Academic Edi o : Robe o Zi ie i
Recei ed: 12 Decembe 2024
Re ised: 23 Decembe 2024
Accep ed: 25 Decembe 2024
Published: 28 Decembe 2024
Ci a ion: A enas-Ramos, V.; Cues a,
F.; Palla es-Lopez, V.; San iago, I.
So wa e In eg a ion o Powe Sys em
Measu emen De ices wi h AI
Capabili ies. Appl. Sci. 2025,15, 170.
h ps://doi.o g/10.3390/
app15010170
Copy igh : © 2024 by he au ho s.
Licensee MDPI, Basel, Swi ze land.
This a icle is an open access a icle
dis ibu ed unde he e ms and
condi ions o he C ea i e Commons
A ibu ion (CC BY) license
(h ps://c ea i ecommons.o g/
licenses/by/4.0/).
A icle
So wa e In eg a ion o Powe Sys em Measu emen De ices
wi h AI Capabili ies
Vic o ia A enas-Ramos 1,*,† , Fede ico Cues a 2,*,† , Vic o Palla es-Lopez 1and Isabel San iago 1
1
Depa amen o de Ingenie ía Elec ónica y de Compu ado es, Campus de Rabanales, Uni e sidad de Có doba,
14071 Có doba, Spain; palla [email p o ec ed] (V.P.-L.); [email p o ec ed] (I.S.)
2Escuela Técnica Supe io de Ingenie ía, Uni e sidad de Se illa, Camino Descub imien os,
E-41092 Se illa, Spain
*Co espondence: [email p o ec ed] (V.A.-R.); [email p o ec ed] (F.C.)
†These au ho s con ibu ed equally o his wo k.
Fea u ed Applica ion: Dis ibu ed Powe Sys ems Moni o ing.
Abs ac : The la es changes on he dis ibu ion ne wo k due o he p esence o dis ibu ed
ene gy esou ces (DERs) and elec ic ehicles make i necessa y o moni o he g id using
a eal- ime high-p ecision sys em. The p esen wo k cen e s on he de elopmen o an
open-sou ce so wa e pla o m ha allows o he join managemen o , a leas , powe
quali y moni o s (PQMs), phaso measu emen uni s (PMUs), and sma me e s (SMs),
which a e h ee o he mos widesp ead de ices on dis ibu ion ne wo ks. This amewo k
could wo k emo ely while allowing access o he measu emen s in a com o able way
o g id analysis, p edic ion, o con ol asks. The pla o m mus mee he equi emen s
o synch onism and scalabili y needed when wo king wi h elec ical moni o ing de ices
while conside ing he la ge olumes o da a ha hese de ices gene a e. The amewo k
has been expe imen ally alida ed in labo a o y and ield es s in wo pho o ol aic plan s.
Mo eo e , eal- ime A i icial In elligence capabili ies ha e been alida ed by implemen ing
h ee Machine Lea ning classi ie s (Neu al Ne wo k, Decision T ee, and Random Fo es ) o
dis inguish be ween h ee di e en loads in eal ime.
Keywo ds: powe quali y moni o ; phaso measu emen uni ; sma me e ; open-sou ce
so wa e; dis ibu ed measu emen sys em; pho o ol aic plan ; load iden i ica ion; a i icial
in elligence; neu al ne wo k; decision ee; andom o es
1. In oduc ion
In ecen yea s, he e has been a signi ican imp o emen in he de ices a ailable o
moni o ing elec ical sys ems ocused on di e en applica ions, in e ms o hei pe o -
mance, a ie y, cos , a ailabili y communica ion capabili ies, acquisi ion, o p ocessing
so wa e. An in e es ing e iew o he di e en moni o ing de ices ha appea in a sma
g id can be ound in [
1
], whe e he undamen al cha ac e is ics and applica ions o h ee o
he mos common ones a e b ie ly desc ibed: a phaso measu emen uni (PMU), a powe
quali y equipmen (PQM), and a sma me e (SM).
This pe o mance imp o emen has been enhanced, among o he s, by he no able
inc ease in dis ibu ed ene gy esou ce (DER) en i onmen s, pho o ol aic (PV) panels, and
elec ic ehicle cha ging, which pose subs an ial challenges om a powe quali y poin
o iew [
2
,
3
]. In his way, moni o ing de ices a e enabling he de elopmen o a ious
Appl. Sci. 2025,15, 170 h ps://doi.o g/10.3390/app15010170
Appl. Sci. 2025,15, 170 2 o 27
applica ions a di e en le els. Fo example, hey ha e been used in aul de ec ion: us-
ing powe quali y moni o ing da a o es ima e aul loca ion on dis ibu ion eede s [
4
],
islanding he loca ion o a g ound aul [
5
], au oma ic aul loca ion, and high impedance
aul s loca ion on dis ibu ion ne wo ks using PMUs [
6
–
8
]. Ano he majo se o appli-
ca ions in ol es he es ima ion o he s a e o he dis ibu ion sys em [
9
–
11
]. In [
12
], a
ecen and comp ehensi e su ey is p esen ed on he es ima ion o he s a e o he powe
sys em using mul iple da a sou ces. Ano he a ea o in e es in ol es he applica ions
o eal- ime powe low moni o ing and con ol in mic og ids [
13
] o a s abili y s udy
using enewable sou ces [
14
]. These p e ious applica ions a e ypically ca ied ou using
bo h PQMs and PMUs, while o Non-In usi e Load moni o ing (NILM), applica ions o
dis ibu ed ene gy esou ces (DERs) [
15
], esiden ial se ings [
16
], o indus ial se ings [
17
]
bo h SMs and PQMs a e used. On a di e en no e, he a ailabili y o la ge olumes o
gene a ed da a has made i common o use A i icial In elligence echniques o hese
asks [
18
]: ecen e iews o machine lea ning echniques in powe sys em secu i y and
s abili y [19,20], aul diagnosis [21], and NILM applica ions [16,17,22] can be ound.
Despi e he bene i s, in eg a ing hese de ices poses signi ican challenges. Each elec-
ical moni o ing de ice has unique cha ac e is ics and equi emen s, such as ansmission
speed and synch oniza ion, ha di e en ia e i om o he s. As a esul , each de ice ypi-
cally ope a es wi hin i s own specialized so wa e. In addi ion, each manu ac u e o en
p o ides p op ie a y so wa e ha is easy o use wi h hei equipmen bu incompa ible
wi h o he s. This c ea es a highly he e ogeneous en i onmen whe e linking da a om
di e en sou ces is challenging. Howe e , in g ids whe e di e en elec ical moni o ing
de ices a e used simul aneously, i is highly desi able o ha e all hese a iables uni ied in
a single da a s eam. A la ge numbe o applica ions would bene i om handling join
da a [23].
I is he e o e o in e es o implemen an a chi ec u e based on open-sou ce ools ha
allows and acili a es he in eg a ion o measu emen s om di e en de ices, p o ides
eal- ime access and in e ac ion wi h he da a, and includes he possibili y o implemen ing
sys ems based on A i icial In elligence. Da a mus be s o ed synch onously, despi e he
a ying le els o synch onism among he de ices. Ha ing synch onized measu emen s
acili a es he co ela ion o measu emen s om di e en and dis an equipmen , e en i
he compa ison is limi ed by he lowe -quali y equipmen .
Al e na i e so wa e ools a e a ailable in hese bibliog aphy [
24
–
27
], al hough hey
ha e limi a ions in e ms o lexibili y and scalabili y when gi en he di e en applica ions
o which hey a e in ended. Fo example, in [
24
], end da a om he PQM a e no aken
in o accoun ; only e en da a egis e ed om he PQM e en eco de a e conside ed.
In [
25
], i was necessa y o ex ac he s o ed da a in he CSV/JSON o ma , which would
be a huge d awback o any applica ion ela ed o A i icial In elligence in eal ime due
o he o ma slowness. In [
26
], a e sion is p esen ed ha imp o es he managemen
o CSV/JSON packages, bu he main p oblem emains. Finally, in [
27
], a solu ion is
p oposed whe e PQM and PMU da a a e admi ed hough he use o he Message Queuing
Teleme y T anspo (MQTT); he ein, cloud compu ing signi ican ly slows down he da a
low, which has led o he conclusion ha his solu ion equi es u he op imiza ion. In
his con ex , his wo k p esen s and implemen s a new al e na i e de eloped using widely
ecognized open-sou ce ools o in eg a ing he e ogeneous elec ical moni o ing de ices.
This app oach aims o mi iga e he abo emen ioned issues while acili a ing he applica ion
o machine lea ning echniques. In addi ion, se e al expe imen al es s a e conduc ed o
demons a e he sys em’s capabili ies.
This a icle con inues in Sec ion 2wi h a desc ip ion o he mos commonly used
elec ical moni o ing de ices. Following his, Sec ion 3p esen s he in eg a ion amewo k
Appl. Sci. 2025,15, 170 3 o 27
we de eloped. Sec ion 4 hen p esen s wo expe imen al es cases. Subsequen ly, Sec ion 5
demons a es he in eg a ion o he p oposed ool wi h eal- ime machine lea ning-based
decision sys ems. Finally, he a icle ends wi h a conclusion sec ion.
2. Moni o ing De ices
As men ioned in he In oduc ion, he e is a wide ange o elec ical moni o ing de-
ices on he ma ke , each wi h di e en cha ac e is ics and pe o mances. The quali y
o he moni o ing sys em will depend on bo h he equipmen i is composed o and i s
dis ibu ion o e he g id. High-quali y equipmen will e u n mo e accu a e and comple e
in o ma ion han o he low-end de ices. Fu he mo e, depending on he cha ac e is ics o
each de ice, i may be ad isable o place i close o o u he away om he consume s o
p oduce s
[3,28–30]
. Fo DERs such as PV plan s, he placemen and quali y o he moni o -
ing de ices a e c ucial o eal- ime moni o ing o ene gy p oduc ion, consump ion, and
s o age, enabling p ecise con ol and op imiza ion o main ain g id s abili y and e iciency.
Figu e 1shows an example o a ypical dis ibu ion ne wo k, whe e PQMs a e placed
nea he consume s, wi h PMUs a he beginning o each line and an SM a each con-
sume . This sec ion explains wha his equipmen consis s o and he easons ha jus i y
i s placemen .
Figu e 1. Example o a dis ibu ion ne wo k wi h meassu emen de ices.
2.1. Powe Quali y Moni o s (PQMs) and Powe Quali y Analyze s (PQAs)
The IEEE S d C37.2-2008 s anda d [
31
] de ines a PQM as a de ice capable o mon-
i o ing he elec ical pa ame e s used in powe quali y (PQ) analysis, speci ically hose
pa ame e s ha de e mine compliance o non-compliance wi h supply quali y egula ions.
These pa ame e s include RMS ol age and cu en alues, equency alues, powe , phase
unbalance, an ha monics and in e ha monics in an elec ical signal. Along wi h hese, pa-
ame e s o o he na u e may also appea . An “ex ended” e sion o his ype o equipmen
is he Powe Quali y Analyze (PQA), which u he p ocesses he da a, checking whe he
he measu emen s comply wi h he s anda ds o no . PQMs and PQAs usually also include
an e en eco de apa om hei end da a. The di e ence be ween a PQM and a PQA is
sub le; howe e , a PQA can be ega ded as an enhanced e sion o a PQM. In gene al, he
pa icula pa ame e s hey measu e and he way hey a e ea ed will s ongly depend on
he objec i es when ins alling a me e .
All measu emen s collec ed by a PQM a e de i ed om hose collec ed by Cu en
T ans o me (CT) and Powe T ans o me (PT) modules, along wi h addi ional measu e-
men s ha equi e u he p ocessing, such as ha monic measu emen s and To al Ha monic
Appl. Sci. 2025,15, 170 4 o 27
Dis o ion (THD) calcula ion. In h ee-phase de ices, mos measu emen s a e p esen ed
h ee imes. Mo eo e , hese measu emen s a e o en shown wi h di e en ime agg e-
ga ions, signi ican ly inc easing he amoun o a ailable da a. As a esul , he olume o
end da a can be qui e subs an ial. The e is no egula ion o s anda d ha indica es which
pa ame e s should be chosen, al hough IEEE S d 1159-2019 [
32
] gi es some ecommenda ions.
2.2. Phaso Measu emen Uni (PMU) and µPMU
A PMU is a de ice de eloped wi h he pu pose o measu ing ol age and cu en
phaso s, equency, and he Ra e O Change O F equency (ROCOF) synch onized wi h an
absolu e ime e e ence h ough a Global Posi ioning Sys em (GPS). P ecisely, one o he
main ea u es o PMUs is hei high sampling a e and ime accu acy. Fo a g id equency
o 50Hz, sample equencies o 10, 25, o 50 F ames pe Second (FPS) a e possible. This
implies measu ing o egis e ing da a e e y 100 ms, 50 ms, o 20 ms, espec i ely. Time
synch oniza ion ensu es p ecise alignmen o he da a, allowing o an e ec i e compa ison
o measu emen s aken a di e en poin s on he g id. This opens up he op ion o mo e
e icien managemen , enhancing p oblem de ec ion and inc easing he abili y o add ess
issues. This is why PMUs a e widely used in bo h he ansmission and dis ibu ion s ages
o he powe sys em [
11
,
12
]. Thei accu acy is also signi ican , eaching
±
1% in magni ude
and ±1◦in phase accu acy.
In ecen yea s, he e has been a la ge g ow h in he use o DERs. These sys ems
in oduce conside able complexi ies and e ec s in he g id ha equi e ho ough s udy.
While PMUs ha e been e ec i e in moni o ing, his inc ease in complexi y calls o a
echnology ha can de ec i egula i ies wi h e en g ea e accu acy and highe esolu ion
han con en ional PMUs.
Tha is why he so-called
µ
PMUs [
33
] we e de eloped. These de ices each a es
o 100phaso s/s o a signal o 50Hz and 120 phaso s/s o 60Hz, meaning hey e u n
ins an aneous g id alues e e y hal cycle. They also exceed he measu emen accu acy
gi en by PMUs, achie ing angula accu acy alues o up o
±
0.010
◦
and ampli ude accu acy
alues o up o ±0.05%. They a e o he wise simila equipmen .
2.3. Sma Me e
A sma me e calcula es consump ion o p oduc ion in a mo e de ailed way han
con en ional me e s. Domes ic me e s in u ban low- ol age ne wo ks a e common. They
allow local se ice p o ide s o moni o and manage he consump ion o each use e ec-
i ely. Thei accu acy anges be ween
±
1% and
±
2%. They usually ha e an adjus able
sending a e be ween 1 and 60 min.
2.4. Applica ions and Cha ac e is ic Compa ison
Each o hese de ices has a common use in a leas one speci ic applica ion. Howe e ,
o en bo h he PMU and he PQM a e de ices ha , due o hei cha ac e is ics and measu e-
men accu acy, lend hemsel es o a la ge numbe o applica ions. SMs a e mo e dedica ed
o hei applica ion o egula ing powe consump ion, bu p ecisely o his eason, hey
a e also e y common. Table 1shows a la ge numbe o common applica ions ound in
he li e a u e ha can be co e ed by hese de ices o a g ea e o lesse ex en . In addi ion,
Table 2shows a compa ison o he communica ions cha ac e is ics o he h ee elec ical
moni o ing de ices. In [
1
,
34
], he e a e mo e in-dep h e iews o bo h he equipmen and
hei applica ions.
Appl. Sci. 2025,15, 170 5 o 27
Table 1. Summa y o applica ions o µPMU, PQM and SM.
Applica ion µPMU PQM 1Sma Me e
Moni o ing Yes [14]Yes: designed o his
pu pose [4,13]Yes [35,36]
Ha monic Analysis No Yes [37] No
E en De ec ion Yes [38–40]Mos de ices include
an e en eco de No
P o ec ion and/o
Con ol
Yes: designed o his
pu pose [41,42]Less ecommended No
Faul Loca ion Yes [6–8] Yes [4,5] Yes [43]
Modeling S a e
es ima ion [9,10,12]Load modeling [2,44]Load
modeling [45,46]
Topology
Iden i ica ion Yes [47,48] No No
Demand Fo ecas ing
and Managemen Less ecommended Yes [49] Yes [50–52]
1Only end da a is conside ed; e en da a is excluded.
Table 2. µPMU, PQM, and SM communica ion capabili y compa ison.
Fea u e µPMU PQM Sma Me e
Sampling Ra e 25,600 m/s a 50 Hz 25,600 m/s a 50 Hz om 1600 m/s up
o 51,200 m/s
T ansmission Ra e om 10 ms up o
100 ms
om 1 s up o
10 min
om 1 s up o
10 min
T ansmission
Speed
C ucial. IEEE S d
C37.118.2 [53]No c ucial No c ucial
Measu es
Vol age and cu en
phaso s. F equency.
ROCOF
PQ da a. F equency,
Vol age, Cu en ,
powe , THD, TDD,
ha monics, e c.
Powe and ene gy
da a. F equency,
ol age, cu en , e c.
Da a Volume Ve y high High Low
Synch onism Ve y impo an .
GPS.
Less impo an .
NTP (usually) Less impo an
Mos Common
P o ocol
IEEE S d. C37.118.2.
Modbus ASCII Modbus RTU
3. P oposed In eg a ion F amewo k
3.1. Requi emen s
The aim o his wo k is o c ea e a so wa e a chi ec u e ha can accommoda e he
measu emen s o di e en elec ical moni o ing de ices while making he da a a ailable
o any a i icial in elligence ask in eal ime. Pa icula a en ion is gi en o PMUs, as
hey exhibi dis inc ly di e en beha io compa ed o he o he equipmen discussed in
he p e ious sec ion (see Table 2). On he con a y, PQMs and SMs shows some simi-
la i ies, as bo h de ices collec powe and ene gy da a. Usually, PQMs include a wide
ange o measu es, including mo e de ailed powe quali y da a. Howe e , he speci ic
lis o measu emen s may a y based on he manu ac u e . They bo h ely on s anda d
communica ions p o ocols on he ma ke .
The case o PMUs is di e en . Fo hem, sample synch oniza ion, e icien s o age,
and high ansmission speed a e essen ial. Since PMUs a e capable o sending up o
100 FPS, e y high da a olumes can be eached in a sho pe iod o ime. This equi es an
Appl. Sci. 2025,15, 170 6 o 27
in as uc u e ha can no only s o e a la ge olume o da a bu also access i e icien ly. The
in as uc u e mus also be able o espec he o iginal ime s amping o he PMU. PMUs
gene ally use he IEEE C37.118.2 s anda d [
53
]. Wi hin his s anda d, he need o a Phaso
Da a Concen a o (PDC), a so wa e ha collec s and manages he phaso s, is speci ied.
This a icle p oposes a alid amewo k o PMUs, SMs, and PQMs ha collec
da a based on comme cial p o ocols while inco po a ing he s anda ds o he IEEE S d.
C37.118.2 and he use o a PDC. Fi s ly, in Sec ion 3.2, a b ie ske ch o he design is shown.
Then, in Sec ion 3.3, he ools chosen a e explained, oge he wi h some al e na i es o
simila pe o mance.
The objec i es ha a e he e o e se o he a chi ec u e as a whole, and which mus
he e o e be me by each and e e y one o he ools used, a e as ollows:
1.
In eg a ion capabili y: Since he main goal o s o ing da a join ly is i s compa ison
and co ela ion, he sys em should be in eg able wi h a wide ange o p o ocols.
2. Scalabili y: The abili y o manage a la ge olume o da a.
3.
Easy use access: Que ies should be use - iendly and e icien in e ms o ime. Da a
mus be accessible o bo h analysis and eal- ime moni o ing.
4.
Open Sou ce: I aims o emb ace he open-sou ce philosophy, making he sys em
accessible o all use s wi hou eliance on p op ie a y so wa e.
5.
Real-Time capabili y: The sys em mus p o ide eal- ime moni o ing capabili ies.
Addi ionally, in o de o suppo AI-d i en applica ions, i should be able o execu e
machine lea ning p ocesses in eal ime.
3.2. Sys em A chi ec u e
Figu e 2shows he ske ch o he amewo k p oposed in his wo k. The sys em
ea u es a modula design, wi h each ool equi ed o mee he speci ica ions ou lined in
Sec ion 3.1. Fi s ly, he elec ical moni o ing de ices a e ep esen ed. They can be ei he
PMUs using he IEEE C37.118.2 s anda d [
53
] o o he de ices such as PQMs o SMs using
gene ic communica ions p o ocols (Modbus, SMTP, MQTT, e c.). An agen is equi ed o
each p o ocol implemen ed in he sys em, as well as a PDC o collec he measu emen s
om he PMUs. Mul iple PDCs o agen s can be added as needed based on he equipmen ’s
dis ibu ion ac oss he g id.
Figu e 2. P oposed in as uc u e.
Nex in he da a low is he da abase. I was designed o be he co e componen o
he sys em a chi ec u e, di e ing om sys ems like he FIWARE pla o m whe e a da a
Appl. Sci. 2025,15, 170 7 o 27
dis ibu ion agen is cen al. This app oach is common in ools whe e da a s o age is
c i ical [
25
,
27
], helping o a oid unnecessa y da a s eams. Howe e , i equi es a da abase
ha can in eg a e wi h mul iple agen s and PDCs. Wi h nume ous da abase managemen
sys ems a ailable, i is c ucial o selec he one ha bes mee s ou needs. Rela ional
da abases (SQL) a e use ul o a oiding da a duplica ion bu a e slow wi h la ge da a
olumes, making hem unsui able o PMU measu emen s. In con as , non- ela ional
da abases (NoSQL) a e designed o handle la ge da a olumes e icien ly and o e as e
da a handling speeds. These da abases a e o en specialized o speci ic da a ypes and
op imized acco dingly. Among NoSQL ypes, ime se ies da abase manage s (TSDBs)
s and ou by s o ing da a as alue– imes amp pai s, making hem mo e e icien han o he
da abase ypes o ime se ies da a [54,55].
Finally, back o he amewo k equi emen s, his a chi ec u e mus be use - iendly.
In his ega d, i was conside ed essen ial o allow eal- ime access o he s o ed da a, bo h
o moni o ing and eal- ime analysis, allowing he inco po a ion o algo i hms ha eed
back he gene a ed da a. Likewise, i will be necessa y o be able o access he da a o line i
a subsequen analysis is equi ed. The so wa e used o his pu pose would depend on
he clien , bu he TSDB will need o be widely compa ible o be e pe o mance.
3.3. P oposed So wa e In eg a ion Tools and Al e na i es
Based on he ske ch d awn in Sec ion 3.2, a se ies o open-sou ce ools we e selec ed,
as shown in Figu e 3. Each o he indica ed ools could be eplaced by a simila one as long
as hey mee he equi emen s desc ibed in Sec ion 3.1. The ools used, along wi h simila
pe o mance al e na i es, a e desc ibed below.
Figu e 3. P oposed amewo k wi h each chosen so wa e ool.
In luxDB e sion 2.0 was chosen as he TSDB, which is an open-sou ce TSDB de-
eloped by he In luxDa a company. I o e s a comp ehensi e ecosys em, including he
Teleg a agen [
56
], which is capable o ecei ing da a om an eno mous numbe o p o o-
cols. Ano he widely used open-sou ce TSDB is MongoDB, al hough i does no ha e i s
own agen ha is as e sa ile as Teleg a , so i would be necessa y o eso o ano he ype
o middlewa e o in eg a e messages, such as Node-RED.In luxDB, which appea s o be
he mos sui able op ion due o i s e sa ili y [54].
Appl. Sci. 2025,15, 170 8 o 27
The chosen PDC is OpenPDC, which is an open-sou ce p og am widely used in he
li e a u e [
24
,
25
,
57
]. This da a concen a o has he capabili y o o wa d he in o ma-
ion ha eaches i o o he PDCs h ough he S eaming Teleme y T anspo P o ocol
(STTP IEEE 2664) [58].
This p o ocol, s ill unde de elopmen by he IEEE, has been op-
imized o exchange ime se ies da a. Due o he p o ocol’s no el y, i has no ye been
implemen ed in he TSDB. To add ess his, a middlewa e called STTP2HIDS, de eloped
by he G id P o ec ion Alliance (GPA), was used o con e he da a in o w i e eques s
o In luxDB. The e a e o he p op ie a y PDCs, such as he enhanced Phaso Da a Con-
cen a o (ePDC) de eloped by Elec ic Powe G oup (EPG), o he VCL-PDC by Valian
Communica ions. Ne e heless, OpenPDC is he only open-sou ce PDC ha has been
widely alida ed. All o hem ha e he same p oblem when ansmi ing da a o a gene ic
TSDB, so STTP2HIDS would ha e o be used anyway. In [
24
], OpenHis o ian, ano he GPA
p oduc , was used as he middlewa e; howe e , i is much mo e bulky and cumbe some
o his pu pose han STTP2HIDS.
Las ly, ha ing such a popula TSDB makes i e y easy o access he da a. Using
In luxDB, i is possible o download he da a in he CSV o ma o analyze hem ex e nally
o line wi h any so wa e, such as GNU Oc a e 9.3 o Py hon 3.11 , o i is possible o access
he da abase online wi h an API capable o pe o ming pe iodic que ies. In his case, a
Py hon API was used. I allows da a o be moni o ed and p ocessed in eal ime. Py hon’s
e sa ili y enables he use o well-known ools such as Tenso Flow 2 o Ke as 3 o pe o m
A i icial In elligence asks di ec ly on he da a. Likewise, i will be possible o moni o he
da a in eal ime wi h a moni o ing so wa e such as G a ana 11.4.
4. Labo a o y and Field Tes s
One o he main ad an ages o his pla o m is i s abili y o s o e and compa e mea-
su emen s o a ious ypes, enabling hem o complemen each o he . To his end, a se ies
o es s we e ca ied ou bo h in he labo a o y and on a dis ibu ion g id using DERs,
wi h bo h comme cial de ices and sel -designed equipmen . A G a ana dashboa d was
designed, al hough he da a we e analyzed o line by downloading hem in he CSV o ma .
4.1. Labo a o y Tes bed and Resul s
The designed so wa e was implemen ed and es ed in he labo a o y using ou uni s
o he mos demanding comme cial elec ical moni o ing de ices. Two Powe side PQMs,
speci ically he h ee-phase PQube3 model and wo
µ
PMUs o he same b and, we e used
as measu ing equipmen . The
µ
PMUs gene a e he measu emen s shown in Sec ion 2.2
o each phase, plus some powe measu emen s ha a e only displayed on he de ice
and a e no ansmi ed. The PQM, on he o he hand, gene a es mo e han 3000 alues
among powe , ene gy, ha monics, and o he PQ measu emen s. As desc ibed in Sec ion 3.1,
µ
PMU uses, as usual, he IEEE S d. C37.118.2. I can be con igu ed o he di e en exis ing
e sions, and in his case, he 2011 e sion was picked. In he case o PQMs, hey use
Modbus TCP/IP. The ou de ices we e connec ed simul aneously o he TSDB, s o ing
he da a hey gene a e in eal ime. Bo h he TSDB and he Teleg a agen we e loca ed on
one PC, while OpenPDC an on ano he PC, bo h wi hin he same local ne wo k. G a ana
moni o ing so wa e was also in he same compu e whe e he TSDB was hos ed. Figu e 4
shows how he equipmen was placed in he labo a o y, while Figu e 5shows he da a low.
Figu e 6shows a dashboa d in G a ana whe e he mos signi ican measu emen s
collec ed by he PQMs a e ep esen ed in eal ime. In he uppe -le co ne , he PQM o
which he da a a e o be displayed is selec ed, whe eas in he uppe - igh co ne , he ime
and da e a e selec ed. A simila dashboa d was c ea ed o he µPMUs.
Appl. Sci. 2025,15, 170 9 o 27
Figu e 4. Expe imen al es bed in he labo a o y.
Figu e 5. Expe imen al es da a low.
Figu e 6. Real- ime moni o ing o a PQM (PQube3) analyze wi h G a ana dashboa d.
In hese es s, he equipmen was con enien ly compa ed in pai s. Fi s , we compa ed
he equipmen o he same ype wi h each o he , ollowed by compa isons be ween he
Appl. Sci. 2025,15, 170 16 o 27
ou ca ego ies acco ding o he LED lamp connec ed o he line a any gi en ime: “None”,
“Lamp1”, “Lamp2”, o “Lamp3”.
As is usual in his ype o echnique, he da a mus be subjec ed o p ep ocessing
be o e being used. Speci ically, a ea u e selec ion was pe o med, ini ially excluding
ha monics ha we e oo low and going om 39 o 21 ha monics. In addi ion, a scaling o
he ype MaxAbsScale was applied so ha each ea u e would be scaled acco ding o i s
maximum absolu e alue, and he alue o he ea u e o aining would be in he ange
[
0.0,1.0
]
. Figu e 16, o illus a i e pu poses, shows he a e age no malized alues o he
21 ha monics conside ed.
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
Ha monic
0.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Vol age (V)
Mean Vol age Ha monics
Lamp1 Lamp2 Lamp3
Figu e 15. A e age ol age ha monics (2nd o 40 h) on each o he h ee lamps.
The da a we e hen app op ia ely di ided in o a aining subse and a es subse
using he ain_ es _spli unc ion om he Sciki _lea n lib a y, which andomly selec s
alues o ensu e a balanced and co ec dis ibu ion o da a om each o he ca ego ies
in bo h subse s. P opo ions o 80% o aining da a and 20% o alida ion we e used.
This esul ed in a subse o scaled aining da a, wi h app oxima ely 1600 en ies o each
ca ego y, and a es subse , wi h app oxima ely 400 alues o each class.
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40
Ha monic
0.0
0.2
0.4
0.6
0.8
1.0
No malized
ol age
Mean No malized Vol age Ha monics
Lamp1 Lamp2 Lamp3
Figu e 16. No malized alues o he conside ed a e age ol age ha monics on each o he h ee lamps.
The i s classi ie implemen ed was a Neu al Ne wo k. Al hough he e is a wide
a ie y o ad anced neu al ne wo ks sui able o hese asks (including RBF, CNN, BRNN,
Appl. Sci. 2025,15, 170 17 o 27
KNN, DNN, GMDH, e c.) [
17
], o he sake o implemen a ion and compa ison wi h
he o he wo me hods, he classic Mul i-laye Pe cep on classi ie s uc u e a ailable in
Sciki -lea n has been used, which allows o supe ised lea ning by op imizing he log
loss unc ion. In his case, he inpu laye will be o med by 21 nodes and he ou pu
laye by 4 nodes (one o each o he conside ed ca ego ies). Con igu ing he NN equi es
op imizing he c oss- uning o a signi ican numbe o hype pa ame e s, which in luence
he accu acy o he ne wo k, as can be seen in Figu e 17. To assis in his ask, Sciki -
lea n implemen s he G idSea chCV unc ion, which pe o ms an exhaus i e sea ch o e
speci ied pa ame e alues op imized by c oss- alida ed g id sea ch o e a pa ame e
g id. In his way, op imized hype pa ame e s we e es ablished wi h a con igu a ion
o wo in e nal laye s o 5 and 10 neu ons, espec i ely, as well as h ough he use o
a s ochas ic g adien descen (SGD) op imize , wi h an adap i e lea ning a e and an
ac i a ion unc ion o he hidden laye o he hype bolic an ype:
(x) = ex−e−x
ex+e−x(2)
(a) Hype bolic an ac i a ion unc ion. (b) Sigmoid ac i a ion unc ion.
Figu e 17. Numbe o neu ons in he wo hidden laye s s. accu acy wi h di e en ac i a-
ion unc ions.
As a esul o aining his model, a p edic o wi h 100% accu acy was ob ained in
bo h he aining and alida ion subse s. Figu e 18 shows he e olu ion o he loss unc ion
du ing aining and he con usion ma ix ob ained wi h he es da a subse .
In gene al, classi ie s based on neu al ne wo ks end o exhibi s ong gene aliza ion
capabili ies. Howe e , some imes, hei black-box na u e makes i challenging o ex ac
in o ma ion abou he c i e ia used o classi y an inpu alue in o speci ic ca ego ies. In
addi ion, he compu a ion ime can be signi ican i a la ge numbe o hidden laye s and
neu ons a e used. In such si ua ions, i may be app op ia e o use a non-pa ame ic classi ie
such as a Decision T ee.
Appl. Sci. 2025,15, 170 18 o 27
0 50 100 150 200 250 300
I e a ions
1.0
1.2
1.4
1.6
1.8
2.0
2.2
2.4
Cos
Loss Cu e
(a) Loss cu e
None
Lamp 1
Lamp 2
Lamp 3
P edic ed
None
Lamp 1
Lamp 2
Lamp 3
Obse ed
414 0 0 0
0400 0 0
0 0 417 0
0 0 0 403
Con usion Ma ix NN
0
50
100
150
200
250
300
350
400
(b) Con usion ma ix o NN on es da a.
Figu e 18. Mul i-Laye Pe cep on aining and es esul s.
The goal o he DT is o cons uc a bina y ee s uc u e ha e icien ly classi ies
inpu alues h ough a p og essi e ea u es e alua ion p ocess un il eaching a e minal
node (lea ) ha co esponds o he ca ego y associa ed o he inpu alue. In his case, he
inpu alue would be he eco ded ha monics. The condi ions a e es ablished based on he
alues o he ha monics, and he esul should be a lea co esponding o each o he ou
ca ego ies. DT aining consis s o op imizing he decision ules a each node, based on
inpu da a, o achie e be e di isions. This p ocess o en u ilizes me ics such as he Gini
Index o guide he op imiza ion:
Gini =1−
n
∑
i=1
P2
i(3)
whe e
Pi
is he p obabili y ha an elemen is classi ied in a gi en ca ego y. Thus, he Gini
Index, which anges om 0 o 1, de e mines he impu i y o a se , indica ing how mixed o
he e ogeneous he se is.
The e o e, using he Sciki -lea n lib a y, a DT was ained o selec he bes spli a
each node, using he Gini Index, and wi h a maximum dep h o 3 o educe compu a ional
ime du ing e alua ion. The DT ob ained achie ed 100% accu acy o bo h he aining and
es subse s. Figu e 19 shows he s uc u e o he Decision T ee and he con usion ma ix
when e alua ing he es da ase .
As can be seen in Figu e 19a he ee managed o es ablish ou lea es wi h pu i y
(
Gini =
0) co esponding o each o he ou ca ego ies using a classi ica ion me hod based
on a p og essi e bina y di ision o he scaled alues o ha monics H29, H5, and H27.
The e ec i eness o his classi ica ion app oach becomes e iden when examining he
dis ibu ion o hese ha monic alues wi hin he a ailable da ase , as shown in Figu e 20.
This esul s in a classi ie ha enables apid classi ica ion, hough i may be mo e
sensi i e o he alues o speci ic ha monics and hus po en ially less obus compa ed o
he NN app oach.
To mi iga e his issue, i is possible o employ a classi ie ha u ilizes a se o DTs,
known as Random Fo es , which combines he esul s om all o hese ees. Al hough
i is possible o use a la ge numbe o ees, in his case, keeping in mind he eal- ime
implemen a ion, only ou ees we e calcula ed. Each o hem was ained on a subse o
he aining da a o acili a e he sea ch o di e se classi ica ions. As a esul , an RF wi h an
accu acy o 100% on he aining and es subse s was ob ained, which was composed o
he ou DTs shown in Figu e 21. As can be obse ed, he decisions we e made based on
he alues o nine ha monics (H3, H5, H9, H16, H23, H27, H33, H34, and H36), he eby
inc easing obus ness wi h espec o he DT.
Appl. Sci. 2025,15, 170 19 o 27
The pe o mance o hese h ee classi ie s applied in eal ime will be shown in he
nex sec ion.
gini = 0.0
samples = 1639
alue = [0, 0, 0, 16 39]
class = Lamp 3
gini = 0.0
samples = 1625
alue = [0, 0, 1625 , 0]
class = Lamp 2
gini = 0.0
samples = 1629
alue = [1629, 0, 0 , 0]
class = None
H27 <= 0.217
gini = 0.5
samples = 3264
alue = [0.0, 0.0, 1 625.0, 1639.0]
class = Lamp 3
H5 <= 0.332
gini = 0.667
samples = 4893
alue = [1629, 0, 1 625, 1639]
class = Lamp 3
T ue
gini = 0.0
samples = 1642
alue = [0, 1642, 0 , 0]
class = Lamp 1
False
H29 <= 0.591
gini = 0.75
samples = 6535
alue = [1629, 164 2, 1625, 1639]
class = Lamp 1
(a) Decision T ee.
None
Lamp 1
Lamp 2
Lamp 3
P edic ed
None
Lamp 1
Lamp 2
Lamp 3
Obse ed
414 0 0 0
0400 0 0
0 0 417 0
0 0 0 403
Con usion Ma ix DT
0
50
100
150
200
250
300
350
400
(b) Con usion ma ix o DT on es da a.
Figu e 19. Decision T ee s uc u e and es esul s.
Figu e 20. Ha monic alues dis ibu ions o each class (None, Lamp1, Lamp2, Lamp3) in ain-
ing da ase .
Appl. Sci. 2025,15, 170 20 o 27
gini = 0.0
samples = 389
alue = [449, 0, 0, 0]
class = None
gini = 0.004
samples = 456
alue = [0, 0, 1, 527]
class = Lamp 3
H3 <= 0.249
gini = 0.498
samples = 845
alue = [449, 0, 1, 527]
class = Lamp 3
gini = 0.0
samples = 445
alue = [0, 0, 506, 0]
class = Lamp 2
H23 <= 0.61
gini = 0.665
samples = 1290
alue = [449, 0, 507, 527]
class = Lamp 3
T ue
gini = 0.0
samples = 450
alue = [0, 517, 0, 0]
class = Lamp 1
False
H27 <= 0.56
gini = 0.749
samples = 1740
alue = [449, 517, 507, 527]
class = Lamp 3
gini = 0.0
samples = 434
alue = [0, 0, 519, 0]
class = Lamp 2
gini = 0.0
samples = 422
alue = [0, 478, 0, 0]
class = Lamp 1
gini = 0.0
samples = 437
alue = [0, 0, 0, 502]
class = Lamp 3
H27 <= 0.534
gini = 0.499
samples = 856
alue = [0, 478, 519, 0]
class = Lamp 2
gini = 0.0
samples = 432
alue = [501, 0, 0, 0]
class = None
T ue
H27 <= 0.213
gini = 0.666
samples = 1293
alue = [0.0, 478.0, 519.0, 502.0]
class = Lamp 2
False
H5 <= 0.242
gini = 0.75
samples = 1725
alue = [501, 478, 519, 502]
class = Lamp 2
gini = 0.0
samples = 414
alue = [486, 0, 0, 0]
class = None
gini = 0.0
samples = 413
alue = [0, 0, 0, 483]
class = Lamp 3
gini = 0.0
samples = 452
alue = [0, 526, 0, 0]
class = Lamp 1
gini = 0.0
samples = 435
alue = [0, 0, 505, 0]
class = Lamp 2
H9 <= 0.038
gini = 0.5
samples = 827
alue = [486, 0, 0, 483]
class = None
T ue
H5 <= 0.769
gini = 0.5
samples = 887
alue = [0, 526, 505, 0]
class = Lamp 1
False
H33 <= 0.391
gini = 0.75
samples = 1714
alue = [486, 526, 505, 483]
class = Lamp 1
gini = 0.0
samples = 415
alue = [0, 0, 0, 480]
class = Lamp 3
gini = 0.375
samples = 15
alue = [0, 0, 4, 12]
class = Lamp 3
gini = 0.268
samples = 431
alue = [0, 80, 423, 0]
class = Lamp 2
gini = 0.259
samples = 451
alue = [0, 440, 78, 1]
class = Lamp 1
H36 <= 0.214
gini = 0.016
samples = 430
alue = [0, 0, 4, 492]
class = Lamp 3
H34 <= 0.276
gini = 0.501
samples = 882
alue = [0, 520, 501, 1]
class = Lamp 1
gini = 0.0
samples = 423
alue = [482, 0, 0, 0]
class = None
T ue
H16 <= 0.413
gini = 0.667
samples = 1312
alue = [0, 520, 505, 493]
class = Lamp 1
False
H9 <= 0.046
gini = 0.75
samples = 1735
alue = [482, 520, 505, 493]
class = Lamp 1
Figu e 21. Random Fo es composed o 4 DTs.
5.2. Online Expe imen al Resul s
To alida e he pe o mance o he classi ie s and o achie e mo e accu a e compa -
isons, he PQM’s capabili y o eco d in o ma ion om h ee phases simul aneously was
aken ad an age o . Consequen ly, one lamp was connec ed o each o he h ee phases.
This allowed o he h ee classi ie s (p e iously ained) o be e alua ed simul aneously
on he h ee lamps wi h he same on and o imes. To accomplish his, s a ing om a
s a e wi h lamps o , hey we e u ned on and o mul iple imes, wi h a ying du a ions as
illus a ed in Figu e 22, o a pe iod o app oxima ely 55 min.
Figu e 22. Expe imen al esul s: Lamp connec ed a each phase s. p edic ions.
Appl. Sci. 2025,15, 170 21 o 27
The da a we e sampled and e alua ed e e y second by applying he h ee classi ie s
o each o he phases. As can be seen in Figu e 22, bo h he NN and he RF p edic ed sa is-
ac o ily mos o he cases. Only in wo ins ances du ing lamp 3 igni ion was i es ima ed as
lamp 2, possibly due o a phenomenon o In ush Cu en [
60
]. In con as , he DT exhibi ed
addi ional e o s in he ini ial swi ch-on and swi ch-o o lamp 1, es ima ing i as lamp 2,
possibly due o i s lowe obus ness o In ush Cu en o empe a u e phenomena [61].
Figu e 23 shows he con usion ma ix o each algo i hm on he expe imen al esul s.
Table 5shows in de ail he pe o mance esul s ob ained du ing he expe imen wi h each
o he models. The s anda d me ics we e calcula ed based on he numbe o T ue Posi i es
(TPs), False Posi i es (FPs), False Nega i es (FNs) and T ue Nega i es (TNs), which a e
de ined as
Accu acy =TP +TN
TP +FP +TN +FN (4)
P ecision =TP
TP +FP (5)
Recall =TP
TP +FN (6)
F1-Sco e =2×P ecision ×Recall
P ecision +Recall (7)
MCC =(TP ×TN)−(FP ×FN)
p(TP +FP)×(TP +FN)×(TN +FP)×(TN +FN)(8)
Addi ionally, o unde s and he esul s o he DT, i is use ul o compa e he ha monics
measu ed du ing he eal- ime es wi h hose a ailable in he lea ning phase. As shown
in Figu e 24, he e a e di e ences be ween he ha monics used in he lea ning phase (see
Figu e 20) and hose measu ed du ing he expe imen al phase. The DT p o ed o be mo e
sensi i e o hese di e ences, while bo h he NN and he RF demons a ed hei capaci y
o gene aliza ion. Fu he mo e, he ol age ha monics o lamp 3 co esponding o he only
wo ins ances in which he NN and RT ailed can be clea ly seen as ou lie s in Figu e 24.
As can be seen, bo h he NN and he RF ob ained he same esul , wi h p ac ically no
ailu es. Howe e , he NN equi ed a compu a ion ime o 5.5 imes ha o he DT, while
he RF equi ed only 1.67 imes ha ime.
None
Lamp 1
Lamp 2
Lamp 3
P edic ed
None
Lamp 1
Lamp 2
Lamp 3
Obse ed
6825 000
0 1071 0 0
0 0 1071 0
0 0 2 1069
Con usion Ma ix NN (Expe imen )
0
1000
2000
3000
4000
5000
6000
(a) Neu al Ne wo k.
None
Lamp
1
Lamp 2
Lamp 3
P edic ed
None
Lamp 1
Lamp 2
Lamp 3
Obse ed
6825 000
0 956 115 0
0 0 1071 0
0 0 2 1069
Con usion Ma ix DT (Expe imen )
0
1000
2000
3000
4000
5000
6000
(b) Decision T ee.
None
Lamp
1
Lamp 2
Lamp 3
P edic ed
None
Lamp 1
Lamp 2
Lamp 3
Obse ed
6825 000
0 1071 0 0
0 0 1071 0
0 0 2 1069
Con usion Ma ix RF (Expe imen )
0
1000
2000
3000
4000
5000
6000
(c) Random Fo es .
Figu e 23. Con usion ma ix o he expe imen al esul s.
Appl. Sci. 2025,15, 170 22 o 27
Table 5. Me ics o he expe imen al esul s.
Class Model Suppo TP FN FP TN Accu acy P ecision Recall F1-Sco e MCC
NN 6825 6825 0 0 3213 1.00000 1.00000 1.00000 1.00000 1.00000
None DT 6825 6825 0 0 3213 1.00000 1.00000 1.00000 1.00000 1.00000
RF 6825 6825 0 0 3213 1.00000 1.00000 1.00000 1.00000 1.00000
NN 1071 1071 0 0 8967 1.00000 1.00000 1.00000 1.00000 1.00000
Lamp1 DT 1071 956 115 0 8967 0.98854 1.00000 0.89262 0.94327 0.93879
RF 1071 1071 0 0 8967 1.00000 1.00000 1.00000 1.00000 1.00000
NN 1071 1071 0 2 8965 0.99980 0.99814 1.00000 0.99907 0.99896
Lamp2 DT 1071 1071 0 117 8850 0.98834 0.90152 1.00000 0.94821 0.94327
RF 1071 1071 0 2 8965 0.99980 0.99814 1.00000 0.99907 0.99896
NN 1071 1069 2 0 8967 0.99980 1.00000 0.99813 0.99907 0.99895
Lamp3 DT 1071 1069 2 0 8967 0.99980 1.00000 0.99813 0.99907 0.99895
RF 1071 1069 2 0 8967 0.99980 1.00000 0.99813 0.99907 0.99895
Figu e 24. Ha monic alues dis ibu ions in aining se (None, Lamp1, Lamp2, Lamp3) s. eal ime
expe imen (NoneRT, Lamp1RT, Lamp2RT, Lamp3RT).
6. Conclusions
The main no el y o his pape is i s designed amewo k capable o in eg a ing elec-
ical moni o ing de ices, such as PMUs, PQMs, and SMs, as well as adding AI capabili ies
o he eco ded da a. This amewo k co e s a necessi y gi en he la ge numbe o appli-
ca ions based on elec ical moni o ing da a and he g ow h o DERs in dis ibu ion g ids,
which inc ease he necessi y o moni o ing and o ecas ing in hese g ids. In addi ion,
only open-sou ce so wa e has been used, which is an ad an age in e ms o anspa ency,
lexibili y, and cus omiza ion. Mo eo e , he amewo k has been p o en o be capable o
execu ing eal- ime machine lea ning models, which is some hing no usually discussed in
he li e a u e o simila amewo ks. This capabili y allows he amewo k o p ocess and
analyze elec ical moni o ing da a mo e e ec i ely, imp o ing applica ions such as aul
diagnosis, s a e es ima ion, and e en de ec ion. In gene al, he chosen scheme was es ed
wi h ema kable success, bo h in he labo a o y and in a dis ibu ion g id.
Appl. Sci. 2025,15, 170 23 o 27
In eg a ion capabili ies we e es ed in bo h labo a o y and ield es s. In he labo a o y,
i was obse ed ha measu emen s we e synch onized among de ices o he same ype,
bu he e was a no iceable delay o abou one second be ween di e en de ices. This is
likely caused by he di e en synch oniza ion me hods used: PMUs a e synch onized ia a
GPS, while PQMs use an NTP. In he ield es , a mo e complex scena io was c ea ed using
con igu able Na ional Ins umen s equipmen . All equipmen was connec ed ia TSN,
enabling he PQMs o use he same GPS signal as he PMUs. In his es , he measu emen s
we e s ongly synch onized, and he delay was less han one second. Ongoing e o s a e
ocused on u he cha ac e izing his ield en i onmen . In conclusion, pe o mance may
a y depending on he measu emen equipmen used, al hough he amewo k adap s o
e en he mos demanding en i onmen s. Due o he obus ness o he so wa e, no da a
loss was obse ed.
Once da a we e p ope ly egis e ed, machine lea ning echniques we e in eg a ed in a
eal- ime load iden i ica ion example. Neu al Ne wo k, Decision T ee, and Random Fo es
classi ie s o eal- ime load iden i ica ion we e implemen ed and expe imen ally alida ed
o iden i y he connec ed LED om he ol age ha monics eco ded by a PQM. O he h ee
op ions, he Decision T ee was ound o be he mos sensi i e o da a a iabili y, while bo h
he Neu al Ne wo k and he Random Fo es had nea 100% accu acy. In all h ee cases,
eal- ime execu ions we e pe o med wi hou isible delay o da a conges ion. E en he
Neu al Ne wo k, which equi ed mo e compu ing ime, was execu ed seamlessly. This
gua an ees he in eg a ion o machine lea ning echniques wi h he amewo k, al hough
he capaci y o execu e mo e complex classi ie s would depend on o he ex e nal ac o s,
such as he p ocessing powe .
Wi h bo h in eg a ion and load iden i ica ion es s, he capabili ies o he amewo k
we e es ed. Looking ahead, he amewo k has he po en ial o suppo a wide ange o
applica ions in ol ing elec ical de ices and AI, including moni o ing and con ol, s a e
es ima ion, e en de ec ion, and aul loca ion. Al hough ce ain so wa e ools, such as he
selec ed TSDB In luxDB, could be subs i u ed wi h simila al e na i es, he ools u ilized in
his s udy deli e ed ou s anding pe o mance and no able success.
Au ho Con ibu ions: Concep ualiza ion, V.A.-R. and F.C.; me hodology, V.A.-R. and F.C.; so wa e,
V.A.-R. and F.C.; alida ion, V.A.-R., F.C. and V.P.-L.; o mal analysis, F.C.; in es iga ion, V.A.-R. and
F.C.; esou ces, V.A.-R., F.C., V.P.-L., and I.S.; da a cu a ion, F.C.; w i ing—o iginal d a p epa a ion,
V.A.-R. and F.C.; w i ing— e iew and edi ing, V.A.-R., F.C., and I.S.; isualiza ion, V.A.-R. and F.C.;
unding acquisi ion, F.C. and V.P.-L. All au ho s ha e ead and ag eed o he published e sion o
he manusc ip .
Funding: This wo k has been suppo ed by he Agencia Es a al de In es igación (AEI)-Spain, unde
g an PID2019-109071RB-I00, and he Spanish esea ch subp ojec Moni o ing And In eg a ion o
ene gy da a wi h Seamless Tempo al Accu acy o pho o ol aic plan s (MISTA) o he coo dina ed
p ojec S a egies o Agg ega ed Gene a ion o Pho oVol aic plan s (SAGPV). Re . PID2019-108953RA-
C22 and PID2019-108953RB-C21.
Da a A ailabili y S a emen : The o iginal da a p esen ed in he s udy a e openly a ailable in he
Hel ia eposi o y a h p://hdl.handle.ne /10396/30154 (accessed on Decembe 2024 ).
Acknowledgmen s: We gi e ou g a i ude o Sola del Valle, owne o he C uz del Doc o and
Cabeza Oli a PV plan s, o hei suppo in he de elopmen o his esea ch.
Con lic s o In e es : The au ho s decla e no con lic s o in e es . The unde s had no ole in he design
o he s udy; in he collec ion, analyses, o in e p e a ion o da a; in he w i ing o he manusc ip ; o
in he decision o publish he esul s.
Appl. Sci. 2025,15, 170 24 o 27
Abb e ia ions
The ollowing abb e ia ions a e used in his manusc ip :
AI A i icial In elligence
BRNN Bidi ec ional Recu en Neu al Ne wo k
CNN Con olu ional Neu al Ne wo k
CT Cu en T ans o me
DER Dis ibu ed Ene gy Resou ce
DNN Deep Neu al Ne wo k
DT Decision T ee
FPS F ames pe Second
GMDH G oup Me hod o Da a Handling
GPS Global Posi ioning Sys em
KNN K-Nea es Neighbo
LAN Local A ea Ne wo k
NILM Non-In usi e Load Moni o ing
NN Neu al Ne wo k
NoSQL Non-Rela ional Da abases
NTP Ne wo k Time P o ocol
PDC Phaso Da a Concen a o
PMU Phaso Measu emen Uni
µPMU Mic o-Phaso Measu emen Uni
PQ Powe Quali y
PQA Powe Quali y Analyze
PQM Powe Quali y Moni o
PT Powe T ans o me
PV Pho o-Vol aic
RF Random Fo es
RBF Radial Basis Func ion
ROCOF Ra e O Change O F equency
SGD S ochas ic G adien Descen
SM Sma Me e
SQL S uc u ed Que y Language
STTP S eaming Teleme y T anspo P o ocol
TC T ans o ma ion Cen e
THD To al Ha monic Dis o ion
TSDB Time Se ies Da abase
TSN Time-Sensi i e Ne wo kig
TVE To al Vec o E o
WAN Wide A ea Ne wo k
Re e ences
1.
Saldaña-González, A.E.; Sumpe , A.; A agüés-Peñalba, M.; Smolnika , M. Ad anced Dis ibu ion Measu emen Technologies and
Da a Applica ions o Sma G ids: A Re iew. Ene gies 2020,13, 3730. [C ossRe ]
2.
Leinakse, M.; Tani, P.; Kil e , J. Impac o Dis ibu ed Gene a ion on Es ima ion o Exponen ial Load Models. IEEE Powe Ene gy
Soc. Gen. Mee . 2019,2019, 8974014. [C ossRe ]
3.
Yu, J.; Sun, Y.; Lu, J.; Zou, Q.; Xie, X.; Fu, B.; Liu, W.; Ye, L.; Chen, Y. Op imal placemen o powe quali y moni o s in complex
dis ibu ion ne wo k based on opological deg ada ion. In P oceedings o he 14 h IEEE Con e ence on Indus ial Elec onics and
Applica ions, ICIEA 2019, Xi’an, China, 19–21 June 2019; pp. 58–61. [C ossRe ]
4.
Kim, J.; Ba an, M.E.; Lampley, G.C. Es ima ion o aul loca ion on dis ibu ion eede s using PQ moni o ing da a. In P oceedings
o he 2007 IEEE Powe Enginee ing Socie y Gene al Mee ing, PES, Tampa, FL, USA, 24–28 June 2007. [C ossRe ]
5.
Sel a aj, G.; Thangaiyan, V.; Rajangam, K. Nume ical me hod o islanding he loca ion o g ound aul in he ma e ial based
dis ibu ion sys em. Ma e . Today P oc. 2021,45, 634–639. [C ossRe ]
Appl. Sci. 2025,15, 170 25 o 27
6.
Lee, J. Au oma ic Faul Loca ion on Dis ibu ion Ne wo ks Using Synch onized Vol age Phaso Measu emen Uni s. In
P oceedings o he Ame ican Socie y o Mechanical Enginee s, Powe Di ision (Publica ion) POWER, Bal imo e, MD, USA, 28–31
July 2014. [C ossRe ]
7.
Cui, Q.; Weng, Y. Enhance High Impedance Faul De ec ion and Loca ion Accu acy ia uPMUs. IEEE T ans. Sma G id
2020,11, 797–809. [C ossRe ]
8.
Yabloko , A.; I ano , I.; Kuliko , F.; Tychkin, A.; Panascha enko, A.; Umno , Y. In es iga ion o aul loca ion based on class M
PMUs wi h simula ed and eal- ield synch ophaso s. Elec . Powe Sys . Res. 2024,232, 110414. [C ossRe ]
9.
Chen, X.; Tseng, K.J.; Ama a unga, G. S a e es ima ion o dis ibu ion sys ems using mic o-synch ophaso s. In P oceedings o
he Asia-Paci ic Powe and Ene gy Enginee ing Con e ence, APPEEC, Nanjing, China, 25–27 Oc obe 2016. [C ossRe ]
10.
Fa ajollahi, M.; Shahsa a i, A.; Mohsenian-Rad, H. T acking S a e Es ima ion in Dis ibu ion Ne wo ks Using Dis ibu ion-
le el Synch ophaso Da a. In P oceedings o he IEEE Powe and Ene gy Socie y Gene al Mee ing, Po land, OR, USA, 5–10
Augus 2018. [C ossRe ]
11.
Dusabimana, E.; Yoon, S.G. A Su ey on he Mic o-Phaso Measu emen Uni in Dis ibu ion Ne wo ks. Elec onics 2020,9, 305.
[C ossRe ]
12.
Cheng, G.; Lin, Y.; Abu , A.; Gomez-Exposi o, A.; Wu, W. A Su ey o Powe Sys em S a e Es ima ion Using Mul iple Da a
Sou ces: PMUs, SCADA, AMI, and Beyond. IEEE T ans. Sma G id 2024,15, 1129–1151. [C ossRe ]
13.
A ale, G.; Ca alio i, A.; Cosen ino, V.; Ca a, D.D.; Guaiana, S.; Panza ecchia, N.; Tine, G. Real-Time Powe Flow Moni o ing and
Con ol Sys em o Mic og ids In eg a ion in Islanded Scena ios. IEEE T ans. Ind. Appl. 2019,55, 7186–7197. [C ossRe ]
14.
Meegahapola, L.G.; Bu, S.; Wadduwage, D.P.; Chung, C.Y.; Yu, X. Re iew on Oscilla o y S abili y in Powe G ids wi h Renewable
Ene gy Sou ces: Moni o ing, Analysis, and Con ol Using Synch ophaso Technology. IEEE T ans. Ind. Elec on. 2021,68, 519–531.
[C ossRe ]
15.
Ja amillo, A.F.M.; La e y, D.M.; Mo ow, D.J.; del Rincon, J.M.; Foley, A.M. Load modelling and non-in usi e load moni o ing
o in eg a e dis ibu ed ene gy esou ces in low and medium ol age ne wo ks. Renew. Ene gy 2021,179, 445–466. [C ossRe ]
16.
Ghosh, S.; Cha e jee, D. Non-in usi e iden i ica ion o ha monic pollu ing loads in a sma esiden ial sys em. Sus ain. Ene gy,
G ids Ne w. 2021,26, 100446. : 10.1016/j.segan.2021.100446 [C ossRe ]
17.
Tanoni, G.; P incipi, E.; Squa ini, S. Non-In usi e Load Moni o ing in indus ial se ings: A sys ema ic e iew. Renew. Sus ain.
Ene gy Re . 2024,202, 114703. : 10.1016/j. se .2024.114703 [C ossRe ]
18. S ock, S.; Babazadeh, D.; Becke , C. Applica ions o A i icial In elligence in Dis ibu ion Powe Sys em Ope a ion. IEEE Access
2021,9, 150098–150119. [C ossRe ]
19.
Alimi, O.A.; Ouahada, K.; Abu-Mah ouz, A.M. A Re iew o Machine Lea ning App oaches o Powe Sys em Secu i y and
S abili y. IEEE Access 2020,8, 113512–113531. [C ossRe ]
20.
Lal, M.D.; Va ada ajan, R. A Re iew o Machine Lea ning App oaches in Synch ophaso Technology. IEEE Access
2023,11, 33520–33541. [C ossRe ]
21.
Vaish, R.; Dwi edi, U.; Tewa i, S.; T ipa hi, S. Machine lea ning applica ions in powe sys em aul diagnosis: Resea ch
ad ancemen s and pe spec i es. Eng. Appl. A i . In ell. 2021,106, 104504. : 10.1016/j.engappai.2021.104504 [C ossRe ]
22.
Chen, T.; Qin, H.; Li, X.; Wan, W.; Yan, W. A Non-In usi e Load Moni o ing Me hod Based on Fea u e Fusion and SE-ResNe .
Elec onics 2023,12, 1909. [C ossRe ]
23.
Cons andache, N.; S anescu, D.M.; Sanduleac, M.; S anescu, C.; T is iu, I.; Mandis, A. Sma me e s, PMU and PQ da a analysis in
ac i e dis ibu ion g ids—Case s udies in MV ne wo ks. In P oceedings o he 2018 In e na ional Con e ence on Applied and
Theo e ical Elec ici y, ICATE, C aio a, Romania, 4–6 Oc obe 2018. [C ossRe ]
24.
Cas ello, P.; Muscas, C.; Pego a o, P.A.; Sulis, S.; Giannuzzi, G.M.; Pede, M.; Maiolini, C.; Pau, P.; Bassi, F.; Coluzzi, C. In eg a ion
o powe quali y and aul da a in o a PMU-based Wide A ea Moni o ing Sys em. In P oceedings o he AMPS 2021—2021 11 h
IEEE In e na ional Wo kshop on Applied Measu emen s o Powe Sys ems, Vi ual Con e ence, 29 Sep embe –1 Oc obe 2021.
[C ossRe ]
25.
Ande sen, M.P.; Kuma , S.; B ooks, C.; Meie , A.V.; Culle , D.E. DISTIL: Design and implemen a ion o a scalable synch ophaso
da a p ocessing sys em. In P oceedings o he 2015 IEEE In e na ional Con e ence on Sma G id Communica ions (Sma G idComm),
Miami, FL, USA, 2–5 No embe 2016; Ins i u e o Elec ical and Elec onics Enginee s Inc.: Pisca away, NJ, USA, 2016; pp. 271–277.
[C ossRe ]
26.
Kosen, I.; Huang, C.; Chen, Z.; Zhang, X.; Min, L.; Zhou, D.; Zhu, L.; Liu, Y. UPS: Uni ied PMU-Da a s o age sys em o enhance
T+D PMU da a usabili y. IEEE T ans. Sma G id 2020,11, 739–748. [C ossRe ]
27.
Sodin, D.; Rudež, U.; Mihelin, M.; Smolnika , M.; And ejˇcampa, A.A.; Shake , R.; Dash i, R. Ad anced Edge-Cloud Compu ing
F amewo k o Au oma ed PMU-Based Faul Localiza ion in Dis ibu ion Ne wo ks. Appl. Sci. 2021,11, 3100. [C ossRe ]
28.
B anco, H.M.; Olesko icz, M.; Cou y, D.V.; Delbem, A.C. Mul iobjec i e op imiza ion o powe quali y moni o ing alloca ion
conside ing ol age sags in dis ibu ion sys ems. In . J. Elec . Powe Ene gy Sys . 2018,97, 1–10. [C ossRe ]
