TELKOM
NIKA Indonesia
n
Journal of
Electrical En
gineering
Vol. 12, No. 8, August 201
4, pp. 5985 ~ 5998
DOI: 10.115
9
1
/telkomni
ka.
v
12i8.631
7
5985
Re
cei
v
ed Ma
y 16, 201
4; Revi
sed
Jun
e
27, 2014; Accepted July 8,
2014
An Improved Reconstruction Algorithm Based on
Compressed Sensing for Power Quality Analysis in
Wireless Sensor Networks of Smart Grid
Yi Zhong*
1
, Jiahou Hua
n
g
2
Schoo
l of Information En
gi
ne
erin
g, W
uhan
Univers
i
t
y
of
T
e
chn
o
lo
g
y
, W
uhan, Hu
be
i, P.
R. Chin
a
*Corres
p
o
ndi
n
g
author, e-ma
i
l
: zhong
yi
@
w
h
u
t.edu.cn
1
, hjh
199
01
14@
w
h
u
t.edu.cn
2
A
b
st
r
a
ct
In recent ye
ars
,
the grow
ing
pow
er qu
ality
prob
le
ms i
n
smart gr
id ca
us
e w
i
despr
ead
concer
n at
ho
me
a
nd
abr
oad. B
e
ca
use
the tra
d
iti
ona
l
p
o
w
e
r qu
a
lity
al
gorith
m
s
w
h
ich
are
b
a
sed
o
n
Nyquist
sa
mp
li
n
g
theory h
a
ve t
h
e draw
b
a
cks o
f
complic
ated,
heavy c
o
mp
utations
an
d p
o
o
r rea
l
-t
i
m
e pe
rforma
nce
w
h
e
n
sampli
ng
an
d
ana
ly
z
i
n
g
co
ntinu
ous
massiv
e
sig
n
a
l
s in
s
m
art
grid. T
h
is
pap
er d
i
scus
s
ed a
n
i
m
prov
ed
reconstructi
on
alg
o
rith
m bas
e
d
on co
mpr
e
s
s
ed sens
ing
d
ue to the spar
sity of pow
er
qua
lity sign
als
in
freque
ncy d
o
m
ain f
o
r p
o
w
e
r
qua
lity a
nalys
i
s
. By usin
g th
e Z
i
gB
ee w
i
re
l
e
ss g
a
tew
a
y for w
i
reless
se
nsor
netw
o
rks an
d
ener
gy
meteri
n
g
chi
p
, w
e
dev
elo
p
a si
ng
le
meter
no
de to
do re
lative
ex
peri
m
e
n
ts. In the
cond
ition of th
e real test-be
d
and sever
a
l c
o
mpar
ed ex
p
e
r
iments, pow
er
qual
ity info
r
m
ation i
n
the hi
g
h
ly
compressi
on r
a
tio has g
o
o
d
perfor
m
a
n
ce a
ccordi
ng to CS
R (Co
m
press
i
o
n
Sampl
i
ng R
a
tio), SNR (Sig
nal
to Noise R
a
tio)
, MSE (Mean Squar
ed Error)
and ERP (En
e
rgy Rec
o
very
Percentag
e) , and w
ill b
e
w
i
dely
used i
n
pow
er
qua
lity an
alysis
.
Ke
y
w
ords
:
co
mpr
e
sse
d sens
ing (CS), pow
e
r
qual
ity (PQ),
w
i
reless sens
o
r
netw
o
rks
Copy
right
©
2014 In
stitu
t
e o
f
Ad
van
ced
En
g
i
n
eerin
g and
Scien
ce. All
rig
h
t
s reser
ve
d
.
1. Introduc
tion
In re
cent yea
r
s, the
growi
ng po
we
r q
u
a
lity
proble
m
s in
singl
e ph
ase
po
wer grid ca
use
wide
sp
rea
d
concern
at ho
me an
d ab
ro
ad. The
po
wer q
uality problem
s mai
n
l
y
lie in several
asp
e
ct
s: Power lo
ad in
sin
g
le-p
ha
se p
o
w
er gri
d
is b
e
comi
ng m
o
re and
mo
re
complicated a
n
d
diversifie
d. Mode
rn ele
c
trical e
quipm
ents,
which
are a
dopte
d
for the sa
ke of imp
r
o
v
ing
prod
uctio
n
eff
i
cien
cy, savin
g
en
ergy
and
de
cre
a
si
ng
p
o
llution
of the
enviro
n
me
nt, are
be
co
min
g
the main
re
source of
po
wer quality p
r
oble
m
s. T
h
e
singl
e-p
h
a
s
e po
wer l
o
a
d
whi
c
h
ha
s the
feature of n
o
n
linea
r, rich
harm
oni
c, impactive an
d
unbal
an
ced
will influen
ce
powe
r
g
r
id
and
cau
s
e
s
ne
w probl
em
s
of power
quality
.
Powe
r cu
st
omers have
i
n
crea
singly d
e
mand
of reli
able
power supply
.
Most
preci
s
i
on el
ect
r
oni
cs e
qui
p
m
ent
s and
p
o
we
r
e
l
ectro
n
ics eq
uipment
s
whi
c
h
are controll
e
d
by comput
ers a
nd micropro
c
e
s
so
rs
are sen
s
itive to the quality of power su
pply.
Electro
n
ics equipm
ents are
m
o
re sensitive
to
the influen
ce of power system th
an
electrome
c
h
a
n
ical e
quipm
ents an
d de
mand for
hig
h
req
u
ire
m
en
ts for po
we
r quality. Once
the
power g
r
id a
ppea
rs p
r
o
b
l
e
ms, ha
rm
s rang from e
c
onomi
c
losse
s
to enda
nge
red po
we
r grid,
equipm
ents
and pe
rsonal
safety, even for the com
m
unity unsta
ble whi
c
h m
a
y affect social
stability. Next-generation smart gri
d
is
compos
ed
of a great many
discrete
power
generati
ng,
transmitting
and
dist
ributi
ng e
quip
m
en
ts. Many
pro
b
lems su
ch as
volta
g
e
bi
as fluctuatio
n
of
regio
nal po
wer grid
s, ha
rmonic p
o
lluti
on and in
cre
a
sin
g
of rea
c
tive powe
r
factor are cau
s
e
d
by
the parall
e
l operation of more a
nd m
o
re small po
wer g
ene
rati
ng equip
m
en
ts su
ch a
s
solar
gene
rato
rs,
wind tu
rbin
e
s
an
d the
r
m
a
l po
wer ge
nerato
r
s. In
orde
r to e
n
sure th
e safe
and
eco
nomi
c
op
eration
of smart g
r
id a
n
d
ke
ep the
stability and self-heali
ng of
power
qualit
y in
microgri
d
, the
re
sea
r
ch of
power h
a
rm
o
n
ic
sup
p
re
ssi
on an
d rea
c
tive com
pen
sa
tion is b
e
com
i
ng
more a
nd mo
re urgent.
2. Related Works
2.1. Introduc
tion of Compress
ed Sen
s
ing
In 2006,
Davi
d L. Dono
ho
et al propo
se
d CS
(Comp
r
essed S
e
n
s
in
g) the
o
ry [8,
9], that
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 598
5 –
5998
5986
spa
r
se si
gnal
with suita
b
le
recon
s
tru
c
tio
n
algo
rithm can be
re
cove
red from a v
e
ry sm
all set
of
measurement
s that fa
r fe
wer than
co
nventional
m
easure
m
ent
l
i
mited
by Nyquist
th
eore
m
.
Acco
rdi
ng to CS, it can sa
mple and
co
mpre
ss PQ
informatio
n sy
nch
r
on
ou
sly without any p
r
ior
kno
w
le
dge. Gene
rally, CS theory ba
sically con
s
i
s
ts of three
steps: findi
ng the sp
arse
st
decompo
sitio
n
of a signal
, designi
ng a
pplicable
co
mpre
ssion re
pre
s
entin
g m
a
trix, which
well
approximate
s
the origin
al
sign
al
x
in lea
s
t co
efficient,
desig
ning
co
rre
sp
ondi
ng reco
nstruction
algorith
m
, wh
ich re
co
nst
r
u
c
ts o
r
iginal
si
gnal len
g
th in
N from ob
se
rved M coeffici
ents.
Acco
rdi
ng to
the theory, if origin
al sig
n
a
l
is
sp
arse o
r
in transfo
rm
domain th
e o
r
iginal
sign
al is
sp
a
r
se,
usi
ng a
ppro
p
ri
ate o
p
timization
al
gorithm
s
can
re
con
s
tru
c
t
origin
al si
gn
als
throug
h a fe
w num
be
r o
f
sample
d si
gnal
s an
d th
e numb
e
r
of sampl
ed
si
gnal
s u
s
ed i
n
recon
s
tru
c
tio
n
can
be far below the
n
u
mbe
r
of sa
mpled
signal
s in the alg
o
r
ithms b
a
sed
on
Nyqui
st theo
rem. CS t
heo
ry is n
o
t the
o
v
erall d
enial
of Nyqui
st th
eore
m
, but
u
s
ing
the
sp
arsity
of sig
nal
s to
re
con
s
truct
origin
al
sign
a
l
s th
roug
h fe
wer sampl
e
d
sig
nal
s than
the al
gorith
m
s
based on
Nyquist sta
nda
rds.
CS (Com
pre
s
sed Se
nsi
n
g
)
theo
ry ta
ke
advant
a
ges of
the spa
r
sit
y
of
sign
als and use
suitabl
e re
co
nstru
c
tion
al
gorithm
s to
reco
nstruc
t
original
signal
s throug
h a v
e
ry sm
all set of
observed
val
ues that far fewe
r tha
n
th
e si
gnal
s limit
ed by
Nyqui
st theore
m
. Compa
r
ed
with
the
previou
s
alg
o
rithm
s
ba
sed on
Nyq
u
i
st sa
mplin
g
theorem, CS theory h
a
s
the foll
owi
ng
advantag
es:
(1)
CS the
o
ry
’s samplin
g
speed i
s
fa
r lo
wer than
Nyq
u
ist’s. An
d CS theory do
e
s
glo
bal
observation
s
rathe
r
than local
sampli
n
g
; what is m
o
re, ea
ch ob
serve
d
value
contain
s
pa
rt o
f
effective info
rmation
of t
he
sign
al. In the
mea
n
w
hile,
CS th
eory
uses d
i
fferent o
b
se
rve
algorith
m
s e
v
ery time to ensure th
e
observ
ed v
a
lue
s
ha
s fe
wer i
n
form
ation re
dun
dan
cy.
Comp
ared
wi
th the tra
n
sfo
r
m
codi
ng in
spa
r
se b
a
si
s,
the
coeffici
e
n
ts’ lo
catio
n
i
s
n
o
lo
nge
r
so
important.
(2) In the a
s
pe
ct of de
cod
e
r, the d
e
co
der
ha
s high robu
stn
e
ss to the missi
n
g
informatio
n fo
r the
rea
s
on
that the
impo
rtance
of
ea
ch
proje
c
tion
co
efficient i
s
th
e same
an
d t
h
e
lose of several coeffici
ents will has fewer influence
to the reconstruction of origi
n
al signals.
(3) Co
mbini
n
g co
mpressio
n with
sam
p
li
ng, algo
rithm
s
ba
se
d o
n
CS theo
ry u
s
e fewe
r
memory
spa
c
e
s
and
co
m
puting resource
s than tr
a
d
itional samp
ling algo
rithm
s
. Applying t
he
saved resources in the later processing will r
educe the cost of sam
p
ling and transmission.
(4)
CS theo
ry can reli
e
v
e the com
put
ation bu
rden of har
d
w
are and le
ave th
e
c
o
mpu
t
a
t
io
ns to
c
o
mp
u
t
ers
in
la
te
r
pr
oc
es
s
a
nd a
c
hieve the sa
me re
con
s
tru
c
tive effects
of
origin
al sig
nal
with the tradi
tional algo
rith
ms by
usi
ng t
he po
we
rful p
a
rallel p
r
o
c
e
s
sing a
b
ilities
of
comp
uters while ke
eping
costs lo
w.
The a
m
ou
nts of sample
d
sign
als
are g
r
eatly
redu
ce
d by the
appl
ication
s
base
d
on
CS
theory, solving the
pro
b
l
e
ms i
n
si
gn
al pr
ocessin
g
, tran
smissi
on
an
d
storage. And th
ose
appli
c
ation
s
d
e
velope rapid
l
y in recent years:
In the aspe
ct
of spa
r
se rep
r
esentation of
signal, literature ten a
nd
el
even [10, 11] sho
w
s
that the Fourier
c
oeffic
i
ents
, wavelet
c
oeffic
i
ents o
f
smooth sig
nal, total variation norm of
boun
ded
vari
ation fun
c
tion
s , th
e G
abo
r
coeffici
ents
of
oscillato
r
sig
nal a
nd
Cu
rv
elet coefficie
n
t
s
of image
si
gn
al which
ha
s
discontin
uou
s edg
es
have
enou
gh
spa
r
sity. Howeve
r, ho
w to find
or
con
s
tru
c
t orth
ogon
al
b
a
si
s for
a
cl
ass of sign
als
in
ord
e
r to
get th
e
best
sp
arse
repre
s
e
n
tation
of
the sign
als i
s
the probl
em n
eede
d to be studied furthe
r.
In the aspe
ct of measure
m
ent
matrix,
literature twel
ve [
12] point
s out that u
n
der the
premi
s
e
of RIP (Re
s
tri
c
ted
Isometry Pro
perty, RI
P) p
r
incipl
e, we
sh
ould redu
ce t
he dime
nsi
o
n
s
of measu
r
e
m
ent matrix while en
su
re the loss
informatio
n of original signal is mini
mal.
No
wad
a
ys the measure
m
ent matrixes
applie
d in
CS theory are: Gau
ssi
an ra
n
dom matrix [10],
binary rand
o
m
matrix (Bo
unerlli m
a
trix), Fourie
r ran
d
o
m matrix [11
], Hadama
r
d
matrix etc.
In the aspe
ct of signal
re
covery
algo
ri
thms
whi
c
h
mean
s recon
s
tru
c
t ori
g
ina
l
sign
al
length in N from observed
M coefficient
s, literatur
e t
h
irteen a
nd fourtee
n
[13, 14], point out that
typical recovery alg
o
rithm
s
a
r
e BP
(B
asi
s
Pu
rsuit, BP) alg
o
rit
h
m, interi
or
point al
gorith
m
,
conj
ugate
g
r
adie
n
t p
r
oj
ection
algo
ri
thm and
iterative th
re
shol
d alg
o
rit
h
m etc.
Other
recon
s
tru
c
tio
n
algorith
m
s are OMP
(ortho
gon
al
matchin
g
pursuit OMP
)
algo
rithm,
TV
recon
s
tru
c
tio
n
algorith
m
a
nd other im
proved algo
rith
ms.
ROMP (Re
g
u
l
arized O
r
tho
gonal M
a
tchi
ng Pursuit, ROMP) alg
o
rit
h
m is a
nothe
r marke
d
improvem
ent
algorith
m
in
tradition
al m
a
tchin
g
p
u
rsuit
algorith
m
. ROMP alg
o
rith
m is devel
op
e
d
from traditio
n
a
l matchi
ng
pursuit alg
o
rit
h
m MP algo
ri
thm [16] and
OMP algo
rithm [17]. RO
MP
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
An Im
proved
Re
con
s
tru
c
tio
n
Algorithm
Base
d on Com
p
re
ssed Sen
s
ing for Power… (Yi Zhong
)
5987
algorith
m
is
on the b
a
si
s
of OMP algo
rithm and
u
s
e
regul
ari
z
atio
n method to
sele
ct elem
e
n
ts
whi
c
h
can
sel
e
ct several el
igible el
emen
ts to supp
ort
set an
d redu
cing th
e time
of power
sig
n
a
l
recon
s
tru
c
tio
n
.What i
s
m
o
re, the
red
u
c
ing
of ti
me is at the exp
e
n
se
of re
con
s
tru
c
tion q
ual
ity
and shoul
d know the
spa
r
sity at first.
T
h
e
pr
oc
es
s
o
f
R
O
MP
a
l
go
r
i
th
m is
to
us
e
reg
u
lari
zat
i
on metho
d
t
o
process bi
gge
st k
inner p
r
od
uct
s
of sen
s
ing
matrix
and resid
ual
y
and
sele
ct one re
quire
d eleme
n
t from the k
inner p
r
o
d
u
c
ts to re
con
s
tru
c
t origin
al sig
nal.
ROMP algo
ri
thm
co
mbine
s
the reg
u
la
ri
zation metho
d
with OMP algorith
m
to achi
eve
the goal of se
lecting m
o
re
element
s in o
ne iterat
io
n. ROMP alg
o
rit
h
m ca
n cla
s
si
fy elements f
a
st
and
sele
ct more
eleme
n
t
s in one ite
r
ation that
i
s
the rea
s
o
n
why RO
MP can g
e
t faster
recon
s
tru
c
tio
n
speed
tha
n
OMP
algo
rithm. Howe
ver, RO
MP
algorith
m
al
so ha
s its o
w
n
dra
w
ba
cks. T
h
is a
r
ticle
pro
posed a n
e
w algorit
hm
wh
ich i
s
ba
sed
on ROMP al
gorithm
but ca
n
achi
eve bette
r perfo
rma
n
ce in power qu
ality analysis.
2.2. Similarity
and Thres
hold Regular
ized Or
thog
onal Matc
hing Pursuit
Traditio
nal
ROMP (Reg
ul
arized
Orth
o
gonal
Matchi
ng Pu
rsuit, ROMP
) alg
o
r
ithm i
s
based on OM
P (Orthog
ona
l Matching P
u
rsuit, OMP)
algorith
m
and
uses
regul
ari
z
ation meth
o
d
to sel
e
ct
ele
m
ent. ROMP
algo
rithm
provides a
ne
w self
-ada
ptive algo
rithm to
achi
eve the
g
oal
of getting faster sp
eed of cl
assifying ele
m
ents an
d re
duci
ng the time of signal
reco
nstructio
n
.
Reg
u
lari
zatio
n
method is
a method wh
ich cl
assifies
element
s accordin
g to the energy
level of
ele
m
ents.
Re
gu
larization
me
thod i
s
de
scribed
a
s
foll
owe
d
: a
set
|
iN
A
xi
I
1
,
2
,
...,
N
IN
is the index set of
i
x
.Cla
ssify
ing all the ele
m
ents follo
w the rule:
||
2
|
|
,
,
mn
k
x
xm
n
I
(
1
)
And cla
ssifie
s
index set
I
into several subset
s
1,
2
,
3
.
.
.
k
Ik
and sele
cts maxi
mu
m
energy sub
s
e
t
0
I
in the last, that is,
0
m
a
x
{
,
1
,
2
,...
,
}
k
II
X
Xk
K
. Regul
arization met
hod can
cla
ssify elem
ents fa
st and
sele
ct mo
re
element
s
in o
ne
iteratio
n which brin
gs
ROMP
algo
rith
m
fas
t
er recons
truc
tion s
p
eed.
Ho
wever,
ROMP algo
rith
m ha
s its o
w
n sh
ort
c
omin
gs. Th
e alg
o
rithm is u
n
re
a
s
on
able
that each tim
e
the algo
rith
m can
only select on
e gro
up whi
c
h
ha
s the maximu
m total energ
y
to
the ca
ndid
a
te
set an
d leav
es oth
e
r g
r
o
u
p
s
whi
c
h h
a
ve simila
r e
n
e
r
gy with the
maximum total
energy to the
next iteratio
n. ROMP al
g
o
rithm b
r
ing
s
a lot of red
u
ndant
comp
u
t
ation whi
c
h i
s
a
wa
ste of re
source
s an
d d
e
mand
s for
h
i
gher
per
fo
rm
ance of equi
pments
whi
c
h lead
s to m
o
re
equipm
ent co
sts. In view of the whole
iterat
ive pro
c
ess, the tasks whi
c
h can
be done in o
ne
iterative are
divided into
several ite
r
atio
ns, wa
stin
g t
he time a
nd
reso
urce
s a
n
d
decre
asi
ng t
h
e
effic
i
enc
y
.
This p
ape
r p
o
ints out that
the maximum
total energy
grou
p and
other g
r
ou
ps
which a
r
e
simila
r to m
a
ximum total e
nergy
gro
up
and h
a
ve
sim
ilar e
nergy wi
th the maxim
u
m total e
nergy
grou
p sh
ould
be put into the can
d
idate
set in the sam
e
iteration.
This pa
pe
r prop
osed ST
ROMP (Simi
l
arity and T
h
re
shol
d Re
gulari
z
e
d
Ort
hogo
nal
Matchin
g
Pu
rsuit)
algo
rith
m whi
c
h
ba
sed on
RO
MP
(Regul
ari
z
ed
Ortho
gon
al
Matchin
g
Pu
rsuit)
algorith
m
. STROMP alg
o
rit
h
m cha
nge
s the rule
s of se
lecting el
eme
n
ts.
The th
re
shol
d pa
ram
e
ter is
a
, ene
rgy i
s
E
, ene
rgy
correlation
is
M
and avera
ge
energy co
rrel
a
tion is
S
:
2
1
1
N
i
i
Ex
N
(2)
Whe
r
e
i
x
is the member of
set
|
1
,
2
,
3
...
i
A
xi
N
,
N
is the total numbe
r of se
t
A
.
1
,
2
,
3
...
ii
ME
E
i
P
(3)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 598
5 –
5998
5988
1
2
1
(,
)
1
p
j
j
SE
E
p
(4)
Whe
r
e
i
E
is the energy of different vecto
r
i
J
.
Use regul
ari
z
ation meth
od
to cla
s
sify inner
p
r
odu
ct
s of
sign
al resid
ual a
nd
sen
s
in
g
matrix into several
group
s
12
3
,
,
...
,
p
JJ
J
J
by energy.
Calculate
the energy
12
1
2
,
,..
.
(
...
)
p
p
EE
E
E
E
E
of
12
3
,
,
...
,
p
JJ
J
J
therefo
r
e
all the g
r
o
u
p
s
whi
c
h e
n
e
r
gy are
ab
ove
1
*
aE
and
the
ene
rgy correlatio
n
(
2
,
3
,
4
...
)
i
Mi
p
is a
bove t
he ave
r
ag
e e
nergy
co
rrela
t
ion
S
are sele
cted to the can
d
ida
t
e set in the same iteratio
n.
Figure 1.
Flow Ch
art of V
o
ltage/Cu
rrent Signal Co
mpressed Sen
s
in
g
The ste
p
s of
STROMP alg
o
rithm are as
follows:
Inputs: Sen
s
ing sig
nal
y
, sensin
g matrix
,
spa
r
sit
y
k
and threshold
coe
fficient
a
In this
way, th
e g
r
oup
s whi
c
h
wo
uld
be
selecte
d
in
several ite
r
atio
ns befo
r
e
are
select
e
d
in one iteratio
n, redu
cing th
e iteration
tim
e
s an
d avoidi
ng unn
ecessary step
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
An Im
proved
Re
con
s
tru
c
tio
n
Algorithm
Base
d on Com
p
re
ssed Sen
s
ing for Power… (Yi Zhong
)
5989
Initialization:
Re
sidu
al r0 = y, support se
t
0
F
=
, iterator t = 1.
Step 1:
Use regula
r
i
z
ation meth
od
to sele
ct
k la
rge
s
t elem
ent
s of inn
e
r
pro
duct of
re
sidu
al and
sen
s
in
g m
a
tri
x
by ab
sol
u
te value.
Ma
rk the
g
r
ou
p
12
3
,
,
...
,
p
JJ
J
J
from larges
t to
s
m
alles
t
by
energy and th
e energy of
12
3
,
,
...
,
p
JJ
J
J
i
s
:
12
1
2
,
,..
.
(
...
)
p
p
EE
E
E
E
E
.
Step 2:
Us
e
S
and
1
*
aE
to cla
ssify the
grou
p
12
3
,
,
...
,
p
JJ
J
J
which
mean
s that
the gro
u
p
s
whi
c
h e
n
e
r
gy
are
ab
ove
1
*
aE
and the
en
ergy co
rrelatio
n
M
betwee
n
t
he g
r
ou
p an
d
1
E
is
above averag
e energy
co
rrelation
S
are selec
t
ed into the c
a
ndidate s
e
t
J
.
Step 3:
Cal
c
ulate the
recon
s
tru
c
tio
n
sign
als by the lea
s
t squ
a
r
e metho
d
:
2
ar
g
m
i
n
t
tx
F
X
yx
, and updatin
g resi
dual.
t
tF
t
ry
x
.
Step 4:
if
2
nk
,then upd
ate the numbe
r of cycles
1
tt
and go to step 1
or exit the loop.
Output:
t
F
x
y
;
The ra
nge of
the threshol
d coeffici
ent
(0
,1
]
a
,
in the re
sea
r
ch sta
ge of this pa
per fo
r
power qu
ality, the reco
nstruction effe
ct will
be be
st when the thre
shold value a i
s
0.6.
Acco
rdi
ng to
the STROMP
(Similarity a
nd Threshold
Regul
ari
z
ed
Orthog
onal M
a
tchin
g
Pursuit) algo
rithm, the flow of the reco
nstruction al
go
ri
thm is sh
own
in Figure 1:
3. Experiment and Per
f
o
r
mance Anal
y
s
is
3.1. Rese
arc
h
Criteria a
n
d Platform Design
There is no
unified
stand
ard i
n
po
we
r qualit
y ge
ne
rally an
d IEC definition fo
r power
quality is th
at power qu
ality is the
physi
cal
cha
r
acte
ri
stics o
f
powe
r
sup
p
ly device’
s
no
t
disturbing
an
d interru
p
ting
user’
s
usin
g
ele
c
tr
icity un
der
no
rmal worki
ng co
nditi
on.
Mea
s
u
r
in
g
the voltage
current a
nd
p
o
we
r in
singl
e ph
ase p
o
wer
grid
in
re
al time i
s
hel
p to
study
a
n
d
analyze the chara
c
te
risti
c
s of power q
u
a
lity.
In stable con
d
ition of linea
r load, voltag
e
and current
signal
s a
r
e b
o
th sine
wave
forms in
50Hz theo
reti
cally. But in
unsta
ble con
d
ition of
nonl
inear lo
ad, they are affect
ed disto
r
ted
by
some in
du
ctances, capa
citan
c
e
s
, or other no
nline
a
r facto
r
s. PQ Harmoni
cs have N*
50
Hz
freque
ncy
affected
si
gnal
s [15]. Voltag
e an
d
curre
n
t sig
nal
s m
a
inly con
s
ist
of pe
riodi
c
or
qua
sipe
riodi
c signal
s in p
r
acti
cal
condi
tion, and it exists a lot of information
redu
nda
ncy i
n
perio
ds o
r
bet
wee
n
peri
o
d
s
.
We
will i
n
tro
duce
several
pe
rform
a
n
c
e
indexe
s
:
CS
R
(Compression Sam
p
ling
Ratio
)
,
SNR
(Signal
to Noise Ratio), MSE (Mean Sq
u
a
red Error), a
nd ERP (En
e
rgy Recove
ry
Percentag
e) to obje
c
tively appraise the
reco
nstructe
d results of PQ sign
als.
100%
c
N
CS
R
N
(5)
2
1
2
1
()
10
l
g
ˆ
()
()
N
i
N
i
fi
SNR
fi
fi
(6)
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 598
5 –
5998
5990
2
1
2
1
ˆ
()
()
10
0%
()
N
i
N
i
fi
fi
MS
E
fi
(7)
2
1
2
1
ˆ
()
()
N
i
N
i
f
i
ES
P
f
i
(8)
Whe
r
e
N
is tot
a
l sa
mpling
numbe
r of
original
sign
als,
c
N
is the
re
se
rved sa
mple
numbe
r of
sign
als after
spa
r
se sa
mpl
i
ng, and
ˆ
()
f
i
is the recon
s
tru
c
t
ed sig
nal.
Based
on the
o
retical re
se
a
r
ch
purpo
se
s ment
ioned
a
bove, we e
s
tablish a set o
f
smart
meter test-be
d
with CS measure
m
ent. In Figure 2,
this
tes
t
-bed platform c
o
ns
is
ts
of three
wirel
e
ss n
o
d
e
s a
s
smart
meters, ZigB
ee wi
rele
ss
g
a
teway, Ethe
rnet ro
uter a
nd termin
al d
a
ta
serve
r
PC. For a singl
e meter nod
e, energy meteri
ng
chip ADE78
7
8
measure
s
a
ll information
of
singl
e-p
h
a
s
e load with
inte
rnal hardware
si
gnal ci
rcu
i
t and o
b
tain
s voltage/curre
n
t value, a
c
tive
power, re
acti
ve powe
r
, ap
pare
n
t po
wer and etc. It transmit
s
sam
p
ling data to
STM32 sy
ste
m
with SPI interface. STM32
system reali
z
es PQ
data storage with
m
u
lti-tasking o
peratin
g
sy
stem
(uC-OS
). In ZigBee wi
rele
ss network, th
e mea
s
ur
i
ng
node tran
smits all of PQ informatio
n to PC
serve
r
. And we ca
n co
nfigure n
ode’
s chargi
ng settings onlin
e with
infrared rem
o
te controller.
Its
LCD display sho
w
s PQ pa
ramete
rs
real
-time dynami
c
ally.
Figure 2.
Smart Meter
T
e
st-bed
with CS
Measurement
In orde
r to comp
are the perform
an
ces
of MP algorithm, OM
P algorithm,
ROMP
algorith
m
an
d STROMP
algorith
m
, we
do a lot of
contrast exp
e
rime
nts whi
c
h is
sho
w
form
Figure 3 to Figure 1
4
and t
he perfo
rma
n
c
e ind
e
xes i
s
sho
w
in Tabl
e 1.
T
abl
e 1.
Stati
s
tic
a
l
Rec
o
very Parameters
of Compre
ssing V
o
lta
ge/
Curre
n
t Data
from Dif
f
erent
R
e
c
o
ns
tr
uc
tion
Alg
o
r
ith
m
MP algorithm
SNR(dB)
MSE (
%
)
ERP (
%
)
voltage of load 1
27.8448
4.0528
97.0766
current of load
1
21.2720
8.6377
99.0027
voltage of load 2
26.8008
4.5704
97.3575
current of load
2
11.6741
26.0794
98.2753
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
An Im
proved
Re
con
s
tru
c
tio
n
Algorithm
Base
d on Com
p
re
ssed Sen
s
ing for Power… (Yi Zhong
)
5991
OMP algorithm
SNR(dB)
MSE (
%
)
ERP (
%
)
voltage of load 1
33.7290
2.0585
99.5871
current of load
1
21.7750
8.1518
99.2397
voltage of load 2
36.1172
1.5636
99.0415
current of load
2
19.4868
106087
99.6135
ROMP algorithm
SNR(dB)
MSE(
%
)
ERP(
%
)
voltage of load 1
37.2835
1.5902
99.9720
current of load
1
28.9687
3.5610
100.5346
voltage of load 2
35.9712
1.6672
99.8083
current of load
2
27.7210
4.1110
99.6759
STROMP algorit
hm
SNR(dB)
MSE(
%
)
ERP(
%
)
voltage of load 1
43.0996
0.6999
100.1646
current of load
1
38.3961
1.2028
100.2397
voltage of load 2
42.0181
0.7927
99.9847
current of load
2
37.2896
1.3662
99.9020
(a)
(b)
(c
)
(d)
Figure 3.
V
o
ltage Recovery Signal and Origin
al Sign
al Comp
ari
s
o
n
throug
h Dif
f
erent
Algo
rith
ms
in Load 1: (a
) is MP
algorit
hm, (b) i
s
OM
P
algorithm, (c) i
s
ROMP
a
l
gorithm a
nd
(d) i
s
STPOM
P
algorithm
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 598
5 –
5998
5992
(a)
(b)
(c
)
(d)
Figure 4.
The
FFT
Comp
arison of V
o
ltag
e Re
covery S
i
gnal an
d Ori
g
inal Signal t
h
rou
gh Dif
f
e
r
ent
Algorithm
s in Load 1: (a
) is
MP
algorithm
, (b) is OMP
algorith
m
, (c) is RO
MP
alg
o
rithm an
d (d
)
is STPOMP
a
l
gorithm
(a)
(b)
(c
)
(d)
Figure 5
.
The
Recover Error of V
o
ltage
Re
covery Sig
nal and O
r
igi
nal Signal through
Dif
f
eren
t
Algorithm
s in Load 1: (a
) is
MP
algorithm
, (b) is OMP
algorith
m
, (c) is RO
MP
alg
o
rithm an
d d is
STPOMP
algorithm
Evaluation Warning : The document was created with Spire.PDF for Python.
TELKOM
NIKA
ISSN:
2302-4
046
An Im
proved
Re
con
s
tru
c
tio
n
Algorithm
Base
d on Com
p
re
ssed Sen
s
ing for Power… (Yi Zhong
)
5993
(a)
(b)
(c
)
(d)
Figure 6
.
Current Re
cove
ry Signal and Origin
al Sign
al Comp
ari
s
o
n
throug
h Dif
f
erent
Algo
rith
ms
in Load 1: (a
) is MP
algorit
hm, (b) i
s
OM
P
algorit
hm, (c) i
s
ROMP
a
l
gorithm a
nd (d) is STPOM
P
algorith
m
(a)
(b)
(c
)
(d)
Figure 7.
The
FFT
Comp
arison of Curre
n
t Recovery Signal and O
r
iginal Signal t
h
rou
gh Dif
f
e
r
ent
Algorithm
s in Load 1: (a
) is
MP
algorithm
, (b) is OMP
algorith
m
, (c) is RO
MP
alg
o
rithm an
d (d
)
is STPOMP
a
l
gorithm.
Evaluation Warning : The document was created with Spire.PDF for Python.
ISSN: 23
02-4
046
TELKOM
NI
KA
Vol. 12, No. 8, August 2014: 598
5 –
5998
5994
(a)
(b)
(c
)
(d)
Figure 8.
The
Recover Error of Cu
rre
nt Re
covery Sig
nal and O
r
igi
nal Signal through
Dif
f
eren
t
Algorithm
s in Load 1: (a
) is
MP
algorithm
, (b) is OMP
algorith
m
, (c) is RO
MP
alg
o
rithm an
d (d
)
is STPOMP
a
l
gorithm
(a)
(b)
(c
)
(d)
Figure 9. V
o
ltage Recovery Signal and Origin
al Sign
al Comp
ari
s
o
n
throug
h Dif
f
erent
Algo
rith
ms
in Load 2: (a
) is MP
algorit
hm, (b) i
s
OM
P
algor
ithm, (c) i
s
ROMP
a
l
gorithm a
nd
d is STPOMP
algorith
m
Evaluation Warning : The document was created with Spire.PDF for Python.