Internati
o
nal
Journal of Ele
c
trical
and Computer
Engineering
(IJE
CE)
Vol.
4, No. 6, Decem
ber
2014, pp. 909~
922
I
S
SN
: 208
8-8
7
0
8
9
09
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
Identifying Appliances using
NIALM with Minimum Featu
r
es
Suni
l
Semw
al
*, R S
Pr
as
ad
**
* Departement o
f
Instrumentatio
n and Con
t
rol Engin
eer
ing, Grap
hic
Era University
Uttarakhan
d
, I
ndia
** Depart
em
ent
of El
ectr
i
c
a
l
and
El
ectron
i
cs
Eng
i
neer
ing,
Graphi
c Er
a Univ
ers
i
t
y
Uttarakh
and,
Ind
i
a
Article Info
A
B
STRAC
T
Article histo
r
y:
Received Sep 19, 2014
Rev
i
sed
O
c
t 24
, 20
14
Accepte
d Nov 5, 2014
Government of
India has decided to inst
all sm
art m
e
ters in fourteen states
.
Sm
art m
e
ters are requir
e
d to id
e
n
tif
y hom
e appl
i
a
nces to fu
lfil
l v
a
rious tasks
in the smart grid environment.
Both
intrusive
and non-intrusiv
e methods
have been suggested for id
entif
icati
on. Howev
e
r, intrusiv
e method is not
suitable for
cost and pr
ivacy
reasons.
On the o
t
her h
a
nd,
techn
i
ques using
non-intrusive applian
ce lo
ad
monitoring (NIALM) are
y
e
t to result in
m
eaningful pra
c
ti
cal im
plem
en
tation
.
Two m
a
jor chall
e
nges
i
n
NIALM
res
earch
ar
e th
e
choi
ce of
fe
atu
r
es
(load
s
i
gnat
u
res
of app
lian
c
es
), and
th
e
appropria
te
algo
rithm
.
Both h
a
v
e
a d
i
re
ct
im
pact on
the
cos
t
o
f
the s
m
ar
t
meter. In
this paper, we
address
the
two issues and propose a pro
cedure with
only
four featu
r
es and a sim
p
le algorithm to identif
y
app
l
iances. Our
experimental setup, on th
e recommende
d specif
i
cations of
the internal
electrical wiring
in Ind
i
an r
e
sid
e
nces, used
common household
appliances’
load sign
atures
of active and
r
eac
tive powers, harmonic
components an
d
their m
a
gnitud
e
s
.
W
e
s
how t
h
at thes
e four
featur
es
are e
s
s
e
ntial an
d
s
u
fficien
t for im
plem
enta
tion of
NIALM
with a
simple algorith
m
. We have
introduced a new approach
of ‘
m
ulti point
sensi
ng’ and ‘group
control’ r
a
ther
than the ‘single point sensing’ and
‘individual control’
, used
so far in
NIALM techniques.
Keyword:
Lo
ad
sign
atur
e
NI
ALM
Sm
art g
r
id
Sm
art m
e
ter
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
:
Sunil Sem
w
al,
Depa
rt
em
ent
of I
n
st
r
u
m
e
nt
at
ion
an
d c
o
nt
r
o
l
engi
nreei
ng
,
G
r
aph
i
c
E
r
a Un
iv
er
s
ity,
56
6/
6
B
e
l
l
R
o
a
d
C
l
em
ent
o
wn
,
De
hra
d
un
,
Ut
t
a
rak
h
a
n
d
,
I
n
di
a.
Em
a
il: su
n
il.111
213
@g
m
a
il.c
o
m
1.
INTRODUCTION
A No
n-in
tru
s
ive App
lian
ce Lo
ad
Mon
itoring
(NIALM) meth
od
is d
e
signed
to
m
o
n
ito
r electrical an
d
electronic a
p
pliances whe
n
they are in
use.
This is carried out
by a
n
ana
l
ysis of the e
x
t
r
acted
features
(loa
d
si
gnat
u
res
)
o
f
t
h
e ap
pl
i
a
nces.
Thi
s
an
al
y
s
i
s
est
i
m
a
t
e
s nat
u
re of i
ndi
vi
d
u
al
l
o
ad si
gnat
u
re
s of t
h
e a
ppl
i
a
nce
s
form
in
g
th
e com
b
in
atio
n
in
switch
e
d
ON st
ates. In
t
r
u
e
NIALM op
eration
,
h
u
m
an
in
terv
en
tion
is
n
o
t
n
eed
ed
for id
en
tificatio
n
o
f
ind
i
v
i
dual ap
p
lian
ces at th
e circu
it b
r
eak
er lev
e
l. Th
u
s
, it facilitat
e
s a v
e
ry co
nven
i
en
t
m
e
t
hod
of
col
l
ect
i
ng i
n
fo
rm
at
i
on a
b
o
u
t
t
h
e
use
of a
p
pl
i
a
nc
es, vi
z,
t
y
pe o
f
t
h
e ap
pl
i
a
nces
an
d t
h
ei
r
p
o
w
e
r a
n
d
ener
gy
co
ns
um
pt
i
o
n
s
p
r
o
f
i
l
e
.
Thi
s
i
n
fo
rm
ation
pr
o
v
i
d
es
d
a
ta wh
ich
are ex
trem
ely v
a
lu
ab
le to
all th
e p
l
ayers,
v
i
z., con
s
u
m
ers,
u
tilities, an
d app
lia
n
ce m
a
n
u
facturers for th
e
pu
rpo
s
e
of
u
p
g
r
ad
i
n
g the p
e
rfo
r
m
a
n
ce wh
en
and
whe
r
e nee
d
ed. Using NIALM, a sm
art
m
e
ter placed
outsi
de a home can dete
rm
ine how
m
u
ch energy
goe
s i
n
t
o
eac
h
ap
pl
i
a
nce i
n
t
h
e
resi
de
nce.
The c
o
m
p
l
e
xi
t
y
of
i
n
st
r
u
m
e
nt
at
i
on i
n
sm
art m
e
t
e
r de
vel
o
pm
ent
increases
with increase in the
num
ber
o
f
featu
r
es wh
ich
resu
lts in
in
crease in
th
e sen
s
o
r
s for ex
tracting
lo
ad
si
gnat
u
res
.
Al
s
o
,
desi
g
n
of t
h
e al
go
ri
t
h
m
impact
s com
p
lex
ity, as in
vo
lv
ing
larg
e nu
m
b
er of feat
u
r
es wo
ul
d
not
hel
p
i
n
de
v
e
l
opi
n
g
a si
m
p
l
e
al
gori
t
h
m
.
B
o
t
h
har
d
ware
r
e
qui
rem
e
nt
and feat
ures
use
d
i
n
al
go
ri
t
h
m
have an
im
pact
on t
h
e ul
t
i
m
a
t
e
cost
of a sm
art
m
e
t
e
r. Fr
om
pract
i
cal
consi
d
er
at
i
ons, i
t
i
s
im
port
a
nt
and
hi
g
h
l
y
essen
tial, p
a
rticu
l
arly fo
r m
a
j
o
rity
o
f
co
ns
u
m
ers i
n
de
vel
o
pi
n
g
c
o
u
n
t
r
i
e
s
l
i
k
e In
di
a a
nd
In
di
an
su
bc
ont
i
n
at
al
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJECE Vol. 4, No. 6, D
ecem
ber 2014
:
909 – 922
91
0
[1
], to
en
su
re t
h
at th
e co
st
o
f
meter is with
in
th
e fin
a
n
c
ial cap
acity o
f
su
ch co
n
s
u
m
ers. It is n
o
t
surp
rising
th
at
i
n
spi
t
e
of
seve
ral
t
ech
ni
q
u
es
pr
o
pose
d
i
n
[
1
]
-[9]
,
n
o
t
eve
n
one
has
f
o
un
d
i
t
s
pract
i
cal
ap
pl
i
cat
i
on si
nce
no
n
e
meets
the
des
i
red goals of the use of
s
m
art
m
e
ters
fu
lly and
so
has no
t
p
r
ov
ed wo
r
t
h
y
of
practical
im
pl
em
ent
a
t
i
o
n. Fr
om
consu
m
ers’ p
o
i
n
t
of
vi
ew, sm
art
m
e
t
e
rs of t
o
day
have
pr
o
v
en t
o
be a bad deal
[
1
0
-
12]
.
W
i
t
h
o
u
t
exce
p
t
i
on,
one
of t
h
e m
a
jor reas
ons
w
h
i
c
h i
s
resp
o
n
si
bl
e f
o
r com
p
l
e
x ha
r
d
wa
re an
d c
o
m
p
l
e
x
so
ft
ware i
n
imp
l
em
en
tin
g
NIALM techn
i
ques, propo
sed
till
d
a
te, is th
e use o
f
a
sing
le p
o
i
n
t
curren
t
sen
s
ing
at
t
h
e i
n
com
i
ng m
a
i
n
s suppl
y
t
o
t
h
e m
e
t
e
r. The cu
rre
nt
se
nso
r
se
nses t
h
e
wave
fo
rm
of a
m
i
xt
ure o
f
el
ect
ri
cal
and el
ect
r
o
ni
c resi
de
nt
i
a
l
app
l
i
a
nces w
h
i
c
h
gene
ral
l
y
ope
r
a
t
e
on
di
ffe
re
n
t
pri
n
ci
pl
es [
1
0]
. Fo
r a m
i
ddl
e cl
ass
In
di
an
fam
i
l
y
the
n
u
m
b
er o
f
s
u
ch
ho
use
hol
d
appl
i
a
nc
es
do
e
s
ha
rdl
y
e
x
cee
d t
h
i
r
t
y
fi
ve
,
w
h
ereas
i
n
t
h
e
U
n
i
t
e
d
States the number m
a
y always be m
o
re
than
fifty
.
The c
u
r
r
e
nt wa
ve f
o
rm
(
C
W
)
sign
ature o
f
h
ybr
id
v
a
riety o
f
appl
i
a
nc
es i
s
b
o
u
n
d
t
o
be m
o
re an
d m
o
re c
o
m
p
l
e
x wi
t
h
i
n
c
r
easi
n
g
num
ber o
f
devi
ces
.
As a c
o
n
s
eq
ue
nce, i
t
becom
e
s essential to use analytical tools requi
ring a
l
a
rg
e num
ber o
f
f
eat
ures a
nd ri
go
r
ous m
a
t
h
em
at
i
cal
alg
o
rith
m
s
to
d
i
sagg
reg
a
te th
e lo
ad
s [2
]. A
m
a
j
o
r im
p
e
d
i
men
t
to
th
e g
r
o
w
t
h
of Ind
i
a’s eco
no
m
y
is t
h
e wi
de
gap
i
n
t
h
e
de
m
a
nd a
nd
su
p
p
l
y
of
p
o
w
er
[
13]
.
Acc
o
r
d
i
n
g t
o
a
re
po
rt
[
14]
,
I
ndi
a’
s ag
gre
g
at
e t
ech
ni
cal
an
d
com
m
e
rci
a
l
(AT & C
)
l
o
ss
es are am
ong
t
h
e hi
g
h
est
i
n
t
h
e w
o
rl
d
,
ave
r
agi
ng
28
.4
4%
(2
00
8
-
0
9
dat
a
). T
h
e
fina
ncial loss has bee
n
estimated at 1.5% of the na
tional GDP, and is growi
ng stea
dily. Faced wi
th this
situ
atio
n
,
Gov
e
rn
m
e
n
t
o
f
In
d
i
a (GOI) set up
a task
force for i
m
p
l
e
m
en
tat
i
o
n
of sm
art
met
e
rs in
stallatio
n p
ilo
t
pr
o
j
ect
s i
n
I
ndi
an
dom
est
i
c
and i
n
d
u
st
ri
al houses i
n
a
phase
d
m
a
nner, start
i
ng
with
selected a
r
eas in fourteen
states o
f
th
e unio
n
[15
]
. Th
is
d
ecision
of
a po
w
e
r
-
d
e
f
i
cien
t I
n
d
i
a h
a
s
p
r
o
v
i
d
e
d
en
ough
opp
or
tun
ity f
o
r
near
ly
h
a
lf a do
zen
mu
ltin
atio
n
a
l com
p
an
ies to
set
u
p
th
eir un
its i
n
Ind
i
a for
p
r
od
u
c
tion
of sm
a
r
t m
e
ters [1
6
]
.
Tabl
e 1.
Li
g
h
t
l
o
ad
a
n
d po
we
r
a
ppl
i
a
nces
Sr.
Light Load Appli
a
nce
Sr.
Light Load Appli
a
nce
Sr.
Pow
er Appliance
s
1
T
L
(
40W)
7
T
ube light (
40W
)
14
Air
conditioner
2 I
L
(
60W
)
8
DSO
15
I
nduction
Coo
k
er
3 CFL
9
CPU
16
M
i
cr
o-
oven
4
L
a
ptop
10
Function Gener
a
tor
17
Roo
m
Heater
5
M
onitor
11
M
obile char
ger
(
N
okia)
18
E
l
ectr
i
c Press
6 Fan
12
T
V
---
--
13
Fr
idge
1.
1 Sm
ar
t
Met
ers fr
om I
ndi
a’s
Perspec
t
i
v
es
C
onsi
d
eri
n
g
t
h
e be
nefi
t
s
of
s
m
art
gri
d
-sm
a
rt
m
e
t
e
r t
echn
o
l
ogy
,
vi
z.
,
red
u
c
t
i
on
of
AT
&
C
l
o
sses a
n
d
p
r
ev
en
tio
n
o
f
o
u
t
ag
es, ap
art fro
m
re
m
o
te a
n
d
tran
sp
ar
en
t
en
erg
y
b
illin
g, th
e task
force h
a
s id
en
tified
th
e
fun
c
tion
a
lities o
f
sm
art
m
e
ter
s
in
each
o
f
th
e selected
areas
in
fou
r
teen states as sh
own
b
e
lo
w:
AM
IR
-
A
dva
nc
e M
e
t
e
r In
fr
a St
ruct
ure
fo
r R
e
si
de
nt
i
a
l
cons
um
ers, AM
I
I
-
A
d
v
a
n
ce M
e
t
e
r
In
frast
ruct
ure
fo
r I
n
d
u
strial con
s
um
ers; OM
-O
utag
e M
a
nagem
e
nt
;
PLM
-
Peak L
o
a
d
M
a
nagem
e
nt
; PQM
-
Power
Qu
ality Man
a
g
e
m
e
n
t; MG-Micro
-g
ri
d
Man
a
g
e
m
e
n
t; DG-Distri
b
u
t
ed
Gen
e
ration
Man
a
g
e
m
e
n
t
.
IEEE Standa
rds Association,
recogniz
ing India as the third largest
m
a
rket for sm
art
m
e
ter, has also
p
i
n-po
in
ted
fou
r
k
e
y ch
allen
g
e
s for sm
art g
r
id
adop
tio
n
in
In
d
i
a,
v
i
z., po
wer th
eft, in
ad
equ
a
te g
r
id
i
n
fra
st
ruct
ure
,
l
o
w m
e
t
e
ri
ng e
ffi
ci
ency
, a
n
d l
ack
of a
w
a
r
ene
ss am
ong t
h
e c
ons
um
ers [
17]
.
Sm
art
m
e
t
e
r b
e
i
n
g
an i
n
t
e
gral
c
o
m
ponent
of a
sm
art
m
i
cro-g
r
i
d
(
d
i
s
t
r
i
b
ut
i
o
n
net
w
or
k)
, I
n
d
i
a i
s
expect
e
d
t
o
i
n
st
al
l
13
0
m
i
ll
i
on
sm
art
meters
b
y
20
20
[18
]
. Th
e ta
sk
force
com
m
i
ttee has also stresse
d
on t
h
e de
vel
o
pm
ent
of a l
o
w cos
t
m
e
t
e
r, wi
t
h
out
com
p
rom
i
si
ng
o
n
t
h
e
assi
gn
ed
fu
nct
i
o
ns
of th
e m
e
ters, so
th
at
th
es
e a
r
e easily affordable to
the significa
ntly large cons
umers in
m
i
ddle and lower i
n
com
e
groups.
Accordi
ng to a recent report [11],
Germ
any
has t
a
ken
a deci
si
o
n
t
o
re
ject
t
h
e
recom
m
endat
i
ons
o
f
t
h
e
Eu
r
opea
n
U
n
i
o
n (
E
U)
f
o
r i
n
st
al
l
a
t
i
on
of
sm
art
meters in
80%
of
hom
es by
2020 be
c
a
use it is lik
ely to prove
too c
o
stly fo
r c
ons
umers. Recently [19],
according
to MOU
si
gne
d betwee
n
GOI and Ge
rm
any,
technical assi
stance from
Germ
any will e
n
able
evacuat
i
on o
f
nearl
y
30
,0
0
0
M
W
of sol
a
r and wi
n
d
po
we
r fo
r in
tegration in
to
th
e n
a
tio
n
a
l g
r
id. GOI is also
encouragi
ng t
h
e cons
um
ers to use t
h
e roof-top s
o
lar
cel
l
s
, t
o
m
eet
t
h
ei
r o
w
n
dem
a
nds
du
ri
n
g
pea
k
l
o
a
d
situ
atio
n
s
.
W
i
t
h
th
is scen
arion
in v
i
ew, t
h
e
o
b
j
ectiv
es
w
h
i
c
h
have
bee
n
s
e
t
fo
r sm
art
m
e
t
e
rs i
n
I
ndi
a a
r
e:
(i) m
o
n
ito
ri
n
g
app
lian
ces’ l
o
ad an
d en
ergy p
r
o
f
ile, (ii)
in
teg
r
ating
i
n
t
e
rm
i
tten
t
an
d
d
i
stribu
ted sources
o
f
en
erg
y
, (iii) mo
n
itoring
h
ealth
of app
lian
ces an
d
equ
i
p
m
en
ts in
sid
e
and
ou
tsid
e (po
w
er
q
u
a
lity) th
e
p
r
e
m
ises,
(iv
)
tran
sp
arent an
d
rem
o
te real ti
me en
ergy b
illin
g
o
f
con
s
u
m
ers, and
(v
) en
ab
li
n
g
sel
f
-h
ealin
g
of grid
s to
p
r
ev
en
t/ redu
ce freq
u
e
n
c
y of
o
u
t
ag
es.
A look
at th
e d
e
sire
d o
b
j
ectiv
es i
n
(i), (iii), an
d
(v) ab
ov
e,
req
u
i
res th
at
co
nsu
m
ers’ lo
ad
s m
u
st b
e
id
e
n
tified
.
If so
m
e
o
f
th
e id
en
tified
ap
p
lian
c
es are requ
ired
to
b
e
switch
e
d
OFF for
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Id
en
tifyin
g
App
lia
n
c
es u
s
ing
NIALM
with
Min
i
mum Fea
t
u
r
es
(S
un
il
S
e
mw
a
l
)
91
1
reason
s i
n
clud
i
n
g con
s
u
m
p
tio
n
o
f
h
i
gh
po
wer during
cr
itical p
eak lo
ad
co
nd
itio
ns, t
h
is
o
p
e
ration
h
a
s t
o
b
e
carried
ou
t rem
o
te
ly b
y
th
e
u
tility. Alth
o
u
g
h
, in
literatu
re, m
a
n
y
NIALM tech
n
i
qu
es
h
a
v
e
b
e
en
repo
rted
[1
]-
[9]
,
t
h
ere
i
s
n
o
di
scussi
on
r
e
gar
d
i
n
g
t
h
e m
e
t
hod use
d
f
o
r swi
t
c
hi
ng
(
O
FF/
ON
) faul
t
y
and/
o
r
hi
g
h
p
o
we
r
co
nsu
m
in
g
ap
plian
ce(s) in
critical lo
ad
con
d
i
tio
n
s
.
Si
nce t
h
e
r
e
ha
s t
o
be
no
us
e of
se
nso
r
s
(
N
I
A
LM
)
i
n
t
h
e ap
pl
i
a
nces’
out
l
e
t
s
, i
t
wo
ul
d
re
qui
re
consum
er’s ass
i
stance for s
w
itching
op
erations. But t
h
ere is
no
guara
n
tee
t
h
at the
cons
umer
would be
present
in
h
i
s/h
e
r residen
ce at th
e ti
me o
f
th
is switch
i
ng
op
eration
.
If th
e u
s
e of sen
s
o
r
s in
th
e ou
tlets o
f
po
ten
t
ially
hi
g
h
po
we
r c
o
nsum
i
ng a
p
pl
i
a
nces i
s
co
nsi
d
ered
u
n
a
voi
da
bl
e, t
h
i
s
am
ount
s t
o
i
n
t
r
usi
v
e
m
e
t
hod
.
A
n
e
x
am
pl
e
o
f
i
n
tru
s
iv
e m
e
th
od
can
b
e
seen
in th
e Fl
orid
a stat
e in USA
where
u
tilities h
a
v
e
i
n
serted
sensors i
n
so
m
e
p
o
t
en
tially h
i
g
h
po
wer con
s
umin
g
app
lian
c
es. Th
ere is a
p
r
i
o
r written
ag
reem
en
t with
th
e con
s
u
m
ers th
at, in
case it is req
u
i
red, th
ese ‘h
armin
g
’
sm
art ap
p
lian
c
es w
ill b
e
switch
e
d
OFF/ ON d
i
rectl
y
b
y
th
e u
tility
. Th
is
t
echni
q
u
e m
a
y be q
u
i
t
e
sat
i
s
fact
o
r
y
fo
r
U
S
A a
nd
ot
he
r
ri
ch a
nd
de
v
e
l
ope
d co
u
n
t
r
i
e
s, b
u
t
m
a
y
n
o
t
be
acceptable a
n
d practicable
for the c
ons
um
e
r
s in India be
c
a
use
of c
o
st involve
d
in t
h
e
use
of se
ns
ors a
nd
pri
v
acy issues. This scenari
o
bri
ngs
us
in
to
th
e subj
ect
m
a
tt
er o
f
o
u
r stud
y in
th
is p
a
p
e
r. We add
r
ess all th
ese
issues usi
ng
NIALM with
non intrusiv
e feat
ure ext
r
action
(NIFE
)
techni
que.
Ming et al in their recent
pape
r
[8]
ha
ve st
ress
ed t
h
at
NI
AL
M
and N
I
FE a
r
e t
w
o
very
di
f
f
ere
n
t
but
rel
a
t
e
d t
ech
ni
q
u
es.
The l
i
m
i
t
a
t
i
ons
of t
h
e
exi
s
t
i
ng
NI
AL
M
t
echni
q
u
es
have
bee
n
ex
p
o
se
d i
n
a re
vi
e
w
pa
per
[1
0]
. In t
h
e e
x
i
s
t
i
n
g
NI
ALM
t
echn
i
ques
,
m
u
l
tip
licit
y o
f
featu
r
es and
al
g
o
rith
m
s
cau
se seriou
s co
nf
us
i
ons
fo
r t
h
e
sm
art
m
e
t
e
r devel
ope
rs.
I
n
t
h
i
s
p
a
per
,
we sh
ow t
h
at
usi
n
g o
n
l
y
st
eady
st
at
e val
u
es of act
i
v
e po
wer P, rea
c
t
i
v
e po
wer Q
,
ha
rm
oni
c com
ponent
s h
,
an
d
th
ei
r m
a
g
n
itu
d
e
s m
h
, it
is p
o
ssib
l
e to
id
en
tify th
e app
lian
ces correctly. In
fact we h
a
v
e
m
o
d
i
fied
our
stu
d
y
in
th
is
p
a
p
e
r fro
m
th
e p
r
ev
iou
s
ly rep
o
rted
p
r
elimin
ary find
ing
s
repo
rted
in
[2
0
]
wh
ere th
i
r
teen
appl
i
a
nc
es
wer
e
use
d
a
n
d l
o
ad si
gnat
u
res
were
ext
r
act
e
d
usi
n
g
o
n
l
y
a
si
ngl
e c
u
r
r
e
n
t
sens
or
pl
ace
d
at
t
h
e
in
co
m
i
n
g
m
a
in
s sup
p
l
y to
the
m
e
ter (sam
e a
s
in
all NIALM tech
n
i
qu
es). Algo
rith
m
u
s
ed
[20
]
was
b
a
sed
on
te
m
p
lates
mat
c
h
i
ng
with
the u
n
k
nown
com
b
in
atio
n
o
f
appl
i
a
nc
es o
p
e
r
at
i
ng si
m
u
l
t
a
neo
u
sl
y
.
H
o
w
e
ver
,
a
seriou
s lim
i
t
at
io
n in
ou
r prev
iou
s
st
ud
y was th
e req
u
i
remen
t
o
f
larg
e
sto
r
ag
e sp
ace
in
m
e
m
o
ry. Fo
r 13
-
appl
i
a
nc
es (si
n
gl
e-
poi
nt
sensi
n
g
)
, t
h
e n
u
m
b
er of t
e
m
p
l
a
t
e
s
fo
r al
l
possi
bl
e
appl
i
a
nce com
b
i
n
at
i
o
ns was
81
9
1
.
For
t
h
e
f
o
ur
fe
at
ures
case,
t
h
e t
e
m
p
l
a
t
e
s wo
ul
d
be
as m
a
ny
t
i
m
e
s hi
ghe
r.
Thi
s
i
s
not
c
o
nsi
d
e
r
ed
a
n
e
f
f
i
ci
ent
and practical
approach despit
e
giving
accurate identification, al
so commented in [2]. T
h
ere
f
ore, in the first
p
a
rt of th
e st
u
dy, in
th
is p
a
p
e
r, we first ex
amin
e wh
ether t
h
e specifications of in
te
rnal el
ectrical wiring
(IE
W
)
o
f
Cen
t
ral Pu
blic Wo
rk
s
Departm
e
n
t
(CPWD),
a GOI
or
gan
i
zatio
n, can
facilitate a ‘n
atu
r
al’
d
i
sagg
reg
a
ti
on
o
f
th
e lo
ad
s i
n
ord
e
r to
enab
le th
e p
r
o
c
ed
ure of tem
p
l
a
tes
m
a
tch
i
n
g
p
r
actical fo
r
th
e Ind
i
an
d
o
mestic
resi
de
nces.
Si
n
ce t
h
e speci
fi
c
a
t
i
ons
of
IE
W
pr
o
v
i
d
es
di
st
ri
but
i
o
n
of a
p
pl
i
a
nces i
n
a
gr
o
up i
n
sev
e
ral
p
a
ral
l
e
l
circu
its,
we
h
a
v
e
u
s
ed
m
u
lti-p
o
i
n
t
sen
s
ing of C
W
rath
er
th
an th
e si
ng
le po
in
t m
o
d
e
of sen
s
i
n
g as
applied
in
reporte
d
works
on
NIALM [1]
-[9],[20]. W
e
explain
later
(next section) t
h
e bi
g adva
ntages that accrue
whe
n
m
u
lt
i
poi
nt
sen
s
i
ng i
s
use
d
, a
nd
we cl
aim
t
h
at
t
h
i
s
m
e
t
hod
i
s
repo
rt
ed f
o
r t
h
e fi
rst
t
i
m
e
in t
h
i
s
pap
e
r. I
n
t
h
e
seco
nd
part
o
f
ou
r st
u
d
y
,
we
sho
w
t
h
at
i
t
i
s
pos
si
bl
e
to
su
ccessfu
lly id
en
tify th
e app
lian
c
es co
rrectly, ex
cep
t
th
at th
e co
n
t
rol o
p
e
ration
(ON/OFF) of applian
ces, on
th
e d
i
rectio
n
o
f
utili
ty, w
ill b
e
p
e
rform
e
d
b
y
‘group
co
n
t
ro
l’
rath
er th
an
b
y
‘ind
iv
idu
a
l con
t
ro
l’. Our exp
e
ri
men
t
s, with
t
w
en
ty fi
v
e
‘lig
h
t
’ lo
ad
app
l
ian
ces,
d
i
stribu
ted in
fou
r
parallel circu
its, and
fiv
e
‘power
’ l
o
ad a
p
pliances, connected
i
n
th
ree
p
a
rallel po
wer
circuits, e
x
actly as per the
IE
W speci
fications,
ha
ve
be
en c
o
nducted
in laborat
o
ry
environm
ent, wit
h
co
nstan
t
supp
ly v
o
ltag
e
en
sured. For id
en
tificatio
n
,
we hav
e
fo
llo
wed
the p
r
o
c
ed
ure of te
m
p
lates
m
a
t
c
h
i
ng
[7
] d
e
sp
ite th
e criticis
m
s
b
y
so
m
e
au
th
o
r
s
[2
].
We sho
w
th
at th
eir criticis
m
o
f
co
m
p
arison
m
e
th
o
d
as an
i
n
effi
ci
e
n
t
an
d
im
pract
i
cal
appr
oac
h
d
o
es
no
t
appl
y
t
o
t
h
e
m
u
lt
i
p
l
e
poi
nt
sensi
n
g.
Fo
r e
x
am
pl
e, i
n
case of
2
5
ap
p
lian
c
es case, u
s
ed
in
t
h
is p
a
p
e
r, t
h
e to
t
a
l o
f
all po
ssi
b
l
e co
m
b
in
ation
s
works ou
t to
ov
er 33
.5
millio
n
s
requ
iring
as m
a
n
y
te
m
p
lates, if on
ly a sin
g
le cu
rren
t
sen
s
o
r
is
u
s
ed
. Tem
p
la
tes
m
a
tch
i
n
g
in
t
h
is case are
clearly
m
eaningless. In our c
a
se, sin
ce all the appliances are distributed
i
n
a num
ber of
paral
l
e
l
ci
rcui
t
s
, and
the curre
nt se
nsors
use
d
a
r
e
the sam
e
in num
b
er as th
e
circu
its,
o
n
e
sen
s
o
r
for each circu
it, t
h
e to
tal o
f
pos
si
bl
e c
o
m
b
i
n
at
i
ons
d
r
ast
i
cal
l
y
reduc
es t
o
t
h
e s
u
m
of
pos
si
bl
e c
o
m
b
i
n
at
i
ons
i
n
eac
h ci
rc
ui
t
.
Ta
ki
ng
t
h
e
sam
e
exa
m
ple
of
25-a
pplia
nc
es case, if
we
distribute th
em in
fo
ur
p
a
rallel circu
its, say
6
-
app
lian
ces in th
ree
and se
ve
n a
ppl
i
a
nces i
n
one
, t
h
en t
h
e t
o
t
a
l
o
f
al
l
possi
bl
e c
o
m
b
i
n
at
i
ons w
o
rks
o
u
t
t
o
m
e
rel
y
316
w
h
i
c
h i
s
an
in
sign
ifican
t
nu
m
b
er in
co
m
p
arison
, and
so
th
e tem
p
late
s match
i
n
g
is
d
e
fin
itely a practical p
r
o
p
o
s
ition
,
as i
n
[7]
.
We
ha
ve, t
h
ere
f
o
r
e,
co
nsi
d
ere
d
t
h
i
s
ap
pr
oach
i
n
o
u
r e
x
peri
m
e
nt
s. Thi
s
pa
per i
s
o
r
ga
ni
zed as
f
o
l
l
o
w
s
:
next
sect
i
on di
sc
uss
e
s t
h
e sal
i
e
nt
feat
ures o
f
IE
W
speci
fi
cat
i
ons
of C
P
WD;
Sec
t
i
on II
I b
r
i
e
fl
y
revi
e
w
s t
h
e st
at
us o
f
research
u
s
i
n
g
NIALM. Here, we al
so d
i
scuss our ap
pro
a
ch
and
th
e su
itab
ility o
f
im
p
l
emen
tin
g
NIALM on
the ha
rd
ware
p
l
atform
pro
v
id
ed by
the e
x
isting
IE
W sp
eci
ficatio
n
s
. Section
IV d
e
scrib
e
s th
e alg
o
rith
m
,
wh
ile
sect
i
on
V
deal
s wi
t
h
t
h
e e
x
pe
ri
m
e
nt
al
proce
d
u
r
es a
n
d s
o
m
e
assum
p
t
i
ons
use
d
i
n
t
h
e e
x
p
e
ri
m
e
nt
ati
ons.
In t
h
e
n
e
x
t
section
we d
i
scu
s
s th
e resu
lts of id
en
tificatio
n
wi
t
h
som
e
exam
pl
es. Fi
nal
l
y
, we
concl
ude
gi
vi
n
g
o
u
r
anal
y
s
i
s
o
f
t
h
e
expe
ri
m
e
nt
s an
d
resul
t
s
,
an
d
f
u
t
u
re
wo
rk
re
q
u
i
r
e
d
f
o
r f
u
rt
he
r i
m
provem
e
nt
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJECE Vol. 4, No. 6, D
ecem
ber 2014
:
909 – 922
91
2
1.
2 IE
W
Speci
fi
cati
ons
o
f
C
P
WD
(I
ND
IA
)
Specific feat
ures of s
p
ecifica
tions of internal el
ectrical w
i
ring (IE
W
), whic
h are rele
vant to the
expe
ri
m
e
nt
s co
nd
uct
e
d
i
n
t
h
i
s
pape
r,
are
gi
v
e
n
bel
o
w.
Ca
pa
city o
f
circu
its
(i) each lighting circuit (appliance ratin
g not exceeding 6
A) shall be loaded
suc
h
that not
m
o
re than ten suc
h
appl
i
a
nc
es
or
a
m
a
xim
u
m
of 80
0
W,
w
h
i
c
he
ver
i
s
l
e
ss,
are
connected;
in c
a
se of
CFL
a
p
pliances, whe
r
e load
may be less, num
ber of such appliances m
a
y be suita
bly increase
d
without exceed
i
ng a
llowable watta
ge.(ii)
each power circuit
can feed
t
h
e
applian
ces
at the powe
r
outlets suc
h
that
(a
)
not m
o
re than t
w
o appliances
of
16
A
(
1
k
W)
ra
t
i
ngs a
r
e c
o
nn
ect
ed t
o
1
6
A
out
l
e
t
s
,
(c)
in
case of l
o
ads
> 1
kW, these
shall be
controlled by
sui
t
a
bl
y
rat
e
d
m
i
ni
at
ure ci
rc
u
i
t
brea
ker
(M
C
B
)/
swi
t
c
h
and cable
size
s
h
all
be decide
d as per
calculations;
Since each li
ghting ci
rcuit ca
n ha
ve a m
a
xim
u
m of 10
a
p
pliances, t
o
tal powe
r rating not excee
ding 800
W,
t
h
en,
o
n
t
h
e re
al
i
s
t
i
c
assu
m
p
t
i
on t
h
at
a m
i
ddl
e cl
ass resi
den
t
i
a
l
house
doe
s
not
ha
ve m
o
re
t
h
an t
h
i
r
t
y
fi
v
e
6 A
appl
i
a
nc
es, t
h
e
s
e can
be
di
st
r
i
but
ed
i
n
t
h
e
f
o
l
l
o
wi
ng
m
a
nn
er:
W
e
can
di
s
t
ri
but
e se
ve
n a
ppl
i
a
nce
s
i
n
ea
ch
of
th
e fi
v
e
ligh
t
circu
its wit
h
ou
t
lo
ad
i
n
g fu
lly,
k
eep
i
n
g on
e or two sp
are circu
its fo
r fu
ture
n
eeds; sim
i
larl
y for
si
x
po
we
r a
p
pl
i
a
nces
(1
6
A
)
i
n
a
h
o
u
se,
we ca
n
di
st
ri
bu
te th
em
in
three
p
o
wer circu
its, two
i
n
each
.
Th
is
mean
s we
req
u
ire fiv
e
sensors for ligh
ting
circu
its and
t
h
ree
fo
r
po
wer
ci
rc
ui
t
s
. T
h
e c
o
n
d
i
t
i
oned
o
u
t
p
ut
s
fr
om
th
ese sen
s
o
r
s are con
n
ected to
th
e in
pu
t po
rt in
th
e m
i
cro
c
on
tro
ller un
i
t
(MCU), each p
i
n
in
th
e i
n
pu
t po
rt
has a uni
que address for a group of app
liances co
nn
ected
in
a circu
it. Similarly,
th
e sen
s
ors ou
tpu
t
s in
po
wer
circuits will also be connect
ed to a
port in
MCU, provi
d
ing
unique
address of each
group
of applianc
es in a
po
we
r ci
rcui
t
.
Theref
o
r
e, f
o
r t
h
e pu
r
pose
of N
I
ALM
,
w
h
i
c
h i
s
descri
bed
next
,
we need a
n
al
go
ri
t
h
m
t
o
m
o
n
ito
r and
iden
tify a m
a
x
i
mu
m
o
f
on
ly ten ap
p
lian
ces at
a ti
m
e
irresp
ectiv
e o
f
ho
w m
a
ny appliances
are in
u
s
e in
a
ho
use. Th
is
p
r
ov
ides th
e ‘first lev
e
l’
o
f
‘nat
ur
al
’ l
o
ad
di
sa
g
g
re
gat
i
o
n,
wh
i
l
e th
e fin
a
l lev
e
l of
d
i
sagg
reg
a
tion
,
wh
ere i
n
d
i
v
i
du
al app
lian
ces
are id
en
tifie
d,
will b
e
carried
o
u
t
b
y
th
e algorith
m
.
No
te that th
e
task
of d
e
v
e
lop
i
ng
th
e algo
ri
th
m
is ren
d
e
red
easy du
e to
th
is n
a
tural first lev
e
l d
i
sag
g
reg
a
tio
n. A comman
d
would
fetch
on-line t
h
e a
g
gregate sign
al
o
f
a gr
o
u
p
o
f
a
p
pl
i
a
nces
fo
rm
ing t
h
e c
o
m
b
i
n
at
i
on
or
gr
o
u
p
i
n
a
circu
it.
2.
NIALM
Th
e con
cep
t
of NIALM has
b
een fu
lly explain
e
d
in
literatu
re
[1
]-[10
].
Sev
e
ral features fro
m
th
e
basi
c v
o
l
t
a
ge a
nd c
u
rre
nt
si
g
n
a
l
s
(b
ot
h st
ea
d
y
st
at
es a
n
d
tran
sien
ts) su
ch
as activ
e powe
r (P), reactive
powe
r
(Q
), p
o
w
e
r fa
ct
or (
p
f
)
, ha
r
m
oni
c co
m
ponent
s (
h
)
,
m
a
gni
t
u
des o
f
h
(
m
h), cu
rre
nt
wavef
o
rm
(C
W)
, V-
I
traj
ectory lo
ad sig
n
a
tures, eig
e
n
v
a
lu
es, switch
i
ng
tr
ans
i
ent waveform shape,
insta
n
taneous adm
i
ttance
wave
form
, energy usage
pattern
of t
h
e loa
d
s, etc., ha
ve
b
e
en
u
s
ed
in
N
I
A
L
M [2
],
[3
].
Th
e algo
r
ith
m
s
u
s
ed
v
a
ry fro
m
a ‘si
n
g
l
e algo
rith
m
’
to
‘m
u
ltip
le
alg
o
rith
m
s
’, th
e latter u
s
ed
in ‘Co
mmittee
Decisio
n
Mech
an
ism
’
(CDM) in
whi
c
h m
a
jority of ‘votes’
(i.e
., a
g
reem
ent), ‘cast’ by each of
the algorithm
s
, decides the final
result
of i
d
ent
i
f
i
e
d a
ppl
i
a
nce
s
[
2
]
.
Som
e
pu
bl
i
s
he
d re
p
o
rt
s
[8]
,
[1
0]
hi
ghl
i
g
ht
t
h
e sh
o
r
t
c
om
ings
o
f
m
o
st
o
f
t
h
e
NI
ALM
t
ech
ni
que
s wi
t
h
res
p
ect
t
o
feat
u
r
e e
x
t
r
act
i
o
n.
Ap
pl
i
cat
i
ons o
f
neu
r
al
net
w
o
r
ks
us
i
ng s
p
ect
ral
fe
a
t
ures
h
a
v
e
also
b
een tried
to
so
lv
e
th
e prob
lem
s
o
f
id
en
tifica
tio
n [
7
].
Th
e au
thor
s [8
] str
e
ss that at present in
m
o
st
o
f
t
h
ese techniq
u
e
s
[
21-
25
] lo
ad
si
gnat
u
re
s are collecte
d
intrusively.
They
f
u
rt
her
em
phasi
ze t
h
a
t
no
n-
in
tru
s
i
v
e iden
ti
ficatio
n
of lo
ad
activ
ities and
n
on-in
tru
s
ive ex
traction
of lo
ad sign
atures are t
w
o relat
e
d
b
u
t
di
ffe
re
nt
pr
o
b
l
e
m
s
. An
ot
he
r s
e
ri
o
u
s om
i
ssi
on i
n
p
r
ese
n
t
-
da
y
NIA
L
M
t
ech
ni
q
u
es i
s
t
h
e a
b
se
nce o
f
di
sc
ussi
o
n
rega
rdi
ng t
h
e t
echni
que
use
d
t
o
swi
t
c
h O
F
F
a faul
t
y
and/
o
r
hi
g
h
p
o
w
er c
ons
um
i
ng de
vi
ce i
n
case of c
r
i
t
i
cal
p
o
wer su
pp
ly
co
nd
itio
ns su
ch
as p
e
ak
l
o
ads [2
6
]
. Th
is
is an
essen
tial req
u
i
rem
e
n
t
in
p
o
wer m
a
n
a
g
e
men
t
.
Pro
b
a
b
l
y
, t
h
e
abse
nce
of
di
s
c
ussi
o
n
i
s
on t
h
e ass
u
m
p
tion
that cons
um
ers are s
uppose
d
to be
prese
n
t a
t
their
p
r
em
ises wh
o
wou
l
d
switch
o
f
f th
e id
en
tified
app
lian
ces
wh
en
in
tim
ate
d
b
y
th
e u
tility
. Howev
e
r, th
ere is n
o
gua
ra
ntee that
the cons
um
ers shall be
prese
n
t whe
n
suc
h
operation
is warranted.
In suc
h
a
case of
em
ergency,
th
e u
tility will
h
a
v
e
no
o
t
h
e
r
o
p
tion
b
u
t
to
switch
OFF th
e su
pp
ly to
th
e
en
tire ho
use. Th
is can
raise seriou
s
ri
g
h
t
s
an
d p
r
i
v
acy
i
ssues of
con
s
um
ers, f
o
r
som
e
of t
h
ei
r
appl
i
a
nce
s
m
a
y
be su
pp
ose
d
t
o
be i
n
O
N
st
at
e
perm
anent
l
y
, a
nd/
o
r
s
o
m
e
appl
i
a
nce(s
)
m
a
y nee
d
t
o
b
e
ke
pt
O
N
w
h
en
t
h
e occ
upa
nt
s
o
f
t
h
e h
o
u
se
ha
ve
t
o
go
out. T
h
is probl
e
m
arises because of single point sensi
ng
of curre
nt signal as the relay, on comm
and from
the
u
tility, will
cu
t-off th
e
p
o
wer su
pp
ly at th
e sin
g
l
e po
in
t of
en
try. In
m
u
lti
-po
i
n
t
sen
s
ing
th
ere will b
e
at least
no
bl
ack
out
o
f
t
h
e ent
i
r
e ho
use;
o
n
l
y
part
i
a
l
appl
i
a
nces i
n
a part
i
c
ul
ar
gr
o
up w
o
ul
d b
e
affect
ed.
Ag
ai
n,
a
serious di
fficulty arises with rega
rd to t
h
e choice
of
feat
ures a
n
d al
go
r
i
t
h
m
s
i
n
t
h
e cont
e
x
t
of ac
hi
evi
n
g a
reasona
b
ly low cost of t
h
e
meter,
si
nce
m
o
re feat
ures
req
u
i
r
e m
o
re s
e
ns
ors,
an
d c
o
m
p
l
e
x nat
u
re
of l
o
ad
sig
n
a
t
u
res
requ
ire so
ph
isticated
d
a
ta acqu
i
sitio
n
ha
rdware
, and com
p
licated
al
gori
t
h
m
s
add t
o
t
i
m
e
co
m
p
lex
ity, cu
l
m
in
atin
g
ov
erall, in
a
v
e
ry co
m
p
lex
h
a
rd
w
a
re
a
nd so
ftwa
re, defeatin
g
t
h
e ob
j
ectiv
e.
Desp
ite
th
e on
go
ing
research
i
n
id
en
tificati
on t
ech
ni
que
s si
nce
ove
r t
w
o
deca
des,
t
h
e u
n
res
o
l
v
e
d
p
r
o
b
l
e
m
t
i
l
l
dat
e
i
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Id
en
tifyin
g
App
lia
n
c
es u
s
ing
NIALM
with
Min
i
mum Fea
t
u
r
es
(S
un
il
S
e
mw
a
l
)
91
3
to
find
an
an
swer to
t
h
e qu
estio
n
s
“wh
a
t are th
e ‘o
p
tim
al’ features of a
ppliance
s
to be consi
d
ere
d
e
ssential
an
d
sufficien
t
for id
en
tificatio
n
”
?
;
an
d
“what is th
e ‘op
timal’ alg
o
r
ith
m
to
b
e
u
s
ed
fo
r id
en
tificatio
n
?
”. An
essent
i
a
l
req
u
i
r
em
ent
of opt
i
m
al
it
y
i
s
t
h
e use of m
i
nim
u
m
sensors t
o
r
e
duce t
h
e ha
rd
ware c
o
m
p
l
e
xi
t
y
and
cost
, a
n
d m
i
ni
m
u
m
tim
e and space
c
o
m
p
l
e
xi
t
y
fo
r t
h
e al
g
o
ri
t
h
m
t
o
be si
m
p
l
e
. If
one
c
h
o
o
ses
t
o
desi
gn
an
d
d
e
v
e
l
o
p a sm
a
r
t m
e
ter fo
llowin
g
[2
] an
d some o
t
h
e
r
pu
b
l
i
s
h
e
d
repo
rts
[3]-[9
] for practical u
s
e in b
illio
n
s
of
resid
e
n
tial h
ouses in
coun
tries lik
e Ind
i
a, the co
st w
o
u
l
d
be so
pro
h
i
b
itiv
e th
at th
e con
s
u
m
ers wou
l
d
l
i
k
e
ly
reject t
h
em
outright. T
h
e t
w
o unre
sol
v
ed issues,
stated earlier, a
r
e
precisely the re
asons
why
NIALM
p
r
esen
tly is th
e m
o
st attractiv
e issu
e
fo
r
research
ers i
n
sm
art
m
e
ter d
e
sign
.
3.
ALGO
RITH
M
Fo
llowing
assum
p
t
i
o
n
s
are m
a
d
e
for th
e im
p
l
e
m
en
tatio
n
of
th
e algo
rith
m
:
(i
) O
n
l
y
si
ngl
e
phase a
p
pl
i
a
n
ces are use
d
b
o
t
h
i
n
l
i
g
ht
an
d p
o
w
er ci
rc
ui
t
s
. Pl
ug
-i
n t
y
p
e
s of l
o
ads a
r
e not
considere
d
.
(i
i
)
T
h
e al
g
o
r
i
t
h
m
i
s
based
st
r
i
ct
l
y
on t
h
e
spe
c
i
f
i
cat
i
ons
of
I
E
W
of
C
P
WD
,
G
ovt
o
f
I
n
di
a.
(iii) Stead
y state v
a
lu
es of featu
r
es P,
Q, h, an
d
m
h
fo
r all p
o
s
sib
l
e app
lian
ce co
m
b
in
atio
n
s
(tem
p
l
ates
)
∑
l
C
r
and
∑
p
C
r
, w
h
e
r
e l
≤
10, r = 1,2,
..,l f
o
r light load appliance
s
; and p
≤
2, r=1,2 for power load applianc
es, are
sto
r
ed
in
m
e
mo
ry lo
cation
s
.
Fo
r a m
a
x
i
m
u
m o
f
1
0
app
lian
ces in
a lig
h
t
circu
it, th
e to
tal co
m
b
in
atio
n
s
wou
l
d
be 1
0
2
3
.
H
o
w
e
ver
,
i
n
t
h
e ex
peri
m
e
nt
s i
n
t
h
e l
a
b envi
r
o
nm
ent
,
a m
a
xim
u
m
of 7 ap
pl
i
a
n
ces has bee
n
u
s
ed i
n
o
n
e
circu
it and 6
app
lian
ces i
n
th
e rem
a
in
in
g
three ligh
t
ci
rcuits.
Sim
ilarly
,
fo
r t
h
ree
p
o
w
er ci
rcuits,
2
po
we
r
applianc
es, which are the m
a
xim
u
m
a
llowable, are c
o
nnec
ted
in
two
p
o
wer circu
its and
o
n
e
i
n
th
e th
ird
.
Th
e
four
features (i
n (iii) above)
of all
possi
ble com
b
inations
of appliance
s
in
each of the light circuits, are store
d
in
th
ree m
a
tric
es, M
1
, M
2
, and M
3
; whe
r
e the colum
n
s of
M
1
(s,5) contain norm
alized va
lues of features P, Q,
I
,
p
f
, V
A
alth
ou
gh
w
e
u
s
e
only
f
eatu
r
es
P
an
d Q in
t
h
e al
g
o
rith
m
;
th
e co
lu
m
n
s of M
2
(
s
, 8
)
c
o
nt
ai
n t
h
e
fi
rst
ei
ght
o
d
d
ha
r
m
oni
cs com
ponent
s
of
feat
u
r
e h, a
nd t
h
e c
o
l
u
m
n
s of M
3
(
s
,8
) co
nt
ai
n t
h
e
m
a
gni
t
ude
s
of
fi
rst
ei
ght
o
d
d
ha
r
m
oni
cs com
ponent
s
of
feat
u
r
es
m
h
;
whe
r
e s
i
s
t
h
e t
o
t
a
l
nu
m
b
er of al
l
po
ssi
bl
e com
b
i
n
at
i
ons o
f
appliances in each cir
c
uit. For the unknow
n combination, which is to be identif
i
ed, th
e corr
esponding extracted featur
es
are s
t
ored in matri
ces
M
4
(1,5), M
5
(1,8), and M
6
(1,8).
The
es
sence
of algorithm
is e
xpl
ai
n
e
d as
f
o
l
l
o
ws:
st
ori
n
g
all the features
of individual
applianc
es, as
well as of
all
po
ssib
l
e co
m
b
in
atio
n
s
o
f
th
e ap
p
lian
c
es in matrices
as stated
above, and
u
s
ing
a sm
a
ll th
r
e
sho
l
d
valu
e
(
±
0
.
0
1
)
fo
r
no
r
m
aliz
ed
m
easu
r
ed
valu
es of
P
and Q
t
o
account for the possi
ble va
riations in s
u
ppl
y
voltage,
the algorithm
co
mpares on-
line t
h
e signature of the
u
nkn
own
co
m
p
o
s
ite lo
ad
(UCLF)
o
f
app
lian
ces
(Ap
p
l
) in
switch
e
d
ON
po
sitio
n, first wi
th
th
e P
v
a
l
u
es in
th
e
sto
r
ed
tem
p
lat
e
s. Sin
ce th
e small
th
resho
l
d o
r
to
leran
ce
valu
e, cho
s
en
after so
m
e
trial
s
, is n
o
t
o
p
timal, it is
q
u
ite lik
ely th
at two
o
r
m
o
re te
m
p
lates
m
a
y
b
e
i
n
d
i
cated in
th
is co
m
p
ariso
n
.
In
t
h
e
n
e
xt step
(secon
d
stag
e)
an
o
t
h
e
r co
m
p
arison
is m
a
d
e
with
th
e
Q v
a
l
u
es wit
h
th
e t
e
m
p
lates o
b
t
ain
e
d
after co
mp
ariso
n
. Th
is,
in
all
lik
elih
o
o
d
,
will furth
e
r
redu
ce
th
e nu
m
b
er of t
e
m
p
lates. In
the th
ird
stag
e of co
m
p
aris
on
with
features h
of
th
e
te
m
p
lates re
main
in
g
as residu
e fro
m
th
e se
co
nd
stag
e, there will p
o
ssi
bly b
e
fu
rth
e
r red
u
c
tion
,
leav
in
g
th
e
residu
e, hop
efu
lly, as th
e s
i
n
g
l
e d
e
sired
can
d
i
d
a
te. If
n
o
t
, th
e fou
r
t
h
stag
e of filterin
g
throug
h
m
h
accom
p
lishes this task
of
fina
l identification.
Note that
t
h
e use of m
h
is based
on the pri
n
ciple tha
t
if a
wave
form
is resolve
d
int
o
a
series of s
p
ec
tral co
m
pone
nts, superposition
of power a
pplies
because
of t
h
e
o
r
t
h
ogo
n
a
lity of sp
ectral co
mp
on
en
ts
o
f
d
i
fferen
t
frequ
en
ci
es.
Tabl
e 2. A
ppl
i
a
nces
i
n
st
al
l
e
d i
n
t
h
e
ci
rcui
t
s
Sr.
Light load appliances connec
t
ed in
lig
ht cir
c
uits
(LC
1
-LC4
)
LC1 LC2
LC3
LC4
1
T
L
(
40W)
CFL
(
8W)
CFL
(
8W)
T
ube light (
40W
)
2
I
L
(
60W
)
CFL
(
15W
)
I
n
cand.
L
a
m
p
(
2
0
0
W
)
Bulb (
60W
)
3
CFL
T
ube light (
40W
)
M
obile char
ger
(
N
okia)
CFL
(
15W)
4
L
a
ptop
DSO
L
a
ptop
L
a
ptop
5 M
onitor
CPU
m
onitor
T
V
6
Fan
Function Gener
a
tor
Fan
Fr
idge
7
Washing m
achine
Washing m
achine
Roo
m
heat
er
Fan
Sr.
Pow
er load appliances connecte
d
i
n
pow
er circuits (
P
C1-P
C3)
PC 1
PC 2
PC 3
1
E
l
ectr
i
c Press
Air
conditioner
M
i
cr
o-
oven
2
I
nd Cooker
I
nduction Coo
k
er
Roo
m
Heater
If t
h
ere still remain
s a situ
atio
n wh
ere th
e
po
ssib
l
e
cand
i
date
s in
i
d
en
tifi
catio
n
at t
h
e
fou
r
t
h
stag
e
o
f
fi
lterin
g
are m
o
re t
h
a
n
one
, s
o
m
e
ot
he
r s
u
g
g
est
e
d
fea
t
ures
suc
h
a
s
I
,
p
f
, a
n
d C
W
w
oul
d
be
use
d
t
o
fi
nal
l
y
i
d
e
n
t
i
f
y
t
h
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJECE Vol. 4, No. 6, D
ecem
ber 2014
:
909 – 922
91
4
appl
i
a
nc
es i
n
t
h
e
un
k
n
o
w
n c
o
m
b
i
n
at
i
on.
F
o
rt
unat
e
l
y
,
ho
weve
r,
i
n
e
x
t
e
nsi
v
e
ex
peri
m
e
nt
at
i
ons
wi
t
h
com
m
on
dom
est
i
c
appl
i
a
nces a
n
d a
fe
w l
a
b
eq
ui
pm
ent
s
(Tab
le
1) we fo
und
th
at
un
iqu
e
id
en
tification
was always
co
nfin
ed
with
in
th
e
first fo
ur st
ag
es
o
f
filtering
. The
salien
t
features o
f
th
e al
g
o
rithm
are sh
own
m
o
re ex
p
lic
itly
b
y
a
flow ch
art in
Fi
g
u
re
1
.
Th
e app
lian
ce
co
d
e
s
are en
tered
in
th
e app
lian
c
e co
m
b
in
atio
n
s
datab
a
se c
o
d
e
matrix
, CM; e.g
.
, fo
r a
6-a
p
pliances case
the
codes
ent
e
re
d i
n
C
M
=[1
1;
2
2;
3
3;
4
4;
5
5;
6
6;
..;
9
1
4
;
1
0
15;
.
.
;
3
9
3
4
6
;
40
3
56;
41
4
5
6
;
..;
6
3
12
3
4
5
6
]
. At
t
h
e
ti
m
e
o
f
d
a
ta acq
u
i
sition
th
e ap
p
lian
c
es are co
d
e
d
i
n
si
n
g
l
e
d
i
g
it nu
m
b
ers,
1
,
2
,
..,7
.
4.
E
X
PERI
MEN
T
AL PR
O
C
EDU
R
E
Dom
e
st
i
c
appl
i
a
nces com
m
onl
y
use
d
i
n
I
n
di
an m
i
ddl
e cl
ass fam
i
li
es have bee
n
s
h
o
w
n i
n
Ta
bl
e
1.Ta
bl
e2
sh
o
w
s t
h
e a
ppl
i
a
nce
s
co
nnect
e
d
i
n
t
h
e f
o
u
r
l
i
g
ht
c
i
rcui
t
s
LC
1 t
o
LC
4 a
nd
p
o
w
e
r ci
rc
ui
t
PC
1 t
o
PC
3.
Du
e to un
av
ailab
ility o
f
a few do
m
e
stic ap
p
lian
ces
fo
r
so
m
e
circu
its, th
ese
h
a
v
e
b
een su
b
s
titu
ted
b
y
a few lab
eq
u
i
p
m
en
ts, su
ch
as
d
i
g
ital sto
r
ag
e
o
s
cillo
sco
p
e
(D
SO), fun
c
tion
g
e
nerato
r
(FG), an
d
m
o
n
ito
r, wh
ich
cont
ri
b
u
t
e
a ri
cher va
ri
et
y
of at
t
r
i
but
es o
f
feat
ures i
n
h and
m
h
. An expe
ri
m
e
nt
al
l
a
borat
ory
set
u
p was
creat
e
d
to represe
n
t an electrical installation strictly as
per C
P
WD s
p
eci
fi
cat
i
ons
on a
fa
bricated wooden fram
e
.
There a
r
e s
o
me comm
on appliances on
th
e
lig
h
t
lo
ad
ci
rcu
its, wh
ich
are g
e
n
e
rally th
e case in
ho
uses, bu
t
powe
r load a
p
pliances in power circ
uits are distinct.
W
ith six de
vices
connected i
n
each of the thre
e light
ci
rcui
t
s
, an
d s
e
ven i
n
t
h
e f
o
urt
h
l
i
ght
ci
rc
ui
t
,
t
h
e t
o
t
a
l
n
u
m
b
er of
pos
s
i
bl
e com
b
i
n
at
ions
w
o
r
k
s o
u
t
t
o
31
6,
each represe
n
t
i
ng differe
n
t com
b
inations of devices
s
w
itched
ON. For six power
appliances, these
were
distributed e
q
ually in three circu
its, the tot
a
l num
ber of
diffe
re
nt co
m
b
inations
being only three
for each
circu
it.
Fig
u
r
e
1
.
Flow ch
ar
t of
p
r
op
osed
al
g
o
r
ith
m
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Id
en
tifyin
g
App
lia
n
c
es u
s
ing
NIALM
with
Min
i
mum Fea
t
u
r
es
(S
un
il
S
e
mw
a
l
)
91
5
The
dat
a
acq
ui
si
t
i
on was i
n
f
o
u
r
st
ep
s:
first
step
con
s
ists of ex
tractio
n of basic
features (P, Q, I,
pf,
VA) of indivi
dual a
ppliance
s
in each circ
uit, second
step was e
x
tracti
on
of l
o
ad si
gnature of thes
e basic
feature
s
of all possi
ble combinations
of the appliances
in each circu
it, third step was extraction of
h
com
pone
nt
s (
f
u
n
d
am
ent
a
l
,
3
rd
., 5
th
, 7
th
, 9
th
, 11
th
, 13
th
, a
n
d
15
th
) of ind
i
v
i
du
al and
all po
ssib
le co
m
b
in
atio
n
s
o
f
applianc
es in
each circ
uit, a
n
d the
fourt
h
step wa
s
for e
x
traction
of si
gnat
u
re m
h
.
Note that for s
p
ectral
com
pone
nt
s, si
gni
fi
cance
o
f
c
ont
ri
b
u
t
i
o
n
o
f
harm
oni
cs
fo
r
t
h
e fi
r
s
t
ei
g
h
t
od
d
ha
rm
oni
cs was
det
e
rm
i
n
ed
by
th
e ratio
of ((m
h
)
ma
x
/ 1
0
)
. If an
y
m
h
v
a
lu
e is less th
an
th
is ratio
, it is ig
n
o
red
(assu
m
e
d
zero
in
v
a
l
u
e), and
corres
ponding
to these ignore
d m
h
va
l
u
es, t
h
e rel
e
vant
har
m
oni
c co
m
pon
ent
s
we
re al
so
i
g
n
o
re
d.
O
n
l
y
t
h
ese
m
o
d
i
fied
v
a
l
u
es were th
en
selected
fo
r th
e tem
p
la
tes to
b
e
sto
r
ed
in m
e
m
o
ry. Th
us,
we
hav
e
n
o
t
no
rm
al
ized
t
h
e m
odi
fi
ed
m
h
val
u
es si
nc
e t
h
ei
r m
a
gni
t
ude
s
were
m
a
de com
p
arabl
e
,
unl
i
k
e
t
h
e
P,
Q,
I,
p
f
,
an
d
V
A
dat
a
.
For each com
b
ination a tota
l of si
x re
a
d
ings (e
ach
10 se
conds a
p
a
r
t) i
n
steady states
were ta
ke
n for each
feat
ure
,
out
of
w
h
i
c
h
onl
y
t
hos
e rea
d
i
n
gs
whi
c
h s
h
o
w
e
d
n
o
va
ri
at
i
o
n
were
use
d
fo
r
st
ori
ng t
e
m
p
l
a
t
e
s i
n
m
e
m
o
ry
. Load
si
gnat
u
res of i
ndi
vi
d
u
al
as wel
l
co
m
b
i
n
at
i
ons o
f
appl
i
a
nce
s
were ext
r
act
e
d
i
n
t
w
o m
odes:
off
-
line and
on-line (real
-tim
e) d
e
scri
be
d
bel
o
w
.
Fi
gu
re
2(a
)
.
Sc
hem
a
ti
c fo
r e
x
peri
m
e
nt
al
set
up
fo
r
dat
a
acq
u
i
si
t
i
on
4.1.
Off-Line
Mode
of Me
as
urement
In t
h
e
of
f-l
i
n
e
m
ode, t
h
e ex
peri
m
e
nt
al
set up co
nsi
s
t
e
d of
a recently
mark
eted
smart
sock
et
met
e
r
(
m
anufa
c
turers
M/s Ha
m
a
G
m
bH & Co KG,
Ger
m
any
)
alon
g with digital
st
orage oscillo
s
c
ope (DSO) with
current
sen
s
or
s.
Plug
meter is an electrical energy
meter
whic
h
p
l
u
g
s
in between an
y app
lian
c
e and
t
h
e
AC
ou
tlet. It
m
easures po
w
e
r
co
ns
um
pt
i
on, I
rm
s
, V
rm
s
with power
factor (c
os
Φ
).
The s
m
art
socket
me
ter was u
s
ed fo
r direct
me
asure
m
ent (steady
state
s
) of P, Q, I, pf, and VA. Only
P and Q feature
s
were used. For extraction of sp
ectral
co
m
pone
nts h a
nd
m
h
, DSO w
a
s used.
A si
n
g
le sm
art so
ck
et m
e
ter was
u
s
ed
in turn fo
r each
o
f
th
e light and
po
we
r circ
uits.
Relay
s
’
Ci
rcu
it
1
Ø
Mains
su
ppl
y
Li
gh
tin
g
ci
rcu
its
Power c
i
rc
uits
Mains
C
u
rrent
sensor
Neu
t
ral lin
k
Neu
t
ra
l li
nk
N
N
N
N
N
N
N
Curre
nt
Sensor
P
P
P
P
P
P
P
N
Swi
t
c
h
box
e
s
fo
r (LC
1
, LC2
,
LC3
,
LC4)
Swi
t
c
h
b
o
x
e
s
f
o
r (P
C1
, PC
2
,
P
C
3
)
Si
gn
al
co
nd
iti
on
er
ci
rcu
it
PT
Analog I/O
Digita
l I
/
O por
t
Arduino
Mega 2560
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJECE Vol. 4, No. 6, D
ecem
ber 2014
:
909 – 922
91
6
Fi
gu
re
2(
b
)
.
Sc
hem
a
ti
c fo
r e
x
peri
m
e
nt
al
se
tup for
s
p
ectral feature (h,
m
h
) extraction
4.2. On-Line
(Real
-
Time)
Mode of Meas
ur
ement
In t
h
e
on
-l
i
n
e
m
ode for P an
d Q m
easurem
ent
,
A
r
d
u
i
n
o
m
i
croco
n
t
r
ol
l
e
r [2
3]
was us
e
d
. A
r
d
u
i
no
Mega 2560 de
velopm
ent boa
rd
receive
s
current a
n
d voltage signat
u
re
from
respective sens
ors t
o
c
o
m
pute the
lo
ad
pr
of
ile (
P
, Q
,
I,
V
,
pf
)
of
co
nn
ected ap
p
l
ian
ces.
As sh
ow
n in
Figu
r
e
2(
a)
t
h
e
f
our
cu
rr
en
t sen
s
o
r
s in
lig
h
t
circuits are
pla
ced at the
exit
of eac
h ci
rcuit
wire
from
th
e D
B
, an
d, sim
i
l
a
r
l
y, th
r
e
e cur
r
en
t sensor
s
h
a
ve b
e
en
u
s
ed
at th
e ex
it po
in
t of each
p
o
wer circu
it fro
m
th
e DB
.
Ad
d
ition
a
lly, one cu
rren
t sen
s
or is used
righ
t
at th
e
i
n
com
i
ng m
a
i
n
s t
o
t
h
e DB
. T
h
i
s
sens
o
r
has
been
use
d
f
o
r
m
oni
t
o
ri
ng a
n
d rec
o
r
d
i
n
g
of
t
o
t
a
l
curre
nt
dra
w
n
fr
om
t
h
e sup
p
l
y
so t
h
at
t
h
i
s
m
a
y
be used i
n
f
u
t
u
re
w
o
r
k
fo
r o
n
-l
i
n
e e
n
e
r
gy
co
ns
um
pt
ion
dat
a
an
d i
t
s
on
war
d
tran
sm
issio
n
to
u
tility,
an
d
h
a
s n
o
t
b
e
en
u
s
ed
fo
r
id
en
tificatio
n
o
f
app
lian
ces.
Figure 2(a) prov
ides th
e
har
d
ware
pl
at
form
for al
l
dat
a
acqui
si
t
i
on
r
e
qui
rem
e
nt
fo
r the basic feat
ures
(P,
Q, I,
pf, VA)
while Figure
2(
b) s
h
o
w
s t
h
e
schem
a
ti
c of spect
ral
feature extractions; all ei
ght current s
e
ns
ors are c
o
nnected to the a
n
alog
i
n
p
u
t
p
o
rt
of
m
i
croco
n
t
r
ol
l
e
r [
23]
w
h
i
c
h
i
s
i
n
t
e
rface
d
wi
t
h
PC
by
u
s
i
ng M
A
TL
A
B
Ar
dui
no i
n
t
e
rface
packa
g
e. M
A
T
L
AB Support
Packa
g
e
fo
r
Ardu
ino
en
ab
les co
mm
u
n
i
catio
n
with
t
h
e m
i
c
r
o
c
on
tro
ller
over a
USB ca
ble, so that the
c
u
rre
n
t signals
from
the sens
o
r
s
can
be
di
rect
l
y
use
d
i
n
M
A
T
L
AB
e
n
vi
ro
n
m
ent
fo
r
si
gnal
pr
ocessi
ng
(F
FT,
W
a
ve
l
e
t
)
.
5.
R
E
SU
LTS
AN
D D
I
S
C
U
SSI
O
N
S
A t
o
t
a
l
of
31
6
‘u
nk
n
o
w
n
’ c
o
m
b
i
n
at
i
ons (3
x 6
3
+ 1 x
12
7)
, i
.
e., f
o
r al
l
possi
bl
e com
b
i
n
at
i
o
ns o
f
ap
p
lian
c
es in
t
h
e fou
r
ligh
t
ci
rcu
its,
were tested
in
o
f
f-lin
e
an
d on
-lin
e (real-ti
m
e
)
m
o
d
e
for valid
atio
n
o
f
t
h
e
al
go
ri
t
h
m
using c
o
m
put
er
pr
og
ram
devel
o
ped
o
n
t
h
e M
A
TL
AB
pl
at
fo
rm
. The ‘
u
n
k
n
o
w
n’
of
f-l
i
n
e
com
b
i
n
at
i
ons
were sel
ect
ed r
a
nd
om
l
y
from
am
ong t
h
e
kn
o
w
n c
o
m
b
i
n
at
i
ons w
h
o
s
e t
e
m
p
l
a
t
e
s were st
or
ed i
n
me
m
o
ry locations a
s
stated e
a
rlier. T
h
e res
u
lts of t
h
ese tests were
found correct
for all th
ese co
m
b
inatio
n
s
.
Next
,
val
i
d
at
i
o
n of t
h
e
pr
oce
d
u
r
e was al
so
carri
ed
out
i
n
real
t
i
m
e
i
n
respect
of P a
n
d
Q val
u
es by
r
a
nd
om
swi
t
c
hi
n
g
of
a
ppl
i
a
nce
s
i
n
O
N
st
at
es i
n
eac
h
of
t
h
e
f
o
ur li
g
h
t
ci
rcu
its. For th
e on-lin
e
valid
atio
n
i
n
resp
ect of
h
an
d
m
h
v
a
l
u
es, th
e exp
e
ri
men
t
s u
s
ing
Ardu
ino
m
i
cro
c
o
n
t
ro
ller are st
ill in
p
r
o
g
ress, and
so
on
ly off-lin
e
m
o
d
e
o
f
th
ese v
a
lu
es
w
e
r
e
u
s
ed
in
v
a
li
d
a
tio
n
of
th
e al
g
o
r
ith
m
.
Th
e r
e
su
lt of
th
e t
e
st w
a
s fo
und v
e
r
y
satisfacto
r
y sin
ce, ex
cep
t
for two
unk
nown
co
m
b
in
atio
ns in
circu
it LC2
,
all o
t
h
e
r test can
d
i
d
a
tes were
uni
quel
y
i
d
ent
i
fi
ed.
N
o
t
a si
ngl
e am
bi
g
u
i
t
y
ever
occ
u
rr
ed.
We
n
o
w
copy
a
fe
w
re
sul
t
s
o
f
t
h
e
pr
og
ram
ex
ecu
tion fo
r
so
m
e
r
a
n
d
o
m
l
y
sw
itch
e
d
o
n
-lin
e unk
now
n
co
m
b
inations.
In the e
x
am
ples, the
form
at of the
resu
lt
d
i
sp
layed
is:
first
n
u
m
b
e
r ind
i
cates th
e lo
cation
in m
e
m
o
ry o
f
t
h
e st
ored tem
p
late fo
r m
a
tch
i
n
g
with
t
h
e
u
nkn
own
co
mb
in
ation
,
th
e seco
nd
nu
m
b
er indicates the code
of the a
p
p
lian
ces, and
the th
ird
nu
m
b
er sh
ows
the stage
of fil
t
ering at wh
ic
h the
ide
n
tification is c
o
m
p
leted. Fi
gure
gi
ven for each of the
dis
p
layed res
u
lts
shows t
h
e
num
b
er of
possibl
e candidates as residue at
each stage
of
filtering
until t
h
e fi
nal ide
n
tification
st
age, a
n
d al
s
o
sho
w
s
t
h
e a
d
dr
esses
of l
o
cat
i
o
ns
of
st
o
r
ed
ca
ndi
dat
e
s’ t
e
m
p
l
a
t
e
s, w
h
i
c
h e
m
erge as
t
h
e
re
si
due
,
m
a
t
c
hi
ng wi
t
h
t
h
e un
kn
o
w
n
com
b
i
n
at
i
on un
de
r t
e
st
. Th
e deco
di
n
g
o
f
t
h
e l
o
cat
i
on code
of t
h
e ca
n
d
i
d
at
e,
w
h
ich
is f
ound
to
m
a
tch
ex
actly w
ith
th
e u
n
k
now
n
com
b
in
atio
n
,
is d
o
n
e
thr
oug
h
p
r
og
r
a
m
as ex
p
l
ain
e
d
b
e
low.
It m
a
y
b
e
n
o
t
ed
th
at t
h
e id
en
tification
is
n
e
v
e
r foun
d to
o
ccur at
th
e 1
st
stag
e, i.e. P.
Th
is is
ob
v
i
o
u
s
because the
tolerance le
vel
of
0.01 selected after a
fe
w tri
a
ls includes
va
lues
(P±0.01P) and
(Q±0.01Q) of
active and
reac
tive powers and so i
n
cl
u
d
es m
o
re
t
h
an
o
n
e candi
dat
e
i
n
P al
way
s
.
W
e
gi
ve
o
n
l
y
one
e
x
am
pl
e
from
each of the light circ
uits, and
one
exam
pl
e from
powe
r circ
uit, for want of s
p
ace.
Circu
it LC1
:
Ex
41
456
4
Fo
r th
is ex
am
ple, Fig
u
r
e
3
d
i
sp
lays th
e resu
lts o
f
id
en
tificatio
n
in
resp
ect of th
e n
u
m
b
e
r of p
o
ssib
l
e
candi
dat
e
s
(res
i
due
) a
v
ai
l
a
bl
e
i
n
P,
Q,
h
,
a
n
d m
h
an
d t
h
ei
r
res
p
ect
i
v
e l
o
c
a
t
i
on a
d
dress
i
n
cl
u
d
i
n
g t
h
e l
o
cat
i
o
n
ad
dress
o
f
t
h
e
can
d
i
d
a
te wh
ich
id
en
tifies th
e u
nkn
own
co
m
b
in
ation
of app
lian
ces Th
e
deco
d
i
n
g
of lo
catio
n
PC
Arduino Mega 2560
Outputs
f
r
o
m
c
urr
e
nt
sensor
s
A7
A6
A5
A4
A3
A2
A1
A0
S
p
ectra
l Feature
extraction
MAT
L
AB
Arduino
Interface (MAI)
D0 D1 D
2
D3
D4
D5 D6
D7
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
ECE
I
S
SN
:
208
8-8
7
0
8
Id
en
tifyin
g
App
lia
n
c
es u
s
ing
NIALM
with
Min
i
mum Fea
t
u
r
es
(S
un
il
S
e
mw
a
l
)
91
7
ad
dress
(41
)
,
sh
own
as 456
, is th
e id
en
tity o
f
in
d
i
v
i
dual ap
p
lian
ces i
n
co
m
b
in
atio
n.
W
e
co
p
y
from th
e
MATLAB
th
e
resu
lts
o
f
th
e ex
ecu
ted
p
r
og
ra
m
and explain
the steps
of e
x
ecution.
>> loc =
13
22
24
29
30
32
34
35
37
39
41
43
48
50
>> locq =
8
11
>> filtq
=
0
.
12
75
0
.
18
24
0
.
13
18
0
.
119
0
0.135
8
0.
1196
0.
1867
0.
1318
0.
1161
0.
1322
>> loch =
1
2
>> filth
=
1
0
5
0
9
0
13
0
1
0
5
0
9
0
13
0
>> locm
h =
41
>> f
ilt
m
h
=
4
.
9
700
0
2
.
5
500
0
1.860
0
0
1
.
11
00
0
We ca
n see the effect
of t
o
l
e
rance
val
u
e resulting
i
n
sel
ect
i
on
of as
m
a
ny
as f
o
urt
e
e
n
p
o
ssi
bl
e P
can
d
i
d
a
tes who
s
e add
r
esses
o
f
storag
e are
g
i
v
e
n
in
‘lo
c
’.
Wh
en
filtered
th
rou
g
h
th
e n
e
xt stag
e Q, th
e nu
m
b
er
of
po
ssi
bl
e ca
n
d
i
d
at
es
drast
i
c
al
l
y
reduces t
o
t
w
o,
w
hos
e
stora
g
e a
d
dresse
s are indi
cat
ed
i
ndi
rect
l
y
by
l
i
nke
d
num
bers i
n
’lo
c
q’
, i.e.
, t
h
e
n
u
m
b
er 8
in
lo
cq
refe
rs t
o
locatio
n
nu
m
b
er 35
i
n
lo
c, and
1
1
in lo
cq
refers to
lo
catio
n nu
m
b
er
4
1
in
l
o
c
(wh
i
ch
is actu
a
lly th
e
d
e
sired
add
r
ess
o
f
un
know
n co
m
b
in
ation
)
. ‘filtq
’ shows two
pos
si
bl
e can
di
dat
e
s i
n
Q, t
h
e col
u
m
n
s 1-
5
repre
s
ent
t
h
e
no
rm
al
i
zed values o
f
fi
ve
fea
t
ures (
P
.
Q
,.
.,
V
A
)
of
te
m
p
lates sto
r
ed
at lo
c 3
5
and 4
1
.
Th
e th
ird
stag
e o
f
filteri
n
g
throug
h
h
still
yie
l
d
s
two
can
d
i
d
a
tes shown
in
‘filth
’
g
i
ven
by lin
k
e
d
nu
m
b
ers
1
,
2
ind
i
cated
b
y
‘lo
c
h
’
wh
ich
refers t
o
th
e link
e
d
nu
mb
ers in
lo
cq
, i.e.,
1
i
n
loch re
fe
rs to s
e
rial num
ber 8
in locq
whic
h
refer
s
to
lo
cat
io
n
n
u
m
b
e
r 35 in
lo
c, sim
i
larly 2
in
lo
ch
refers to
1
1
in
l
o
cq wh
i
c
h
refers to 41
in
lo
c.
Fi
gu
re 3.
I
d
ent
i
fi
cat
i
on of
ap
p
l
i
a
nces
(LC
1
,
LC
2, LC
3,
LC
4)
Since, e
v
e
n
at
the 3
rd
sta
g
e (h) the ca
ndidate
s
for identif
ica
tion are
m
o
re than one, t
h
e
4
th
stag
e (m
h
)
o
f
filter
actio
n
is g
o
n
e
th
ro
ugh
wh
ich n
o
w u
n
i
q
u
e
ly
id
en
tifies d
i
rectly
th
e lo
catio
n
nu
m
b
er 4
1
in
‘lo
c
m
h
’ where th
e
sto
r
ed
te
m
p
late o
f
co
m
b
in
ation
m
a
tch
e
s th
e u
nkn
own
app
lian
ces co
m
b
in
atio
n
,
wh
ile th
e v
a
lu
es
o
f
m
h
fo
r th
is
0
10
20
30
40
50
60
70
0
0.
1
0.
2
0.
3
0.
4
0.
5
0.
6
0.
7
0.
8
0.
9
1
Loc
at
i
on addres
s
of
c
a
ndi
da
t
e
s
P
o
s
s
i
bl
e
c
andi
dat
es
f
o
r s
e
l
e
c
t
i
on at
di
f
f
erent
s
t
ages
(P
,
Q,
h,
m
h
)
P
Q
h
mh
U
n
k
now
n appl
i
anc
e
c
o
m
b
i
nat
i
o
n
obt
ai
ned at
l
o
c
a
t
i
on 1
2
at
4t
h
s
t
age
0
10
20
30
40
50
60
70
0
0.
1
0.
2
0.
3
0.
4
0.
5
0.
6
0.
7
0.
8
0.
9
1
L
o
c
a
t
i
on a
ddr
es
s
of
c
a
ndi
dat
es
P
o
s
s
i
bl
e
c
a
n
d
i
dat
es
f
o
r
s
e
l
e
c
t
i
o
n at
d
i
f
f
e
r
ent
s
t
ages
(
P
,
Q
,
h
,
m
h)
P
Q
h
mh
Un
k
n
o
w
n
co
m
b
-
nat
i
o
n
of
appl
i
a
-
c
e
s
obt
ai
ned
at
l
o
c
a
t
i
o
n
41
at
4t
h s
t
age (
m
h)
0
10
20
30
40
50
60
70
0
0.
1
0.
2
0.
3
0.
4
0.
5
0.
6
0.
7
0.
8
0.
9
1
Lo
c
a
t
i
on
ad
dre
s
s
o
f
c
a
n
d
i
d
at
e
s
P
o
s
s
i
b
l
e
c
a
nd
i
d
at
e
s
f
o
r
s
e
l
e
c
t
i
o
n a
t
di
f
f
er
en
t
s
t
a
g
e
s
(
P
,
Q
,
h
,
m
h)
P
Q
h
mh
U
n
k
n
own ap
pl
i
a
nc
e
c
o
m
b
i
n
at
i
on o
b
t
a
i
n
ed a
t
l
o
c
a
t
i
o
n
6
2
at
4
t
h s
t
age
(m
h
)
0
20
40
60
80
100
120
1
4
0
0.
1
0.
2
0.
3
0.
4
0.
5
0.
6
0.
7
0.
8
0.
9
1
Loc
at
i
on
addr
es
s
of
c
andi
dat
es
P
o
s
s
i
bl
e
c
andi
dat
es
f
o
r
s
e
l
e
c
t
i
on at
di
f
f
e
r
ent
s
t
ages
(
P
,
Q
,
h,
m
h
)
P
(
21)
Q
(
10)
h(3
mh
(
1
U
n
k
now
n appl
-
anc
e c
o
m
b
i
na-
t
i
on obt
ai
ned at
l
o
c
a
t
i
on 74 at
4t
h
s
t
age
a
b
cd
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
SN
:
2
088
-87
08
IJECE Vol. 4, No. 6, D
ecem
ber 2014
:
909 – 922
91
8
u
nkn
own
co
m
b
in
ation
are sho
w
n
in
‘filtm
h
’
.
W
e
can
also
see th
ese lo
catio
n
add
r
esses of po
ssib
l
e can
d
id
ates
P,
Q,
h,
an
d t
h
e m
h
i
n
Fi
g
u
re
3.
Circu
it LC2
Ex
62
23456
3
Circu
it LC3
Ex
12
23
4
Circu
it LC4
As s
h
ow
n i
n
T
a
bl
e 2
,
t
h
i
s
ci
r
c
ui
t
has
seve
n
appl
i
a
nc
es
wh
ose t
o
t
a
l
n
u
m
b
er o
f
p
o
ssi
bl
e
com
b
i
n
at
i
ons
wo
rk
s
out
t
o
1
2
7
.
N
o
t
e
t
h
at
t
h
i
s
ci
rcui
t
cont
ai
ns T
V
an
d fri
d
g
e a
m
ong t
h
e appl
i
a
nces.
W
e
s
h
o
w
t
h
ree t
y
pi
cal
cases
o
f
i
d
en
tification
with
th
e
same in
terp
retations as g
i
v
e
n
in t
h
e earlier e
x
amples.
As a
n
e
x
a
m
pl
e, we co
py
fr
om
MATLAB
the execute
d result
.
Ex
74
1345
4
C
u
rrent
w
a
ve (
c
w)
f
e
at
ures:
C
u
r
r
ent
wa
vef
o
rm
of a
p
pl
i
a
n
ces co
n
n
ect
ed i
n
ci
rc
ui
t
s
LC
1,
LC
2,
LC
3
,
LC
4 a
r
e s
h
o
w
n i
n
Fi
gu
re
4.
Fi
gu
re
4.
C
u
rre
nt
wa
ve
f
o
rm
of LC
1, LC
2,
L
C
3, LC
4
5.
1.
Summ
a
ry
of Light Circu
its
Iden
tifica
tion
Tab
l
e 3
shows id
en
tificatio
n
freq
u
e
n
c
y at d
i
fferen
t
stag
es
o
f
filtering
in
th
e fo
ur lig
h
t
circu
its, LC1,
LC2, LC3, and LC4. N
o
te tha
t
, here, the
fre
quency of id
e
n
tification at m
h
stage is
pr
og
re
ssi
vel
y
i
n
creasi
ng as
we proceed
from
LC1 to LC4. T
h
is is be
cause s
o
m
e
of the appliance
s
use
d
in LC
2, LC3, a
n
d L
C
4 are
pr
o
g
ressi
vel
y
r
i
cher i
n
harm
oni
cs, a
n
d i
n
t
h
ei
r spect
ral
m
a
gni
t
u
des
,
m
h
.
Thi
s
beha
vi
or
i
s
su
pp
o
r
t
e
d
b
y
t
h
e
C
W
sign
atures o
f
app
lian
ces in
Fi
g
u
re
4
for
a case wh
en
all th
e in
stalled
ap
p
lian
c
es are si
m
u
ltan
e
o
u
s
ly
in
ON
states in
th
e fou
r
circu
its. As
sh
own
,
C
W
for LC1
is
n
early
sin
u
s
o
i
d
a
l, while fo
r th
e remain
in
g
three circu
its,
t
h
ei
r C
W
si
gna
t
u
res i
n
c
r
easi
n
gl
y
depa
rt
fro
m
si
nusoi
dal
n
a
t
u
re. T
h
e P, Q
,
I, pf
, V
A
an
d h, m
h
val
u
es o
f
t
h
es
e
cir
c
u
its ar
e show
n in
Figu
r
e
5.
a
b
c
d
0
10
0
20
0
30
0
400
50
0
60
0
70
0
800
-0.2
-0.15
-0.1
-0.05
0
0.
05
0.
1
0.
15
0.
2
Sa
m
p
l
e
-----
-----------
-----------
->
A
m
p
l
i
t
u
d
e
-
---
--
--
--
--
--
--
-->
0
10
0
20
0
30
0
40
0
50
0
60
0
70
0
80
0
-0
.0
5
-0
.0
4
-0
.0
3
-0
.0
2
-0
.0
1
0
0.
01
0.
02
0.
03
0.
04
0.
05
Sa
m
p
l
e
-
---
--
--
---
--
---
--
--
---
--
---
>
A
m
p
l
i
t
u
d
e
-
-
---
--
--
--
--
--
---
->
0
10
0
20
0
30
0
400
50
0
60
0
70
0
800
-0.2
-0.15
-0.1
-0.05
0
0.
05
0.
1
0.
15
Sa
m
p
l
e
---------
-----------
----->
A
m
p
l
i
t
u
d
e
------
-----
------
->
0
100
200
300
400
500
600
700
80
0
-0
.
2
-0.15
-0
.
1
-0.05
0
0.
05
0.
1
0.
15
Sa
m
p
l
e
-------
-------------
->
A
m
p
l
i
t
u
d
e
-
------
-----
--->
Evaluation Warning : The document was created with Spire.PDF for Python.