Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
V
o
l.
6, N
o
. 1
,
Febr
u
a
r
y
201
6,
pp
. 12
~20
I
S
SN
: 208
8-8
7
0
8
,
D
O
I
:
10.115
91
/ij
ece.v6
i
1.8
901
12
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
Electri
c
i
t
y P
e
ak Load
Demand using De-n
oising W
a
vel
e
t
Transform integrated with Neural Network Methods
Pituk B
unn
oon
Department o
f
Electrical Engin
e
ering,
R
a
jamangala Univ
ersity
of
Techno
lo
g
y
Sriv
ijay
a
, Muang So
ngkhla, Th
ailand
Article Info
A
B
STRAC
T
Article histo
r
y:
Received Aug 25, 2015
Rev
i
sed
No
v 3, 201
5
Accepted Nov 20, 2015
One of m
o
s
t
important
elem
en
ts
in el
ec
tric pow
er s
y
s
t
em
plann
i
ng is
loa
d
forecasts. So, in
this paper proposes
the load dem
a
nd forecast
s
using de-
noising wavelet transform (DNWT) inte
gr
ated
with n
e
ural network (NN)
methods. This r
e
search
, th
e case stud
y
uses peak load d
e
mand
of Thailand
(Electricity
Gen
e
rating Authority
of Th
ailand: EGAT). The data of demand
will be
ana
l
yz
ed with m
a
n
y
influen
c
ing v
a
riab
les for se
lec
ting an
d
clas
s
i
f
y
ing fa
cto
r
s
.
In the res
earc
h
, th
e de-noising
wavelet tr
ansfor
m uses for
decomposing the peak lo
ad signal into 2 comp
onents these ar
e detail and
trend
components. Th
e for
e
casting
method
using th
e neural n
e
twor
k
algorithm
is used. Th
e work r
e
sults
ar
e shown
a good p
e
rformance of th
e
model proposed. The result may
be taken
to the o
n
e of decision in
the power
s
y
stems operatio
n.
Keyword:
De-
n
oisy
W
a
v
e
let
M
i
d-t
e
rm
ener
gy
dem
a
nd
Neu
r
al
net
w
or
k
Peak l
o
ad fore
casts
Weat
he
r a
n
d
e
c
on
om
i
c
fact
or
s
Copyright ©
201
6 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
:
Pitu
k
B
u
n
noo
n,
Depa
rtem
ent of Elect
ri
cal
E
n
gi
nee
r
i
n
g,
Facu
lty of
En
gin
eer
ing
,
Raj
a
man
g
a
la Un
i
v
er
sity o
f
Technolo
g
y
Sr
iv
i
j
aya,
Son
gkh
la
1
Ratch
a
d
a
m
n
o
e
nno
k Ro
ad
,
Bo
Yang
, Mu
an
g So
ngk
h
l
a
Th
ailan
d
900
00
.
Em
a
il: dr.pituk.b@ieee.org,
Pituk.b@rm
utsv.ac.th
NO
MEN
C
LA
TURE
AI
Artificial
in
tellig
en
ce
DW
T
Discret
wav
e
let
tran
sform
s
DN
De-
n
oisy
V STLF
Very s
h
ort term
load Forecas
ting
STLF
Short term
loa
d
Forecasting
MTLF
Medium
ter
m
load Forecasting
LTLF
Long term
load Forecasting
EGAT
Electricity Generatin
g Au
tho
r
i
t
y o
f
Th
ailand
NN
Neu
r
al
net
w
or
k
GDP
G
r
o
ss
do
m
e
sti
c
pr
odu
ct
CPI
C
ons
um
er pri
c
e i
nde
x
T
ma
x
M
ont
hl
y
m
a
xim
u
m
t
e
m
p
erat
ure
T
mi
n
M
ont
hl
y
m
i
ni
m
u
m
t
e
m
p
erat
ure
T
avg
Mont
hly avera
g
e tem
p
erature
H
M
ont
hl
y
h
u
m
i
di
t
y
1.
INTRODUCTION
Peak load dem
a
nd forecasting, is one of predicted
dem
a
nds for consum
ption lo
ad of the country.
It
shows
a m
a
xim
u
m
required in each hour, m
onth, and y
ear
in the future. Load dem
a
nd, it is depended on
m
a
ny
i
n
fl
uenci
ng fact
ors. These fact
ors
are econom
i
c
vari
abl
e
s such
as i
ndust
r
i
a
l
fact
or, consum
er pri
ce
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
6, No
. 1, Feb
r
uar
y
20
1
6
:
1
2
– 20
13
i
ndex, or GDP of t
h
e count
ry
, and weat
her vari
abl
e
s for i
n
st
ance t
e
m
p
erat
ure, hum
i
d
i
t
y
, ul
t
r
avi
o
l
e
t
,
rai
n
fal
l
,
and wind speed. Thus, the peak load dem
a
nd forecasting
is significantly for the pow
er system
operation.
The
case of the
forecasting can be
classified into 4
types such as
very short-term
, short-term
,
m
e
dium
-term
,
and long-term
load forecasts. Each for
ecasting has advantage as follows: VSTLF result is used
fo
r p
o
w
er system
o
p
e
ratio
n
;
STLF resu
lt u
tilizes fo
r p
o
w
er system
m
a
in
ten
a
n
ce an
d
o
p
e
ratio
n
;
MTLF
resu
lt
uses for
fuel reserve
planning; LTLF
result is
related to
power plant pla
nning
in the
future. This research
proposes the forecasting interval MTLF m
odel.
In the past of
paper, m
a
ny m
e
thods
are used fo
r load
dem
a
nd forecasting in
the power system
.
There
are
t
w
o m
a
jor approaches. These
m
e
t
hods are st
at
i
s
t
i
c
m
e
t
hod such
as exponent
i
a
l
sm
oot
hi
ng (ES) [1]
,
m
u
ltip
le lin
ear reg
r
essio
n
(MLR) [2
], an
d
au
to
reg
r
essiv
e
in
teg
r
ated
m
o
v
i
n
g
av
erag
e (ARIMA) [3
-6
]. So
m
e
m
e
thods are conventional approaches. The presented m
e
thod, artificial
intelligence
is used a lot in
load
predi
c
t
i
on such as art
i
f
i
c
i
a
l
neural
net
w
ork (NN) [7-16]
, support
vect
or m
achi
n
e (SVM
) [17-21]
, fuzzy
l
ogi
c
[22-26].
Many papers
at the present
propose the for
ecasting
research using wavelet
transform
with support
vector m
achine
or neural
network a
pproach [27].
Several papers
interest
the
forecasts in
short-term
load
prediction
[28-40], m
i
d-term
load
forecasts [20] [33]
,
and long-term
load
for
ecasting [41-43]. Also, som
e
paper interests to forecast
by focus
in
m
u
lti-area forecasts [44] and
on the
influence
of weather factors
[32].
W
h
i
l
e
, som
e
researches use AI for ot
her work i
n
power sy
t
e
m
[45-46]
.
This research
presents the
load dem
a
nd
for
ecasts
using denoising
wavelet transform
(DNW
T)
i
n
t
e
grat
ed wi
t
h
neural
net
w
ork (NN)
m
e
t
hods. The
case st
udy
used
peak l
o
ad
dem
a
nd of
Thai
l
a
nd (EGAT)
i
s
proposed. The peak load dem
a
nd data
will be analysed with influenc
ing variables fo
r selecting and
classifyin
g
facto
r
s. Th
e d
e
-n
o
i
sin
g
wav
e
let tran
sfo
r
m
will b
e
u
s
ed
fo
r d
eco
m
p
o
s
in
g
th
e m
o
n
t
h
l
y p
eak
lo
ad
signal
into two com
ponents (detail and trend com
ponent
s)
after separating. The m
a
in forecast m
e
thod using
t
h
e neural
net
w
ork al
gori
t
h
m
i
s
used.
Thi
s
paper i
s
st
ruct
ured as fol
l
o
ws. Sect
i
on 2
provi
des an i
n
t
r
oduct
i
on t
o
el
ect
ri
ci
t
y
dem
a
nd
and
i
n
fl
uenci
ng
fact
ors of Thai
l
a
nd. Sect
i
on 3 proposes
t
h
e t
h
eory
and i
m
pl
em
ent
a
t
i
on m
e
t
hodol
ogi
es. Sect
i
on 4
gives the data
testing and forecasting
results. Section
5 provides the
discu
ssion, and finally
the conclusion
is
drawn i
n
sect
i
on 6.
2.
ELECTRICITY DE
MAND AND
INFLUENCING FACTORS
Load
charact
eri
s
t
i
c
s vary
si
gni
fi
cant
l
y
am
ong di
fferent
el
ect
ri
ci
t
y
sy
st
em
s;
t
h
erefore, i
t
i
s
very
im
portant to study the energy
dem
a
nd pattern of th
e system
before construc
ting the dem
a
nd
forecasting
m
odel
i
n
t
h
e research. In t
h
i
s
sect
i
on, t
h
e paper di
sc
usses about
energy
l
o
ad dem
a
nd and Infl
uenci
ng fact
ors
of t
h
e dem
a
nd i
n
Thai
l
a
nd.
Fi
gu
re
1.
Ene
r
gy
dem
a
nd
(M
W)
an
d e
n
er
gy
co
ns
um
pt
i
on l
o
ad
dem
a
nd
(
M
W
h
)
2.1. E
n
ergy Demand
Electric energy consum
ption is the
form
of energy
consum
pt
i
on t
h
at
uses electric energy. It is
the
act
ual
energy
dem
a
nd m
a
de
on exi
s
t
i
ng el
ect
ri
ci
t
y
suppl
y
.
It
can be
cl
assi
fi
ed i
n
t
o
t
w
o
t
y
pes t
h
at
are
energy
dem
a
nd (
W
) and
energy
consum
pt
i
on
dem
a
nd (
Wh
). The
W
is a
m
easure of
power, while
Wh
is
a m
easure of
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
El
ect
ri
ci
t
y
Peak Lo
a
d
Dem
a
n
d
usi
n
g
De-
n
oi
si
ng
Wavel
et
T
r
an
sf
or
m i
n
t
e
g
r
at
ed
w
i
t
h
…
(Pitu
k Bunno
on)
14
en
erg
y
. Th
is p
a
p
e
r
will b
e
ex
p
l
ain
e
d
in
MW
o
r
m
e
g
a
watt, in
Th
ailan
d
electricity, Electricity
Gen
e
ratin
g
Aut
hori
t
y
of Thai
l
a
nd (
EGAT
). The peak l
o
ad dem
a
nd of t
h
e count
ry
i
s
shown i
n
Fi
gure 1.
Fi
gure 1, shows t
w
o
charact
eri
s
t
i
c
s of
energy.
There are energy
dem
a
nd (
MW
) and
consum
pt
i
on
dem
a
nd (
MWh
). These dat
a
were recorded from
y
ear
1997 t
o
2004 i
n
m
ont
hl
y
t
y
pe. The el
ect
ri
ci
t
y
energy
has
the trend of dem
a
nd to
increase every year. Mean
wh
ile, th
e en
erg
y
d
e
tails in
all m
o
n
t
h
sh
o
w
th
e
behavi
our of peak l
o
ad dem
a
nd com
p
l
i
a
nce wi
t
h
i
n
fl
uenci
ng fact
ors.
2.2. Infl
uenci
n
g Factors
In
the power system
, m
i
d-term
load forecasting,
the weather and
econom
y are usually the dom
inant
vari
abl
e
s
i
n
dri
v
i
ng t
h
e
el
ect
ri
ci
t
y
dem
a
nd of
t
h
e count
ry
. Fi
rst
l
y
,
t
h
e weat
her fact
ors
m
ean t
h
e t
e
m
p
erat
ure,
h
u
m
id
ity, rain
fall, an
d
win
d
sp
eed
. Th
e
tem
p
eratu
r
e f
actors for
instance m
a
xim
u
m
tem
p
erature,
m
i
nim
u
m
tem
p
erature,
and average tem
p
erature
are very affectively in the behavi
our com
ponent of the dem
a
nd in each
m
ont
h. It
can i
ndi
cat
e t
h
e cust
om
er behavi
our caused by
the tem
p
erature change. Fo
r exam
ple, especially,
if
today is higher tem
p
erature, the consum
ers will open air
conditioner a lot. It m
a
kes m
o
re dem
a
nds of the
area
or t
h
e count
ry
. The
l
a
st
, econom
i
c
vari
abl
e
s, t
h
ere
are m
a
ny
fact
ors such
as gross dom
est
i
c
product
(GDP),
consum
er
pri
ce i
ndex (C
PI), and
i
ndust
r
i
a
l
i
ndex (ID
I). These econom
i
c
fact
ors show t
h
e econom
i
c
growt
h
of
t
h
e count
ry
, t
h
e
growt
h
t
r
end. The
si
gni
fi
cant
i
ndex i
s
GDP. In sum
m
a
ry
,
bot
h weat
her and econom
i
c
in
flu
e
n
c
in
g
facto
r
s in
d
i
cate th
e
lo
ad
d
e
m
a
n
d
o
f
all
areas.
Th
ese facto
r
s
will affect with
th
e
tren
d
o
f
th
e
p
eak
dem
a
nd
and the custom
er behaviour of each m
onth. So
,
in the research, both weather and econom
ic factors
will
be used
for the feature
input of
peak
load m
odel
forecasting which th
e
feature input is
described in the
next
sect
i
on.
3.
THEORY AND
IMPLEME
N
TA
TIO
N
METHODOL
O
G
IES
3.1.1. De-noi
sy Wavel
e
t T
r
ansforms
Th
e
wav
e
let tran
sfo
r
m
th
eo
ry is ap
p
licab
le to
sev
e
ral su
b
j
ects in
th
e research
. All wav
e
let
t
r
ansform
s
(W
T) m
a
y
be consi
d
ered pat
t
e
rn form
of t
i
m
e-frequency
represent
a
t
i
on for cont
i
nuous
t
i
m
e
si
gnal
s
(anal
og si
gnal
)
and so
are rel
a
t
e
d
t
o
harm
oni
c an
alysis. Alm
o
st
all p
r
actically
u
s
efu
l
d
i
screte
W
T
u
s
e
d
i
screte-tim
e filter
b
a
n
k
s
(FB).
Th
is FB
is called
th
e wav
e
let
an
d
scalin
g
co
efficien
ts in
wav
e
let
nom
encl
at
ure.
Thi
s
FB m
a
y
cont
ai
n ei
t
h
er fi
ni
t
e
i
m
pul
se
response (FIR) or
i
n
fi
ni
t
e
i
m
pul
se response (IIR)
filters. The wavelets
form
ing a
continuous wavelet
tran
sform
(CW
T
) are
subject to
the uncertainly
principle
of Fouri
e
r t
r
ansform
anal
y
s
i
s
.
The product
of t
h
e
uncert
a
i
n
t
i
e
s of t
i
m
e and
frequency
response scal
e has
a
l
o
wer bound.
Thus, i
n
t
h
e scal
eogram
of a
cont
i
nuous wave
l
e
t
t
r
ansform
of t
h
e
si
gnal
.
Al
so,
di
scret
e
wavel
e
t
t
r
ansform
(DW
T
) bases m
a
y
be consi
d
ered i
n
t
h
e
cont
ext
of ot
her form
s of t
h
e uncert
a
i
n
l
y
pri
n
ci
pl
e.
Fi
nal
l
y
,
wavelet transform
s
are divided into three type
s: continuous, discrete,
and m
u
lti-resolution.
Basically, wh
en
d
ealin
g wit
h
wav
e
let an
alysis, two
typ
e
s
of tran
sfo
r
m
s
can
b
e
u
s
ed
: th
e
co
n
tinuo
us
wav
e
let an
d the d
i
screte
wavelet tran
sfo
r
m
s
.
Fi
rst
l
y
, cont
i
nuous wavel
e
t
t
r
ansform
(C
W
T
), t
h
e
cont
i
nuous wavel
e
t
t
r
ansform
i
s
t
h
e sum
over al
l
t
i
m
e
of scal
ed and shi
f
t
e
d val
u
e of a wavel
e
t
funct
i
on
ψ
. C
W
T, when appl
i
e
d on t
h
e ori
g
i
n
al
si
gnal
f
(
t
), is
expressed as:
1
_,
*
t
CW
T
f
s
f
t
d
t
s
s
(1
)
W
h
ere
s
represents the
scale param
e
ter,
τ
rep
r
esen
ts th
e
tran
slatio
n
p
a
ram
e
ter,
an
d
represents the
wav
e
let fu
n
c
tio
n
.
Lastly, d
i
screte W
a
v
e
let tran
sfo
r
m
(D
W
T
), i
n
pract
i
ce, si
gnal
s
acqui
red experi
m
e
nt
al
l
y
are
not
cont
i
nuous i
n
t
i
m
e,
but
sam
p
l
e
d as
di
scret
e
t
i
m
e i
n
t
e
rval
s.
Previ
ousl
y
, we have
seen t
h
at
t
h
e C
W
T
perform
s a tim
e-frequency
resolution (or
m
u
ltiresolution)
by scaling
(contraction a
nd
dilation) and
shifting a
wavelet function. Recently, it has
been shown that
such analysis
can actually be perform
ed
using
m
u
ltireso
l
u
tio
n
filter b
a
n
k
s
an
d
wav
e
let
fu
n
c
tio
n
s
, resu
ltin
g
in
th
e
Discrete W
a
v
e
let
Tran
sfo
r
m
(DW
T
).
Th
e
DW
T is realized
b
y
m
ean
o
f
th
e
filters
h
[
k
]
, g
[
k
] that are related
to each other through the equation
as
fo
llo
win
g
.
1
1
1
0
,
1
,
...,
1
k
gk
h
N
k
k
N
(2
)
W
h
ere
N
rep
r
esen
ts
th
e len
g
t
h
o
f
th
e sig
n
a
l
filter. Th
ese
filter are co
n
s
tru
c
ted
fro
m
th
e
wav
e
let
kernel
(
t
) and t
h
e com
p
ani
on scal
i
ng funct
i
on
ϕ
(
t
) t
h
rough t
h
e rel
a
t
i
ons bel
o
w.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
6, No
. 1, Feb
r
uar
y
20
1
6
:
1
2
– 20
15
22
k
th
k
t
k
(3
)
22
k
tg
k
t
k
(4
)
Al
so,
de-nosy
wavel
e
t
t
r
ansform
s
t
h
e basi
c i
d
ea:
one
area where wavel
e
t
s
have
proven t
o
be very
useful
i
s
t
h
at
of nonparam
e
t
r
i
c
regressi
on or si
gnal
de
-noi
si
ng.
Thi
s
i
s
a way
of rem
ovi
ng random
noi
se from
a seri
es i
n
order t
o
l
eave
t
h
e t
r
ue si
gnal
.
Thi
s
i
s
done
wi
t
h
no knowl
e
dge
of t
h
e form
of t
h
e
underl
y
i
ng
funct
i
on.
To begi
n wi
t
h
l
e
t
us assum
e
t
h
at
a noi
sy
si
gnal
has t
h
e form
,
ii
i
yg
x
e
(5
)
Hence
g
(
x
i
) represent
s
t
h
e t
r
ue
funct
i
on t
o
be est
i
m
at
ed
and
e
i
i
s
som
e
form
of random
noi
se
usual
l
y
assum
e
d
t
o
be norm
a
l
l
y
di
st
ri
but
ed.
The basi
c procedure for
de-noi
si
ng i
s
t
h
en t
o
t
a
ke a t
r
ansform
of t
h
i
s
n
o
i
sy sig
n
a
l. Th
e tran
sfo
r
m
e
d
sig
n
a
l will th
en
h
a
v
e
th
e fo
rm
,
*
dd
(6
)
where
d
represent
s
t
h
e t
r
ansform
of t
h
e t
r
ue si
gnal
and
ε
presents the transform
of the noise.
3.1.2. Neural
Netw
ork Al
gori
t
hm
Neural
net
w
ork al
gori
t
h
m
(NN), i
s
an
i
n
form
at
i
on processi
ng paradi
gm
t
h
at
i
s
i
n
spi
r
ed
by
t
h
e way
bi
ol
ogi
cal
nervous sy
st
em
s, such as t
h
e
brai
n process i
n
form
at
i
on. The key
el
em
ent
of t
h
i
s
paradi
gm
i
s
t
h
e
novel
st
ruct
ure of t
h
e i
n
form
at
i
on processi
ng sy
st
em
. It
i
s
com
posed of a l
a
rge num
ber of hi
ghl
y
i
n
t
e
rconnect
ed
processi
ng el
em
ent
s
(neurones) worki
ng i
n
uni
on
t
o
sol
v
e speci
fi
c probl
em
s. NN, l
i
k
e peopl
e,
learn
by exam
ple. The NN is configured for a
specifi
c appl
i
cat
i
on, such as pat
t
e
rn recogni
t
i
on or dat
a
cl
assi
fi
cat
i
on,
t
h
rough a l
earni
ng process. Learni
ng i
n
bi
ol
ogi
cal
sy
st
em
i
nvol
ves adjust
m
e
nt
s t
o
t
h
e sy
napt
i
c
co
n
n
ectio
n
s
th
at ex
ist b
e
tween
th
e n
e
u
r
o
n
e
s. Th
is is
tru
e
o
f
NN as well. Netwo
r
k
arch
itectu
r
e, in
th
is
research, shows
m
u
ltiple-input neur
on. Typically,
a neuron
has m
o
re
than
one input.
A neuron
with
R
i
nput
s
i
s
shown i
n
Fi
gure 2.
The i
ndi
vi
dual
i
nput
p
1
, p
2
,..., p
R
are each
weighted by
corresponding elem
ents
w
1,1
,
w
1
,
2
, ...,w
1
,R
of wei
ght
m
a
t
r
i
x
W
.
Fig
u
re 2
.
Mu
ltip
le-Inp
u
t
n
e
u
r
al
The neuron has a bi
as
b
, whi
c
h i
s
sum
m
e
d wi
t
h
t
h
e wei
ght
ed i
nput
t
o
form
t
h
e net
i
nput
n
:
1,
1
1
1
,
2
2
1
.
...
RR
nw
p
w
p
w
p
b
(7
)
Th
is ex
p
r
essio
n
can
b
e
written
in
m
a
trix
fo
rm
:
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
El
ect
ri
ci
t
y
Peak Lo
a
d
Dem
a
n
d
usi
n
g
De-
n
oi
si
ng
Wavel
et
T
r
an
sf
or
m i
n
t
e
g
r
at
ed
w
i
t
h
…
(Pitu
k Bunno
on)
16
,
nW
p
b
(8
)
where
t
h
e m
a
t
r
i
x
W
for t
h
e si
gnal
neurons
case has onl
y
one row. Now t
h
e
neuron out
put
can be
written
as
s
fW
p
b
(9
)
This
paper uses the neural network
for training
and forecasting the
history peak load dem
a
nd and
predi
c
t
i
on i
n
t
h
e fut
u
re. The NN of t
h
e m
odel
i
s
descri
bed i
n
t
h
e next
sect
i
on.
3.1.3. Peak Load Forecasting Model
In this paper,
proposes a
new forecasting
m
odel.
There
is the
com
b
ination of
two m
e
thods
for
forecasting. These are de-noising
wavelet transform
s
a
nd artificial
neural network approaches. The
exam
ple
m
odel shows in
figure 3
which is the
de-noisy wavelet
transform
s
integrated with
neural
network forecasting
m
odel
based on 3-l
e
vel
.
From
t
h
e fi
gure
3, can be di
vi
ded
i
n
t
o
t
h
ree part
s
t
h
ese are t
h
e
part
of t
h
e hi
st
ori
cal
dat
a
(peak dem
a
nd and fact
ors), t
h
e part
of de-noi
sy
wavel
e
t
t
r
ansform
s
, and neural
net
w
ork part
.
Firstly, historical
data are
record
ed from
m
a
ny
organi
zat
i
ons.
There ar
e three
m
a
jor historical factor
dat
a
. These fact
ors are peak l
o
ad dem
a
nd, weat
her fact
or, and econom
i
c
fact
or.
Secondl
y
,
peak l
o
ad dem
a
nd
preprocessi
ng usi
ng de-noi
sy
W
T
,
i
n
t
h
i
s
st
age use
for t
h
e peak
l
o
ad
dem
a
nd
si
gnal
de-noi
sy
before decom
posi
ng i
n
t
o
1-l
e
vel
,
2-
lev
e
l, 3
-
lev
e
l, 4
-
lev
e
l, an
d
5
-
lev
e
l. Each
lev
e
l,
bot
h t
h
e det
a
i
l
(D) and t
h
e t
r
end (A) com
ponent
s are shown.
Finally,
this part is forecasting st
age
using the neural network. It
uses three m
a
jor features. These
are peak dem
a
nd i
n
hi
st
ori
cal
det
a
i
l
and
t
r
end com
pone
nt
s, weat
her vari
abl
e
s,
and econom
i
c
vari
abl
e
.
These
features
are significant data for trai
ning and forecasting the peak load dem
a
nd in the future. After this stage,
the forecasting data will be reconstructed to the actual data.
In concl
ude, t
h
e
bl
ock di
agram
i
n
fi
gure
3 shows
especi
al
l
y
t
h
e
peak l
o
ad
dem
a
nd decom
posi
t
i
on
i
n
th
ree-lev
e
l. It sh
o
w
s m
a
j
o
r th
ree p
a
rts wh
ich
are sig
n
i
fican
t fo
r m
o
d
e
l p
r
ed
ictio
n
in
th
is research
.
Fi
gu
re
3.
De
-n
oi
sy
wa
vel
e
t
t
r
ansf
o
r
m
i
n
t
e
grat
ed wi
t
h
ne
u
r
al
net
w
or
k
fo
re
cast
i
ng m
odel
base
d
on
3
-
l
e
v
e
l
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
6, No
. 1, Feb
r
uar
y
20
1
6
:
1
2
– 20
17
4.
DAT
A TESTI
NG
A
N
D
FO
REC
A
STI
N
G
RES
U
LTS
The
peak load and factors of Thailand count
ry from
January 1, 1997 to Decem
ber 31, 2006 are
chosen
as t
h
e t
r
ai
ni
ng and t
e
st
i
ng dat
a
for experi
m
e
nt
s in the research. The data
show the recorded in
m
onthly
type (January to Decem
ber). The unit of load
dem
a
nd
is in m
e
gawatt (MW
)
. Also, in the paper will
use
t
h
e i
n
fl
uenci
ng
fact
ors such as
weat
her and econom
i
c
vari
abl
e
s of
t
h
e count
ry
i
n
m
ont
hl
y
t
y
pe. The
recorded data cam
e from
the sam
e
year.
In this
research, we
use the
m
ean absolute
percentage error
(MAPE) to
m
easure the
forecast
perform
ance. The
MAPE is
defined as
the ratio
between
absolute
forecasting error
and the
the actual values
of t
h
e dem
a
nd. The M
A
PE (%) equat
i
ons can be cal
cul
a
t
e
d as fol
l
o
ws.
1
1
.
100
2
n
af
pp
MA
P
E
pa
,
(1
0)
where
n
i
s
t
h
e num
ber
of observat
i
on,
p
a
represents
the actual peak load dem
a
nd,
and
p
f
represents
the forecasting load dem
a
nd. In Table I, shows th
e results of peak dem
a
nd forecasting all 5-level.
The
investigative results from
figure 3 (for
exam
ple
P3LT) and
Tabl
e I have
t
h
e m
a
ny
pat
t
e
rn vari
es
from
P1LT
t
o
P5LT
m
odel
s
. These
m
odel
s
com
e
from
t
h
e
num
ber of
com
ponent
s i
n
t
h
e wavel
e
t
t
r
ansform
s
decom
posi
t
i
on. The neural
net
w
ork approach requi
res
a num
ber of
NN m
odel
equal
t
o
t
h
e num
ber
of
com
ponents
after W
T
decom
position. All of
the studyresu
lts from
5 m
odels above
can be shown
in Table I.
Table I, in the first colum
n
, explai
ns the result of the load forecasting
outcom
e
in 12 m
onths from
January
to
Decem
ber
of the
year 2007,
a period
that is
forecasted. It
uses the
five W
T
m
odels for
load prediction. After
th
at, wil co
m
p
are th
e resu
lts o
f
th
e lev
e
l 1
,
lev
e
l 2
,
lev
e
l 3
,
lev
e
l 4
,
an
d
Lev
e
l 5
with
d
e
-n
o
i
sy W
T
. Th
e fin
a
l
resul
t
s
show i
n
t
h
e Tabl
e I. The P1LT, P i
s
t
h
e m
a
xi
m
u
m
l
o
ad or peak l
o
ad m
ont
hl
y
dem
a
nd, 1L i
s
t
h
e
1
-
lev
e
l
o
f
wav
e
let tran
sfo
r
m
after sig
n
a
l d
e
-n
o
i
sy, an
d
T is
th
e test (testin
g
)
. Each
co
lu
m
n
will sh
o
w
th
e p
eak
load
forecasting dem
a
nd in
norm
a
lized value. It can
convert
this value
to the actual
value
or peak load value
u
s
in
g
b
a
se v
a
lu
e m
u
ltip
lier o
f
th
e research
. Th
en
it can
m
a
k
e
b
ack
to
th
e m
a
x
i
m
u
m
lo
ad
v
a
lu
e (MW
)
. In
th
e
result, the percent
error in
each m
ont
h
is shown
in the PE
(%), which
will display
all five
levels and
will
finally have the com
p
lete the m
ean
absolute percent error (MAPE) of each
m
odel. The results are m
easured
by MAPE of each approach, and
are com
p
ared. In su
m
m
a
ry, the MAPE
of the De-P1LT is 4.39%. The
De-
P2LT m
odel
,
t
h
e M
A
PE i
s
5.85%. The M
A
PE of t
h
e De-P3LT i
s
7.91%. The De-P4LT, t
h
e M
A
PE i
s
8.88%.
Fin
a
lly, th
e De-P5
LT, MAPE is 1
0
.
7
2
%
, all o
f
wh
ich
are d
i
sp
layed
in
th
e tab
l
e I.
Th
e
resu
lt
will
b
e
d
i
scu
ssed
in
th
e fo
llo
win
g
sectio
n
.
Tabl
e 1.
Th
e resu
lt of
p
eak lo
ad
d
e
m
a
n
d
fo
recasti
n
g
(M
W in
n
o
rm
aliza
tio
n
)
in
each
lev
e
l
Lev
e
l
De-1
(P1
L
T)
De-2
(P2
L
T)
De-3
(P3
L
T)
De-4
(P4
L
T)
De-5
(P5
L
T)
Month Forecasted
(M
W
)
PE(%)
Forecasted
(M
W
)
PE(%)
F
orecasted
(M
W
)
PE(%)
F
orecasted
(M
W
)
PE(%)
Forecasted
(M
W
)
PE(%)
Januar
y
0.
7266
-
0
.
05
0.
7322
-
1
.
82 0.
7733
-
7
.
54
0.
7600
-
5
.
69
0.
7446
-
3
.
55
Febr
uar
y
0.
7249
4.
64
0.
7395
7.
81
0.
7630
4.
88
0.
7642
4.
73
0.
7385
7.
94
M
a
r
c
h
0.
8254
13.
82
0.
7940
17.
09
0.
7663
19.
99
0.
7584
20.
82
0.
7312
23.
66
Apr
il
0.
9999
0.
01
0.
8251
17.
49
0.
7994
20.
05
0.
8028
19.
71
0.
7457
25.
42
M
a
y
0.
8336
3.
71
0.
8389
3.
10
0.
7915
8.
58
0.
7793
9.
98
0.
7515
13.
19
June
0.
8247
8.
12
0.
8228
2.
77
0.
7904
11.
95
0.
7772
13.
42
0.
7507
16.
36
July
0.
7972
-
1
.
86
0.
8287
-
5
.
77 0.
7866
-
.
050
0.
7731
1.
22
0.
7546
3.
58
August
0.
7786
5.
96
0.
8092
2.
27
0.
7711
6.
86
0.
7563
8.
66
0.
7473
9.
74
Septem
ber 0.
7885
7.
11
0.
8011
5.
62
0.
7771
8.
45
0.
7566
10.
86
0.
7521
11.
39
October
0.
8108
-
1
.
05
0.
7837
2.
31
0.
7919
1.
29
0.
7618
5.
04
0.
7621
5.
00
Novem
b
er
0.
8091
-
2
.
25
0.
8054
-
1
.
78 0.
8015
-
1
.
29
0.
8288
-
4
.
74
0.
7769
1.
81
Dece
m
b
er
0.
8118
3.
16
0.
8184
2.
37
0.
8088
3.
52
0.
8242
1.
68
0.
7794
7.
02
M
A
PE
4.
39
5.
85
7.
91
8.
88
10.
72
5.
DIS
C
USSI
ON
From
the result in the previous s
ection, the forecasts are relatively well during the level 1 and 2 of
the de-noisy wavelet transform
.
The fo
recasting errors
are quite higher in
le
vel three, four,
and five level
of
W
T
decom
posi
t
i
ons. W
e
concl
ude t
h
at
, t
h
e
de-noi
sy
wavel
e
t
for
si
gnal
separa
t
i
on shoul
d
be enough
m
a
de
only at one or two
level appreciably. Som
e
m
onth, the
error is still high because
of the weather change
is
non-stationary in each year,
especia
lly, the weather
factor of each
seas
on. For exam
ple,
the hot season
has
very
hot
for a
l
ong t
i
m
e com
p
ared
wi
t
h
t
h
e past
. Therefore,
furt
her research shoul
d
st
udy
t
h
e dy
nam
i
cs
of
clim
ate
data from
the past to the pres
ent to im
prove the forecasting results.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
El
ect
ri
ci
t
y
Peak Lo
a
d
Dem
a
n
d
usi
n
g
De-
n
oi
si
ng
Wavel
et
T
r
an
sf
or
m i
n
t
e
g
r
at
ed
w
i
t
h
…
(Pitu
k Bunno
on)
18
6.
CO
NCL
USI
O
N
The
m
a
i
n
cont
ri
but
i
ons m
a
de i
n
t
h
e
present
work are present
e
d as
fol
l
o
ws:
i
n
t
h
i
s
paper proposes,
the load
dem
a
nd forecasting
by using
de-noising wavelet
transform
(DNW
T)
com
b
ined
with neural network
(NN) approach. For this research,
th
e case st
udy
of Thai
l
a
nd
(El
ect
ri
ci
t
y
Generat
i
ng Aut
hori
t
y
of
Thai
l
a
nd:
EGAT) is proposed. The
peak load da
ta will
be analysed with m
a
ny
in
fluencing factors
for selecting
and
classifying
factors used in
the res
earch. The
de-noi
si
ng wavel
e
t
t
r
ansform
i
s
used for
decom
posi
ng t
h
e peak
load signal into 2 m
a
jor
com
ponents. These com
pone
nts are
detail and trend ingredients.
Theforecasting
m
e
t
hod usi
ng t
h
e neural
net
w
ork m
e
t
hod
i
s
used. The research resul
t
s
are
shown a good perform
ance of
t
h
e
m
odel
proposed. The resul
t
m
a
y
be t
a
ken t
o
t
h
e one of deci
si
on i
n
t
h
e power sy
st
em
s.
AC
KN
OWLE
DG
MENT
The
aut
hors woul
d l
i
k
e t
o
t
h
ank t
h
e edi
t
o
r and t
h
e
anony
m
ous revi
ewers for t
h
ei
r val
u
abl
e
com
m
e
nt
s
and suggest
i
ons t
o
i
m
prove t
h
e paper and t
h
e present
a
t
i
on.
REFERE
NC
ES
[1]
Xianghe Zhu, and Min Shen, "Based on the ARIMA
m
odel with
grey
theory
for short term
load forecasting
m
odel",
2012 International
conference on systems and informatics
, pp. 564-567, 2012.
[2]
Am
ra
N., Ozveren C.C., King D.,
"S
hort term
lo
ad forecasting using
m
u
ltiple linear regression",
42nd International
confer
ence of Univer
s
ity
power engineering
, pp. 1192-1198, 2007.
[3]
Suhartono, Indah Puspitasari,
M. Sjah
id Akbar,
and Muhammad Hisy
am
Lee,
"Two-level seasonal
model based
on
hy
brid ARIMA-ANFIS for forecasting
short-
term
electricity
load in
Indonesia",
2012 International
confer
ence on
statistics in science, business and engineering
, pp. 1-5, 2012.
[4]
Suhartono, Indah Puspitasari,
M. Sjah
id Akbar,
and Muhammad Hisy
am
Lee,
"Two-level seasonal
model based
on
hy
brid ARIMA-ANFIS for forecasting
short-
term
electricity
load in
Indonesia",
2012 International
confer
ence on
statistics in science, business and engineering
, pp. 1-5, 2012.
[5]
Xianghe Zhu, and Min Shen, "Based on the ARIMA
m
odel with
grey
theory
for short term
load forecasting
m
odel",
2012 International
conference on systems and informatics
, pp. 564-567, 2012.
[6]
Jian-Chang Lu, Dong-Xiao, and Zheng-Yuan Jia, "A st
udy
of shortterm load forecasting based on ARIMA ANN",
International conference
on machine learning and cybernetics
, vol. 5, pp. 3183-3187, 2004.
[7]
A.S.
Khwaja, M. Naeem
, A.
Anpalaga
n, A. Venetsanopoulos, and
B. Venka
tesh, "Im
p
roved short-term
load
forecas
ting us
ing bagged neural networks
",
Electr
i
c Power
Sys
t
em Res
e
ar
ch
, vol. 125 (2015), pp. 109-115, 2015.
[8]
Kenji Nose-Filho,
Anna Diva
Plasenci
a, and
Carlos Roberto
Minussi, "Short-
term
m
u
ltinodal
load forecasting using
a modified general regression neural network",
IEEE T
r
ans. Power Del.
, vol. 26, no. 4, pp. 2862-2869, October
2011.
[9]
Osm
a
n Z.H., Awad
M.L., and
Mahm
oud T.K.,
"Neural networ
k based
approach for
short-term
load
forecasting",
PSCE09 IEEE/PES Power
system conference and exposition
, pp. 1-8, 2009.
[10]
Sajjad
Kouhi, Farshid Key
n
ia, and Saja
d Najafi Ravadanegh,
"A new short-
term
load forecast m
e
thod based on
neuro-evolutionary
algorithm and chaotic feature
selection",
International Journal of
Electrical Power
and Energy
Systems
, vol. 62 (2014), pp. 862-867, November 2014.
[11]
Toly
Chen, "A collaborative fuzzy
-neural approach
for long-term
load forecasting in Taiwan",
Computers and
industrial engineering
, vol. 62 (2012), pp. 663-670, November 2012.
[12]
Sajjad, and Farshid Key
n
ia,
"A new cas
cade NN
based m
e
thod to shortterm
load forecast in deregulated
electricity
ma
rke
t
",
Energy conversion
and management
, vol. 71 (2013), pp. 76-83, July
2013.
[13]
Yongli
W
a
ng, Dongxiao
Niu, and Li
Ji,
"A
SOM-based hy
brid linearneural
m
ode
l for
short-term
load forecasting",
Neurocomputing
, vol. 74, no. 17, pp. 2874-2885, Octorber 2011.
[14]
Long-term
load forecasting, "Long-term
load forecasting base
d on adaptive neural fuzzy
inference sy
stem
using real
energy
data",
Energy
Pr
ocedia
, vol. 14, pp. 794-799, 2012.
[15]
Ruey
-Hs
un Liang, and Ching-Chi Cheng, "S
hort-term
load forecas
ting by
a neuro-fuzzy
bas
e
d
approach",
International Journal of Electrical
Power and Energy Systems
, vol. 24, no. 2, pp. 103-111, February
2002.
[16]
Xianghe Zhu, and Min Shen, "Based on the ARIMA
m
odel with
grey
theory
for short term
load forecasting
m
odel",
International conference
on system and informatics
, pp. 564-567, 2012.
[17]
Zhongy
i
Hu, Yukun
Bao, and Tao
Xiong, "Comprehensive
learning particle swarm
op
timization based memetic
algorithm
for m
odel selection in short-term
lo
ad forecasting using support vector regression",
Applied
Soft
Computing
, vol. 25 (2014), pp. 15-25, 2014.
[18]
Ervin
Ceperic, Vladim
ir Ceperic,
and
Adrijan Baric,
"A strategy
for
short-
term
load forecasting
by
support vector
regres
s
i
on m
achines
",
IEEE
Trans. Power Syst.
, vol. 28, no. 4, pp. 4356-4364, November 2013.
[19]
Rong Gao, Liy
u
an, and Xiaohua Liu, "Short-term
load
forecasting based on least square support vector m
achine
com
b
ined with fuzzy
control",
10th World congress on intelligent control and automation
, pp. 1048-1051, 2012.
[20]
Xinxing
Pan, and Lee B., "A
comparison of support vector
machines
and artificial neur
al networks for mid-term
load forecas
ting",
2012 IEEE
International conference on industrial technology
, pp. 95-101, 2012.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJEC
E V
o
l
.
6, No
. 1, Feb
r
uar
y
20
1
6
:
1
2
– 20
19
[21]
Zhongy
i
Hu, Yukun Bao, Ray
m
ond
Chiong,
and Tao Xiong, "Midterm
interval load forecasting using m
u
lti-output
support vector regression
with a memetr
ic algorithm
for feature
selection",
Energy
,
vol. 84 (2015),
pp.
419-431,
May
2015.
[22]
Woo-Joo Lee, Jink
y
u
Hong, "A
H
y
brid
d
y
namic
and fuzzy
time s
e
ries model for
mid-term
power
load for
e
casting"
,
International Jo
urnal of
Electrical Pow
e
r and
En
ergy Systems
, vo
l. 64
(2015), pp.
1057-1062, 201
5.
[23]
Riswan Efendi, Zuhaimy
Ismail, and Must
afa
Mat Deris, "A new linguistic out-s
ample approach of fuzzy
time series
for daily
forecas
ting of M
a
lay
s
ian electricity
load dem
a
nd",
Applied Soft Computing
, vol. 28 (2015), pp.
422-430,
2015.
[24]
Torkzadeh Roozbeh, Mirzaei Ahm
a
d,
Mirjalili Moham
m
a
d Mehdi,
Anaraki Alireza, S
e
hhati
Moham
m
a
d Reza,
and
Behdad F
a
rideh, "Medium
term
load forecasting in
di
stribution sy
stem
s based on
m
u
lti linear regression
and
principal
component analy
s
is: A
novel approach",
2014
19th Conference on
electrical power distribution networks
,
pp. 66-70, 2014.
[25]
Toly
Chen, and
Yu-Cheng W
a
ng,
"Long-term
load for
ecasting
by
a
collaborative fuzzy
-neural approach",
International Journal of Electrical
Power and Energy Systems
, vol. 43, no. 1, pp. 454-464, Decem
ber 2012.
[26]
Rustum
Mam
l
ook, Om
ar Badran,
and Em
ad Abdulhadi,
"A fu
zzy
inference m
odel
for short-term
load
forecasting",
Energy policy
, vol. 37, no. 4, pp. 1239-1248, April 2009.
[27]
Ming-Yue
Zhai, "A new m
e
thod
for short-term
load fo
recasting based
on fractal interpretation and wavelet
analy
s
is
",
International Journal of
Electrical Power and Energy Systems
, vol. 69 (2015), pp. 241-245, 2015.
[28]
G. S
udheer, A. S
u
s
eelatha,
"S
hort term
load
forecas
ting
us
ing wavelet trans
f
orm
com
b
ined with Holt-W
i
nters
and
weighted
nearest neighbor models",
International Journal of Elect
rical Power and Energy Systems
, vol. 64 (2015),
pp. 340-346, 2015.
[29]
Nian Liu, Qingfeng Tang, Jianhua Zha
ng, W
e
i Fan, and Jie Liu, "A Hy
br
id forecasting m
odel with param
e
ter
optim
ization for s
hort-term
load forecas
ting of m
i
cro-grids
"
,
Applied Energy
, vol. 129 (2014), pp. 336-345, 2014.
[30]
Efstathios
P
a
paroditis and Theofanis
S
a
patinas, "S
hort-te
rm
load
forecasting: The sim
ilar
shape functional tim
e-
s
e
ries
predictor",
IEEE T
r
ans.
Powe
r Sy
st.
, vol. 28, no. 4, pp. 3818-3825, November 2013.
[31]
Shu Fan, Luonan Chen, and
W
e
i-Jen Lee, "Short-term
lo
ad forecasting
using com
p
rehensive com
b
ination based
on
m
u
ltim
eteorological inform
ation",
IEEE T
r
ans. Ind. App.
, vol. 45, no. 4, pp. 1460-1466, July
/August 2009.
[32]
Dam
i
en
Fay
,
and John V.
Ringwood, "On the influence
of
weather forecast errors
in
short-term
load forecasting
models",
IEEE T
r
ans. Power
Syst.
, vol. 25, no. 3, pp. 1751-1758, August 2010.
[33]
Yang W
a
ng, Qing Xia, and Chongqing Kang,
"Secondary
forecasting based on devi
ation analy
s
is for
short-term
load forecas
ting",
IEEE T
r
ans.
Powe
r Sy
st.
, vol. 26, no. 2, pp. 500-507, May
2011.
[34]
Xiaorong
Sun, Peter B. Luh, and Kw
ok W.
Cheung, "An efficient approach
for short-term substation load
forecas
ting",
2013 IEEE Power
and energy society general meeting
, pp. 1-5, 2013.
[35]
Adel
Abdoos, Moham
m
a
d Hem
m
a
ti, Ali A
kbar Abdoos, "Short term
load for
ecasting using a hy
brid intelligent
method",
Knowledge-Based
Systems
, vol. 76 (2015), pp. 139-147, March 2015.
[36]
Pandey
A.S., Singh D.,
Sinha S.K., "Intelligent hy
brid
wavelet m
odels for short-term
load forecasting",
IEEE T
r
ans.
Powe
r Sy
ste
m
s
, vol. 25, no. 3, pp. 1266-1273, 2010.
[37]
Tao
Hong, Pu W
a
ng, and
Laura W
h
ite, "W
eather st
ation selection
for electric load forecasting",
International
Journal of Forecasting
, vol. 31 (2015), pp. 286-295, 2015.
[38]
Liy
e
Xiao, Jianzhou wang,
Ru Hou, and
Jie Wu, "A co
mbined
model based on
data pre-analy
s
is and
weight
coefficients
optim
ization for electrical load forecas
ting",
Energy
, vol. 82 (2015), pp. 524-549, 2015.
[39]
Gu
Chaojun, Dazhi Yang, P
a
nida
Jirutitijaroen, W
ilfred M.
W
a
lsh, and
Thom
as Reindl, "S
patial load forecasting
with
com
m
unication failure us
ing time-forward kriging",
IEEE T
r
ans.
Power Syst.
, vol. 29, no. 6, pp. 2875-2882,
November 2014.
[40]
Franklin L.
Quilum
b
a, W
e
i-Jen
Lee,
Heng Huang,
David Y. Wa
ng,
and Robert
L. Szabados,
"Using smart meter
data
to im
prove the accuracy
of
intraday
load for
ecasting considering custorm
e
r
behavior sim
ilarities",
IEEE T
r
ans.
Smart grid
, vol. 6, no. 2, pp. 911-918, January
2014.
[41]
Zhiheng Zhang, and Shijie Ye, "Long term
load forecasti
ng and recom
m
e
ndations for China based
on
support
vector
regression",
2011 International
conference
on information m
anagement, innovation
management
and industrial
engineering
, pp. 597-602, 2011.
[42]
Tao Hong, W
ilson J.,
and Jingrui Xie, "Long
term
proba
bilistic load forecasting
and norm
a
lization with
hourly
information",
IEEE T
r
ans.
Smart Grid
, vol. 5, no. 1, pp. 456-462, 2014.
[43]
Tao Hong, Jason
W
ilson, and
Jingrui
Xie, "Long
term
probabilistic
load forecasting
and norm
a
lization
with hourly
information",
IEEE T
r
ans.
Smart grid
, vol. 5, no. 1, pp. 456-462, January
2014.
[44]
W
e
n-Chen Chu, Yi-P
ing Chen, Zheng-W
e
i Xu, W
e
i-Jen
Lee, "Multiregion short-term
load forecasting in
consideration
of HI and load/weather diversity
",
IEEE T
r
ans. Ind.
App.
, vol. 47, no. 1, pp. 232-237, January
/
February
2011.
[45]
Kam
a
l Yavarian,
Farid Hashem
i,
and Am
ir
Moham
m
a
dian,
“
D
esign of
intelligent PID
controller for
AVR sy
stem
us
ing an adaptive neuro-fuzzy
inference s
y
s
t
em
”,
Internation Journal of Electri
cal and Computer Engineering
, vol.
4, no. 5, October 2014.
[46]
May
a
devi
N., Vinodchandra S.S., S. Us
hakumari, "A review on expert sy
st
em application is power plants",
Internation Journal of Electri
cal and Computer Engineering
, vol. 4, no. 1, February
2014.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJECE
ISS
N
:
2088-8708
El
ect
ri
ci
t
y
Peak Lo
a
d
Dem
a
n
d
usi
n
g
De-
n
oi
si
ng
Wavel
et
T
r
an
sf
or
m i
n
t
e
g
r
at
ed
w
i
t
h
…
(Pitu
k Bunno
on)
20
BIOGR
AP
H
Y
O
F
AUTH
O
RS
Pituk
Bunnoon was
born in Songkhla,
Thailand, on
Se
ptem
ber 4, 1973.
He received
the B.S. in
electrical engineering
from
Kingm
ongkut’s
Institute
of Technology
Ladkrabang, Thailand, in
2000;
M.S.degree in electrical engineering, and
the
Ph.D. degree in electrical engineering from
Price
of Songkla University
, Songkhla, Thailand,
in
2005 and 2013, respectively
.
He is currently
working
as
lecturer in the electrical engineer
ing department of Rajamangala University
of
Technology
Srivijay
a
, Songkhla, Thailand. His
re
search interests include applied artificial
intelligence to the power sy
stem
.
Evaluation Warning : The document was created with Spire.PDF for Python.