter
tak
e
n in
to co
n
s
id
eration
is ty
pe of t
h
e
d
a
y
wh
ere we re
m
a
r
k
ed thr
ee typ
e
s of
d
a
ys; wor
k
i
n
g days, Fr
id
ays th
at
are real
wee
k
e
nds
a
n
d Saturdays that
are
for som
e
Algeria
n
a worki
n
g day
s
.
We
have
, s
u
cc
essfully, i
n
troduced the e
ffect
of the
t
e
m
p
er
at
ure a
n
d t
y
pe
of t
h
e
day
as i
n
p
u
t
m
a
t
r
i
x
use
d
i
n
t
h
e
pr
o
cess o
f
t
r
ai
ni
n
g
,
so
we
h
a
ve
a
fast
an
d
rel
i
a
bl
e l
o
ad
f
o
rec
a
st
i
n
g
usi
n
g
ANF
I
S
.
REFERE
NC
ES
[1]
Abdel-Aal R
.
E
,
“
M
odeling, For
ecast
i
ng El
ec
tric
Dail
y
Pe
ak Lo
a
d
s Using Abductive Ne
tworks,”
Elsevi
er, Oxford
,
united
kingdom,
Vol. 28
, No. 2, p
p
. 133-141
,
200
6.
[2]
Cha
t
urv
e
di D. K
,
P
r
e
m
da
yal
S
.
“
M
odified Neur
al
and Neuro-fu
zz
y Approa
ch for
S
hort T
e
rm
Lo
ad
F
o
recas
ting
,
”
2
nd
International Co
nference on
Po
w
e
r, Control and
Embedded S
y
stems
, 2012.
[3]
Ma
hala
na
bis
A
.
K
,
K
o
th
ar
i D
.
P
,
A
h
s
o
n
S
.
I
,
“
C
om
p
u
t
er
Aided
P
o
we
r
Sy
s
t
e
m
Ana
l
y
s
i
s
and
C
ont
rol,
”
Tata Mc
G
r
aw
Hi
ll
Pu
b
l
i
s
h
i
n
g
C
o
mp
an
y
li
mi
t
e
d,
NewDe
l
h
i
,
1988.
[4]
B
ox
G
.
E
,
Jen
k
i
ns
G
.
M,
“Ti
m
e
Se
r
i
es An
aly
s
i
s
For
e
ca
s
t
in
g
a
nd c
o
nt
r
o
l,
”
Hold
en
-
D
a
y
, Sa
n
F
r
a
n
s
i
sc
o,
19
76
.
[5]
Mogh
ra
m I
,
R
a
hman
S,
“
A
n
a
ly
si
s an
d
evalu
a
ti
on
of
fi
ve s
h
ort
-
t
e
r
m
loa
d
forec
a
s
t
i
n
g
tech
ni
qu
es,
”
IEEE
T
r
an
s
a
c
t
io
nson P
o
w
e
r
Sys
t
em
s
,
V
o
l. 4
,
No. 4
,
p
p
. 14
84
-
1
491
, 1989
.
[6]
C
h
e
n
C.
H. “Fuzzy
logic an
d
Ne
ur
a
l
Netw
o
r
k Han
d
boo
k,” Mc
G
r
a
wHi
ll
c
o
mpu
t
er
En
gg., 199
6
.
[7]
Kli
s
G.
J
,
F
o
lg
er T.
A
.
“Fuzzy
sets un
cer
taint
y
an
d
in
f
o
rm
ation
.
P
r
e
n
tice
H
a
l
l
o
f
I
ndia pr
iv
ate l
i
mite
d
,” 1993.
[8]
Shing J, Jang R, “ANFIS: Adapt
i
ve Ne
twork-Based-Fuzzy
Inf
e
r
e
nce S
y
s
t
em,”
I
EEE transact
i
on
on systems, man
and cyb
ernetics
,
Vol. 23
, No. 3, 1
993.
[9]
“ANFIS and the
ANFIS Editor GUI,”
Ma
tlab do
cumentation
cen
ter,
MATLAB R
2013a.
[10]
Jang J. S. R, Sun C.
T. “Neu
rofuzzy
modeling
and
contro
l,”
Pro
c
.I
EEE83
,
Vol. 3
,
pp. 378–406
, 19
95.
[11]
Jang J. S. R, Sun C.
T, Mizu
tani
E, “Neurofuzzy
and
soft computing,”
Pr
en
tice H
a
ll, pp. 510-514,
1997.
[12]
Bunke H, Kandel A, “Neuro
-fuzzy
P
a
tt
ern R
ecog
n
ition,”
W
o
rl
d Sc
i
e
nt
i
fi
c Publishing,
Singapor
e, 2
000.
[13]
Xu G,
Wa
ng Z
,
“Powe
r
Sy
st
em L
o
a
d
Flow
Distribution Res
earch b
a
sed on Adaptive Neur
o-Fuzzy
Inf
e
ren
ce
S
y
ste
m
s,
”
IE
EE
,
2012.
[14]
Bonissone P, “Adaptiv
e Neur
al F
u
zzy
Inf
e
ren
c
e Sy
st
ems (ANFIS): Analy
s
is
a
nd A
pplications,” 20
02.
[15]
Equilibr
e
s,
la
let
t
re de
la
commission de régu
latio
n de
l’é
l
ec
trici
t
é
et du
Gaz (
C
REG)
, No. 14, 2011
.
[16]
S
o
ciét
é na
tion
a
l
e
d’é
l
e
c
tri
c
it
é
et
de ga
z
algé
rienn
e
(S
ONELGAZ)
.
Newsletter
pres
s, No. 09
, 2012
.
[17]
Archive d
e
cour
bes de
charg
e
qu
otidienn
e,
Oper
a
t
eur
de
s
y
s
t
eme
ele
c
tr
ique
,
http://www.
ose.
dz/courbes.
php
.
[18]
Historical weath
e
r data website
h
ttp://www.wunderground.com/histor
y
.
Evaluation Warning : The document was created with Spire.PDF for Python.