Academic Edi o : Jose Luis
Cal o-Rolle
Recei ed: 13 Janua y 2025
Re ised: 31 Janua y 2025
Accep ed: 4 Feb ua y 2025
Published: 6 Feb ua y 2025
Ci a ion: Syl es in, G.R.; Maciel,
J.N.; Amo im, M.L.M.; Ca mo, J.P.;
A onso, J.A.; Lopes, S.F.; Ando Junio ,
O.H. S a e o he A in Elec ic
Ba e ies’ S a e-o -Heal h (SoH)
Es ima ion wi h Machine Lea ning: A
Re iew. Ene gies 2025,18, 746.
h ps://doi.o g/10.3390/
en18030746
Copy igh : © 2025 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/).
Re iew
S a e o he A in Elec ic Ba e ies’ S a e-o -Heal h (SoH)
Es ima ion wi h Machine Lea ning: A Re iew
Gio ane Ronei Syl es in 1,2, Joylan Nunes Maciel 1,2 , Ma cio Luís Munhoz Amo im 3, João Paulo Ca mo 3,
José A. A onso 4,* , Sé gio F. Lopes 5and Oswaldo Hideo Ando Junio 1,2,6,*
1In e disciplina y Pos g adua e P og am in Ene gy & Sus ainabili y (PPGIES), Fede al Uni e si y o La in
Ame ican In eg a ion—UNILA, Pa aná Ci y 85867-000, PR, B azil;
[email p o ec ed] (G.R.S.); [email p o ec ed] (J.N.M.)
2Resea ch G oup on Ene gy & Ene gy Sus ainabili y (GPEnSE), Academic Uni o Cabo de San o
Agos inho (UACSA), Fede al Ru al Uni e si y o Pe nambuco (UFRPE),
Cabo de San o Agos inho 54518-430, PE, B azil
3G oup o Me ama e ials Mic owa es and Op ics (GMe a), Depa men o Elec ical Enginee ing (SEL),
Uni e si y o São Paulo (USP), A enida T abalhado São-Ca lense, N . 400, Pa que Indus ial A nold Schmid ,
São Ca los 13566-590, SP, B azil; [email p o ec ed] (M.L.M.A.); [email p o ec ed] (J.P.C.)
4Cen e o Mic oelec omechanical Sys ems (CMEMS), Uni e si y o Minho, 4800-058 Guima ães, Po ugal
5Cen o Algo i mi/LASI, Uni e si y o Minho, 4704-553 Guima ães, Po ugal; se [email p o ec ed]
6
Sma G id Labo a o y (LabREI), Cen e o Al e na i e and Renewable Resea ch (CEAR), Fede al Uni e si y
o Pa aiba (UFPB), João Pessoa 58051-900, PB, B azil
*Co espondence: [email p o ec ed] (J.A.A.); [email p o ec ed] (O.H.A.J.)
Abs ac : The sus ainable euse o ba e ies a e hei i s li e in elec ic ehicles equi es
accu a e s a e-o -heal h (SoH) es ima ion o ensu e sa e and e icien epu posing. This
s udy applies he sys ema ic P oKnow-C me hodology o analyze he s a e o he a in
SoH es ima ion using machine lea ning (ML). A bibliog aphic po olio o 534 pape s
( om 2018 onwa d) was cons uc ed, e ealing key esea ch ends. Public da ase s a e
inc easingly a o ed, appea ing in 60% o he s udies and eaching 76% in 2023. Among
12 iden i ied sou ces co e ing 20 da ase s om di e en li hium ba e y echnologies,
NASA’s P ognos ics Cen e o Excellence con ibu es 51% o hem. Deep lea ning (DL)
domina es he ield, comp ising 57.5% o he implemen a ions, wi h LSTM ne wo ks used
in 22% o he cases. This s udy also explo es hyb id models and he eme ging ole o
ans e lea ning (TL) in imp o ing SoH p edic ion accu acy. This s udy also highligh s he
po en ial applica ions o SoH p edic ions in ene gy in o ma ics and sma sys ems, such as
sma g ids and In e ne -o -Things (IoT) de ices. By in eg a ing accu a e SoH es ima es
in o eal- ime moni o ing sys ems and wi eless senso ne wo ks, i is possible o enhance
ene gy e iciency, op imize ba e y managemen , and p omo e sus ainable ene gy p ac ices.
These applica ions ein o ce he ele ance o machine-lea ning-based SoH p edic ions in
imp o ing he esilience and sus ainabili y o ene gy sys ems. Finally, an assessmen o
implemen ed algo i hms and hei pe o mances p o ides a s uc u ed o e iew o he
ield, iden i ying oppo uni ies o u u e ad ancemen s.
Keywo ds: s a e o heal h; ba e y; machine lea ning; P oKnow-C; public da ase s; ene gy
in o ma ics; sma g ids; in e ne o hings; deep lea ning
1. In oduc ion
The wo ldwide inc ease in ba e y usage is e iden in a ious ields, especially in
elec ic ehicles. Resea ch e o s aim o imp o e ba e y e iciency, ex end li espan, and
educe cha ging ime, d i en by he demands o a g owing global ma ke . Alongside
Ene gies 2025,18, 746 h ps://doi.o g/10.3390/en18030746
Ene gies 2025,18, 746 2 o 77
ad ancemen s in echnology, ehicle ba e y euse has eme ged as a key a ea o ocus.
Ba e ies can se e au omo i e pu poses un il hei capaci y d ops o abou 80% o he
nominal alue. Beyond his poin , eplacemen is necessa y o mee he powe equi e-
men s o ehicles [
1
]. Howe e , he cells om hese ba e ies can s ill be epu posed o
o he applica ions, such as s a iona y ene gy s o age sys ems connec ed o pho o ol aic
gene a ion de ices. This p ocess, known as second use, o e s a sus ainable way o ex end
ba e y li e.
The epu posing o a second-use ba e y is s ill a p ocess ha equi es imp o emen
because o cons uc ion and sa e y di icul ies. Accu a e es ima ion o ba e ies’ SoH is
pi o al in ad ancing sus ainable ene gy solu ions. By in eg a ing SoH p edic ions in o
sma g ids and IoT sys ems, i is possible o op imize ene gy managemen , enhance sys em
esilience, and educe was e, aligning wi h b oade ene gy in o ma ics and sus ainabili y
goals. Wi h inc easing compu a ional ad ances, he p esence o sma senso s, and he
e a o big da a, he e has been g owing esea ch in e es in applying machine-lea ning
(ML) algo i hms o a i icial in elligence [
2
–
4
]. Accu a e SoH cha ac e iza ion is essen ial
o assessing cells sui able o euse. As new da ase s become a ailable, he olume o
esea ch connec ing ML o SoH es ima ion con inues o g ow, as demons a ed by nume ous
ecen s udies [5–9].
The es ima ion o he SoH o ba e ies is a c i ical s ep o enhancing hei li ecycle man-
agemen , especially in applica ions whe e eliabili y and pe o mance a e pa amoun [
1
].
Commonly employed me hods o SoH es ima ion can be b oadly classi ied in o elec o-
chemical app oaches and model-based, da a-d i en, and hyb id me hods [
10
–
13
]. Elec o-
chemical app oaches, al hough less common in ope a ional en i onmen s because o hei
in asi e na u e, o e unpa alleled p ecision o unde s anding ba e y deg ada ion mecha-
nisms. Fo ins ance, di e en ial ol age analysis and di e en ial capaci y analysis [
14
,
15
]
a e used o ack speci ic aging signa u es by analyzing ol age–capaci y p o iles du ing
cha ge/discha ge cycles. These me hods, combined wi h echniques like cyclic ol am-
me y [
16
] o ad anced elec ochemical impedance spec oscopy [
17
], p o ide de ailed
insigh s in o phenomena such as li hium pla ing and ac i e ma e ial loss. Al hough such
me hods a e ypically applied in labo a o y se ings, ecen ad ances in senso echnology
and signal p ocessing aim o make hem mo e easible o eal- ime SoH es ima ion [15].
Model-based me hods ely on elec ochemical o equi alen -ci cui models o p e-
dic he SoH by cap u ing he physical and chemical beha io s o he ba e y [
18
]. Tech-
niques such as elec ochemical impedance spec oscopy [
19
], Kalman il e ing [
18
], and
pa icle il e ing [
20
] a e widely used in his ca ego y. These me hods o e p ecise in-
sigh s in o ba e y pe o mance bu o en equi e complex pa ame e uning and a e
compu a ionally in ensi e [18,20].
Da a-d i en me hods, on he o he hand, u ilize ML and DL algo i hms o analyze
la ge da ase s and unco e pa e ns indica i e o ba e y deg ada ion [
10
]. These me hods
excel in modeling nonlinea ela ionships and adap ing o di e se ba e y chemis ies and
usage pa e ns [
10
,
11
]. Howe e , hei eliance on la ge amoun s o labeled da a and chal-
lenges in in e p e abili y limi hei di ec applica ion in some
scena ios [10,11,13].
Hyb id
me hods combine he s eng hs o model-based and da a-d i en app oaches, le e aging
physical models o enhance he in e p e abili y and obus ness o ML-based p edic ions.
In [21,22],
hyb id app oaches in eg a ing equi alen -ci cui models wi h ML echniques a e
p oposed, achie ing a balance be ween accu acy and compu a ional complexi y. Despi e
hei p omise, hyb id me hods o en equi e signi ican domain-speci ic expe ise and
ex ensi e compu a ional esou ces [23].
Recen ad ancemen s in compu a ional powe and he a ailabili y o la ge da ase s
ha e signi ican ly boos ed he p ominence o da a-d i en me hods, making hem a co ne -
Ene gies 2025,18, 746 3 o 77
s one in SoH es ima ion o la ge-scale applica ions, like elec ic ehicles and s a iona y
ene gy s o age
sys ems [10,11].
None heless, challenges pe sis ega ding da a a ailabili y,
algo i hmic gene aliza ion, and sys em in e p e abili y [13].
The wo k p esen ed in [
10
] p o ides a lis o ad an ages and disad an ages o using
ML algo i hms, highligh ing he need o open pla o ms o da a sha ing and modeling
echniques as a necessa y s ep o he ad ancemen o he esea ch ield. In he con ex
o challenges and p ospec s, s udies [
24
–
26
], which ocus on explo ing DL echniques
o es ima ing he emaining ba e y li e, a e also no ewo hy, while in [
27
], his heme is
e iewed om he pe spec i e o ans e -lea ning usage. In [
28
], challenges and p ospec s
a e add essed conside ing he impo ance o ea u e ex ac ion, cons uc ion, and selec ion
o heal h s a e modeling. The impo ance o ba e y heal h cha ac e iza ion is p esen ed
in [
12
], unde he challenges o scaling second-use ba e ies. In [
11
], a ele an e iew
o
s a e-o -cha ge
(SoC) and -heal h es ima ion is p esen ed, whe e he au ho s e eal
compa a i e esul s mainly conside ing neu al ne wo ks, such as eed o wa d neu al
ne wo ks (FFNNs), ecu en neu al ne wo ks (RNNs), and long sho - e m memo ies
(LSTMs). S udies [
29
–
31
] also p o ide a e iew ocused on compa ing echniques o
s udying ba e y deg ada ion. In all he ele an e iew pape s in ecen yea s ha ha e
been analyzed, a common gap can be poin ed ou : he absence o a s uc u ed me hodology
ha unde pins he analysis po olio and leads o he au ho s’ conclusions.
Al hough signi ican esea ch has been conduc ed on da a-d i en algo i hms o SoH
es ima ion, sys ema ic me hodologies a e lacking o ensu e he selec ion o highly ele an
s udies o cons uc ing a eliable s a e-o - he-a o e iew. The absence o such app oaches
makes i di icul o iden i y eme ging ends in he ield. In his con ex , and gi en he
ele ance o he opic, his wo k aims o explo e he ecen
s a e-o - he-a
pano ama, om
he las 5 yea s, o he es ima ion o ba e ies’ SoHs. To achie e his, we s a wi h he
explana ion and demons a ion o a s uc u ed and sys ema ic me hodology, known as
P oKnow-C (Knowledge De elopmen P ocess Cons uc i is ) [
32
], o ob ain a cen al-
ized bibliog aphic po olio on he opic o p edic ing he heal h s a es o ba e ies, using
ML. P oKnow-C is a s uc u ed p ocess designed o assis esea che s in sys ema ically
iden i ying, selec ing, and analyzing a bibliog aphic po olio aligned wi h hei esea ch
objec i es [
32
–
34
]. I s ands ou as a comp ehensi e app oach o conduc ing li e a u e
e iews because i combines quan i a i e and quali a i e c i e ia, ensu ing he inclusion
o highly ele an and impac ul s udies while minimizing biases o en p esen in manual
selec ion p ocesses [
32
,
35
]. By applying his me hodology, we aim o cons uc a obus
bibliog aphic po olio ha p o ides a eliable ounda ion o e alua ing he s a e o he
a in ba e ies’ SoH es ima ions. In his way, his pape ’s con ibu ions a e summa ized
as ollows:
Applica ion o he P oKnow-C Me hodology: The p esen a ion and demons a ion
o he P oKnow-C sys ema ic me hodology o building a bibliog aphic po olio. This
sys ema ic app oach allows o igo ous and s uc u ed li e a u e e iews.
Cha ac e iza ion o he Resea ch Scena io (S a e o he A ): Cha ac e iza ion o he
cu en scena io o s udies in he heal h-s a e es ima ion o ba e ies, using ML by analyzing
534 ele an a icles published be ween 2018 and 2024. This p o ides a comp ehensi e
s a e o he a in he cu en esea ch ield.
Public Da ase Compila ion: P esen a ion, de ailing, and summa y o 20 public
da ase s om 12 di e en sou ces, p ima ily om uni e si y esea ch cen e s, o selec ing
sui able da ase s o esea ch in SoH, SoC, and ba e y ene gy s o age sys ems.
Machine-Lea ning and Deep-Lea ning Algo i hms: Resea ch o he main ML algo-
i hms used in s udies p edic ing esponse a iables ela ed o ba e y deg ada ion, includ-
ing deep-lea ning, hyb id, and ans e -lea ning models.
Ene gies 2025,18, 746 4 o 77
Pe o mance Analysis o SoH Es ima ion Models: A comp ehensi e pe o mance
analysis o s a e-o -heal h es ima ion models, ocusing on a ious esponse a iables,
including SoH, emaining-use ul-li e, cu en -li ecycle, capaci y, ajec o y, and ea ly-
use ul-li e p edic ions. The compa ison in ol es 21 s udies, allowing o bo h ai and
b oade compa isons.
Fi s S udy Applying P oKnow-C o Ba e ies’ SoHs: This s udy is he i s o apply
he P oKnow-C sys ema ic e iew me hod in he con ex o ba e ies’ s a e-o -heal h es i-
ma ions. This pionee ing applica ion se s a new s anda d o s uc u ed li e a u e e iews
in his ield.
The emainde o his pape is o ganized as ollows: Sec ion 2de ails a sys ema ic
e iew using he P oKnow-C me hodology, enabling igo ous and s uc u ed li e a u e
selec ion, mapping he s a e o he a in s udies and expe imen s and a ailable open
da ase s o he applicabili y o hese echniques. Sec ion 3discusses in de ail he con en
analysis o he bibliog aphic e iew, as well as he analyses o he pape s composing he
bibliog aphic po olio on his opic, including he main da abases and ML algo i hms,
along wi h he key li e a y s udies. Addi ionally, his sec ion explo es he po en ial p ac ical
applica ions o SoH es ima ion in ene gy in o ma ics, sma g ids, and IoT sys ems, high-
ligh ing i s ole in enhancing ene gy e iciency, sus ainabili y, and ope a ional esilience.
Finally, Sec ion 4o e s concluding ema ks, summa izes he main esul s, and p o ides
sugges ions o u u e esea ch, explo ing po en ial de elopmen s based on he in eg a ion
o a i icial in elligence in a ious scena ios while iden i ying gaps and oppo uni ies o
u u e esea ch.
2. Sys ema ic Re iew
The adop ion o sys ema ic p ocesses o bibliog aphic su eying allows o op imizing
he quali y o he ma e ial ob ained on a pa icula opic, as i makes he p ocess mo e
analy ical and igo ous, he eby imp o ing he eliabili y o he esul s ound. As his is
an ini ial and undamen al s age o all esea ch de elopmen , me hods ha inc ease he
obus ness o a bibliog aphic po olio a e essen ial [36].
In his pape , he sys ema ic me hod P oKnow-C is employed o ob ain a ecen
and scien i ically ele an bibliog aphic po olio on he use o ML in es ima ing SoHs o
ba e ies. P oKnow-C was de eloped a he Labo a o y o Mul ic i e ia Me hodologies
in Decision Suppo (LabMCDA) a he Fede al Uni e si y o San a Ca a ina (UFSC) and
pa en ed in 2010 [
32
]. This me hod has been applied in esea ch in a ious a eas, and some
examples o P oKnow-C applica ions can be obse ed in [33,35,36].
Wi hin he ield o elec ic ba e ies, he P oKnow-C me hod was applied in [
34
] o
de ine he s a e o he a in li hium-ion-ba e y ecycling. Al hough ela ed, he p esen
s udy speci ically ocuses on analyzing he s a e o he a in SoH es ima ion using ML
me hods. To da e, no simila s udy applying P oKnow-C has been obse ed.
The P oKnow-C me hod consis s o ou main s ages [36]:
•
Selec ion o a po olio o pape s on he esea ch opic: This in ol es de ining esea ch
keywo ds, sea ching in da abases, and il e ing a icles based on alignmen wi h he
esea ch objec i e, ci a ion me ics, and ele ance;
•
Bibliome ic analysis o he po olio: This s age examines scien i ic indica o s, such as
he numbe o a icles, ci a ion coun s, au ho s, and jou nals, o assess he po olio’s
comp ehensi eness and scien i ic impac ;
•
Sys emic analysis: The selec ed a icles a e deeply analyzed o insigh s and pa e ns
and he iden i ica ion o possible esea ch gaps;
Ene gies 2025,18, 746 5 o 77
•
De ini ion o he esea ch ques ion and objec i e: The esul s om he p e i-
ous s ages a e syn hesized o e ine he scope and o mula e p ecise esea ch
ques ions and objec i es.
This pape p esen s he esul s o he i s h ee s ages o he P oKnow-C me hod, along
wi h he analysis o he selec ed ele an pape s. These s ages ep esen a comp ehensi e
s a e-o - he-a e iew o he b oade esea ch ield, se ing as a basis o e ining he ocus
o a mo e speci ic and well-de ined niche.
O he well-known sys ema ic e iew me hods can be ound in he li e a u e and
may be used as al e na i es o P oKnow-C. The PRISMA (P e e ed Repo ing I ems o
Sys ema ic Re iews and Me a-Analyses) [
37
] emphasizes anspa ency and eplicabili y
h ough s ic adhe ence o p ede ined inclusion and exclusion c i e ia, making i widely
ega ded as a gold s anda d in ields such as ene gy sys ems, en i onmen al science,
a i icial in elligence, and o he echnical domains [
38
–
40
]. Howe e , PRISMA does no
include a bibliome ic e alua ion phase o ools o mul ic i e ia decision-making, which a e
cen al o P oKnow-C. Simila ly, SALSA (Sea ch, App aisal, Syn hesis, and Analysis) [
41
]
ocuses mo e on syn hesizing and analyzing e idence bu lacks he po olio alignmen
capabili ies o P oKnow-C, which ensu es a a ge ed and ele an selec ion o a icles.
Ano he me hod, Scoping Re iews, is designed o map he b ead h and dep h o he
li e a u e on a opic, making i well-sui ed o explo a o y s udies o iden i ying gaps in he
li e a u e [
42
]. Al hough Scoping Re iews p o ides a b oad o e iew, i is less s uc u ed
in e ms o bibliome ic e alua ion and o en does no employ mul ic i e ia ools o e ine
he po olio, which a e key s eng hs o P oKnow-C [43].
Howe e , as wi h any me hod, P oKnow-C has i s limi a ions. The subjec i e align-
men analysis s age, al hough use ul o ailo ing he po olio o speci ic objec i es, may
educe epea abili y [
35
]. Addi ionally, i s eliance on ci a ion me ics migh o e look
eme ging bu seldom-ci ed s udies [35].
2.1. Bibliog aphic Po olio Selec ion
This sec ion desc ibes he selec ion o he bibliog aphic po olio, ini ially, he se
o axes and keywo ds ha encompass he heme o his esea ch, i.e., he use o ML in
es ima ing SoH, was de ined. As shown in Table 1, axis 1 co esponds o he s udy objec ,
which is ba e ies. Axes 2 and 3 encompass e ms ela ed o he de ini ion o SoH and i s
es ima ion, espec i ely. Axis 4 includes e ms ela ed o a i icial in elligence algo i hms,
machine lea ning, deep lea ning, and ensembles.
Table 1. Resea ch axes o he bibliog aphic po olio selec ion.
Axis 1 Axis 2 Axis 3 Axis 4
ba e y
s a e o heal h es ima ion machine lea ning
cycle li e p edic ion neu al ne wo k
li e ime ea u es ans e lea ning
aging second use a i icial
in elligence
deg ada ion boos ing
use ul li e quan ile eg ession
ensemble
deep lea ning
Ene gies 2025,18, 746 6 o 77
The axes we e combined using he condi ional logic AND, esul ing in 192 combi-
na ions sea ched in he Scopus da abase. Fil e s we e es ablished o documen s o he
ypes o pape s and e iews, sea ching o he keywo ds in i les, keywo ds, and abs ac s,
as well as de ining a esea ch ho izon o publica ions o up o 5 yea s old. The Scopus
da abase was selec ed because o he la ge olume o pape s e u ned compa ed o o he
da abases, such as Web o Science, as well as he p esence o jou nals ocused on a eas
possibly ela ed o he esea ch. An example o a condi ion esul ing om he combina ion
o axes was (TITLE-ABS-KEY(ba e y) and TITLE-ABS-KEY(s a e o heal h) and TITLE-
ABS-KEY(p edic ion) and TITLE-ABS-KEY(neu al ne wo k) and PUBYEAR > 2017 and
(LIMIT-TO (DOCTYPE, “a ”) OR LIMIT-TO (DOCTYPE, “ e”)).
Table 2p esen s he adhe ence me ics o he keywo ds used in he combina ions o
axes. The pe cen ages shown quan i y he po ions o he o al numbe o aw a icles in
which a pa icula keywo d was included in he pe o med combina ions. Wi hin axis 2
combina ions, a highe adhe ence a e o he e m “s a e o heal h” is obse ed, while in
axis 3 and 4 combina ions, he keywo ds “p edic ion” and “neu al ne wo k” s and ou ,
espec i ely. These adhe ence me ics sugges ha wi hin he heme o esea ch ela ed
o ba e y’s s a e o heal h, he e m “s a e o heal h” ends o be mo e applied, o en in
connec ion wi h “p edic ion” s udies u ilizing “neu al ne wo ks”. I is impo an o no e
ha al hough some e ms show low adhe ence a es, hey emain ele an o iden i ying
po en ially impo an pape s ha may explo e eme ging ends in an a ea o esea ch
s ill unde explo ed.
Table 2. Adhe ence o he esea ch axes.
Axis Keywo d Keywo d Adhe ence Ra e
Axis 2
s a e o heal h 29.3%
deg ada ion 20.6%
aging 17.7%
use ul li e 13.6%
cycle li e 10.9%
li e ime 7.9%
Axis 3
p edic ion 36.5%
es ima ion 30.9%
ea u es 24.9%
second use 7.8%
Axis 4
neu al ne wo k 36.9%
machine lea ning 28.6%
deep lea ning 14.6%
ensemble 6.2%
ans e lea ning 5.5%
a i icial in elligence 4.5%
boos ing 3.4%
quan ile eg ession 0.3%
This sea ch was conduc ed on 14 Janua y 2024, esul ing in a o al o 6032 pape s
(wi h 275 pape s om 2024). Al hough he e we e pape s om 2024, o he calcula ion o
a publica ion ho izon o up o 5 yea s, esea ch om 2018 onwa d was conside ed, hus
ha ing 6 comple e yea s o publica ions o analysis plus wo weeks o publica ions in 2024.
The pape s we e expo ed in “.cs ” o ma in each i e a ion o he 192 combina ions o axes.
Ene gies 2025,18, 746 7 o 77
The ini ial low p oposed by P oKnow-C is p esen ed in Figu e 1. The objec i e o his
i s s age is o signi ican ly educe he olume o pape s in he RPD ( aw pape da abase)
ob ained om he combina ions o esea ch axes. To achie e his, il e s a e applied o
pe o m a p elimina y selec ion o a icles ela ed o he esea ch heme. The ollowing
il e s a e applied:
Figu e 1. Flow I o ob aining he bibliog aphic po olio.
Redundancy Fil e : This is he i s s ep o P oKnow-C, whe e he RPD pape s a e
analyzed o duplica ion. In his s age, he “.cs ” iles esul ing om he axis combina ions
we e p ocessed h ough a Py hon sc ip ha pe o med conca ena ion and emo al o
duplica es acco ding o he i le and publica ion yea ields. A o al o 4682 samples we e
emo ed om he RPD.
Ti le Alignmen Fil e : This in ol es eading he pape s’ i les o assess whe he hey
a e aligned wi h he esea ch heme, as iden i ied by he esea che s. Ou o he 1350 pape s
emaining a e he p e ious il e , 722 we e deemed o be no aligned wi h he esea ch.
Among he pape s no selec ed we e s udies ocused on SoH analysis in elec ochemical
con ex s and labo a o y expe imen al phases, which a e conside ed as p elimina y s eps
be o e explo ing da abases and implemen ing ML models.
Scien i ic Recogni ion Fil e : This s ep in ol es analyzing he numbe o ci a ions
wi hin he RPD. In his s ep, he emaining 672 pape s a e so ed in descending o de by
ci a ion coun . Acco ding o he cumula i e pe cen age o ci a ions and a p ede ined cu o
pe cen age, he po olio is di ided in o wo eposi o ies: K and P. The K eposi o y consis s
o pape s conside ed as scien i ically ecognized, con aining 80% o he ci a ions in he
inpu po olio o his il e , o aling 85 publica ions in he case s udy. The cu o pe cen age
is de e mined by he esea che s, wi h [
32
,
36
] ecommending a ange be ween 70% and
90%. The P eposi o y comp ises 543 pape s exceeding he de ined cu o h eshold.
The second low o a icle selec ion s eps o he bibliog aphic po olio, using
P oKnow-C, is p esen ed in he lowcha in Figu e 2. In his second phase o he me hod,
he objec i e is o e i y he alignmen o he pape s emaining om he i s phase wi h
he con en p esen ed in hei abs ac s. The ollowing s eps a e applied:
Ene gies 2025,18, 746 8 o 77
Figu e 2. Flow II o ob aining he bibliog aphic po olio.
K Reposi o y Alignmen : The abs ac s o he pape s conside ed as scien i ically
ecognized a e analyzed by he esea che (s) o de e mine whe he he esea ch aligns wi h
he in ended esea ch objec i es. I he a icle is aligned, i emains in he P oKnow-C low;
o he wise, i is excluded. The 85 a icles in eposi o y K we e conside ed as aligned wi h
SoH es ima ions using ML.
C ea ion o he Au ho Da abase (AD) and Reposi o y A: This s ep in ol es iden i ying
he au ho s o he pape s app o ed in he p e ious s ep and c ea ing a da abase o au ho s
deemed as ele an o he esea ch heme. The selec ed pape s a e conside ed as aligned
wi h he esea ch heme and o m eposi o y A, which cons i u es he i s pa o he
inal po olio.
P Reposi o y Alignmen : This s ep analyzes he pape s ha did no each he le el
o scien i ic ecogni ion. These pape s a e di ided in o wo ca ego ies based on hei yea
o publica ion. A icles published mo e han wo yea s ago a e p e-selec ed i one o hei
au ho s is p esen in he AD co esponding o eposi o y A. I no ma ch is ound in he
au ho da abase, he pape s a e excluded. The emaining pape s a e hen e alua ed o
abs ac alignmen , and i he expec ed alignmen is con i med, hey a e app o ed in he
low and included in eposi o y B. Recen a icles a e no e alua ed based on he AD;
ins ead, hei abs ac s a e di ec ly analyzed o alignmen , and app o ed pape s a e added
o eposi o y B. In his case s udy, in he ini ial analysis o eposi o y P, which con ained a
o al o 543 pape s, 518 we e ecen publica ions om he pas wo yea s. O he wen y- i e
a icles olde han wo yea s, eigh een we e excluded because o he absence in he AD,
and he emaining se en we e added o he g oup con aining he 518 ecen pape s. O he
525 pape s analyzed in his s age, 449 we e ound o be aligned and o med eposi o y B.
C ea ion o Reposi o y C: This s ep in ol es combining eposi o ies A and B o o m
he inal bibliog aphic po olio esul ing om he applica ion o P oKnow-C. The inal
po olio comp ises pape s aligned wi h he esea ch heme, including scien i ically ec-
ognized s udies in e ms o ci a ions, ecen a icles wi h po en ial, and publica ions by
esea che s deemed as ele an o he ield.
The union o eposi o ies A and B o ms he inal bibliog aphic po olio wi h
534 pape s,
ep esen ing abou 9% o he ini ial aw pape po olio. F om his g oup,
a high deg ee o alignmen wi h he esea ch is expec ed, along wi h he abili y o desc ibe
he cu en s a e o he a , se ing as a basis o he de elopmen o he a ge esea ch.
Table 3p esen s he 40 mos ele an pape s in e ms o he numbe o ci a ions in he inal
po olio. This numbe is based on he ecommenda ion om [
32
] o e alua e an ideal ol-
Ene gies 2025,18, 746 9 o 77
ume o be ween 20 and 40 pape s. Howe e , each ield o esea ch and de elopmen phase
has i s own cha ac e is ics ha in luence he ideal olume o pape s. Because his wo k
aims o e eal he cu en esea ch scena io wi hin SoH es ima ion using ML, a po olio
app oxima ely en imes la ge han he olume ecommended by [
21
] was cons uc ed
o enable mo e obus in e ences ega ding he algo i hms employed, da ase s used, and
pe o mances achie ed.
Table 3. Top 40 pape s om he bibliog aphic po olio.
Ti le Ci a ions Re .
Da a-d i en p edic ion o ba e y cycle li e be o e
capaci y deg ada ion 1453 [1]
Long sho - e m memo y ecu en neu al ne wo k o
emaining-use ul-li e p edic ion o li hium-ion ba e ies 880 [44]
Da a-d i en heal h es ima ion and li e ime p edic ion o
li hium-ion ba e ies: A e iew 749 [10]
A da a-d i en app oach wi h unce ain y quan i ica ion
o p edic ing u u e capaci ies and emaining use ul li e
o li hium-ion ba e ies
434 [45]
P edic ing he s a es o cha ge and heal h o ba e ies
using da a-d i en machine lea ning 405 [46]
Random o es eg ession o online capaci y es ima ion
o li hium-ion ba e ies 398 [47]
Remaining-use ul-li e p edic ion o li hium-ion ba e ies
based on a hyb id model combining he long sho - e m
memo y and Elman neu al ne wo ks
316 [48]
Remaining-use ul-li e p edic ion o li hium-ion ba e ies:
A deep-lea ning app oach 313 [49]
A da a-d i en au o-CNN-LSTM p edic ion model o
li hium-ion-ba e ies’ emaining use ul li e 291 [50]
S a e-o -heal h es ima ion and emaining-use ul-li e
p edic ion o he li hium-ion ba e y based on a a ian
long sho - e m memo y neu al ne wo k
284 [51]
Machine lea ning applied o elec i ied- ehicle-ba e ies’
s a e-o -cha ge and s a e-o -heal h es ima ions: S a e o
he a
267 [11]
Modi ied Gaussian p ocess eg ession models o cyclic
capaci y p edic ion o li hium-ion ba e ies 262 [52]
A deep-lea ning me hod o online capaci y es ima ion o
li hium-ion ba e ies 260 [53]
Machine-lea ning pipeline o ba e ies’ s a e-o -heal h
es ima ions 246 [54]
A neu al-ne wo k-based me hod o RUL p edic ion and
SOH moni o ing o li hium-ion ba e ies 245 [55]
A no el es ima ion me hod o he s a e o heal h o
li hium-ion ba e ies using a p io -knowledge-based
neu al ne wo k and a Ma ko chain
239 [56]
Ene gies 2025,18, 746 16 o 77
Figu e 9. Dis ibu ion o he numbe o publica ions by jou nal in he bibliog aphic po olio.
2.2.4. Rele ance o Keywo ds
The bibliog aphic po olio consis s o 2753 keywo ds, o which 1044 a e unique.
Figu e 10 p esen s he dis ibu ion o he keywo ds, whe e a signi ican concen a ion is
obse ed o he e ms “li hium-ion ba e y” and “s a e o heal h”, which a e, indeed, he
objec s and cen al hemes o his esea ch. The p ominence o “s a e o heal h” ein o ces
i s posi ion as a pi o al concep in his s udy, guiding much o he esea ch e o s in his
ield. Addi ionally, “machine lea ning” appea s as he i h mos equen e m, e lec ing
he c i ical ole o a i icial in elligence in ad ancing ba e y SoH es ima ion.
The e m “ emaining use ul li e” is also no able, co esponding o one o he main
esponse a iables in he s udy o he SoH. Rega ding echniques, he equen appea ances
o “LSTM neu al ne wo ks” and “deep lea ning” highligh he inc easing adop ion o ad-
anced compu a ional models. This e lec s he g owing sophis ica ion in p edic i e analy -
ics, as esea che s seek mo e accu a e and obus app oaches o model
ba e y deg ada ion.
Figu e 11 depic s he dis ibu ion o keywo d connec ions wi hin he selec ed pape s.
The pa e n o e ms mi o s he p e ious dis ibu ion, showing how cen al keywo ds,
such as “s a e o heal h” and “machine lea ning”, b anch ou ac oss di e se con ex s. This
in e connec edness illus a es he mul idisciplina y na u e o SoH esea ch, b idging ields
like ene gy sys ems, a i icial in elligence, and sus ainabili y.
Ene gies 2025,18, 746 17 o 77
Figu e 10. Dis ibu ion o keywo ds in he bibliog aphic po olio.
Figu e 11. Numbe o connec ions o keywo ds wi hin he bibliog aphic po olio.
Figu es 10 and 11 oge he emphasize he impo ance o keywo ds in s uc u ing and
ad ancing he ield. Al hough he dominan e ms e lec he cu en esea ch ocus, he
a ie y and connec ions among keywo ds indica e he e ol ing bounda ies o he ield and
i s esponsi eness o eme ging challenges and echnologies.
3. Con en Analysis
The bibliog aphic po olio o 534 pape s, ollowing he P oKnow-C me hodology
analyzed in he p e ious sec ion, was explo ed o cha ac e ize he cu en scena io in he
ield o SoH es ima ion in ba e ies. Fi s , we ou line he su ey o public da abases ound
in he bibliog aphic po olio, ollowed by he echniques and algo i hms implemen ed
in he pape s. The algo i hms a e u he analyzed acco ding o ca ego ies o modeling,
Ene gies 2025,18, 746 18 o 77
including DL, algo i hmic hyb idiza ion, and ans e lea ning. A si ua ional o e iew
o he pe o mance is highligh ed, and key e iew s udies in he ield a e analyzed. This
sec ion concludes wi h implica ions o ene gy in o ma ics and in elligen sys ems.
3.1. Po olio O e iew
Excep o e iew pape s, he objec i es o he s udies a e di ec ly ela ed o es ablish-
ing a ious o ms, ei he algo i hmically o h ough di e en app oaches, o pe o m SoH
es ima ion. In he p esen a ion o he mos ci ed pape s in Table 3, i is no ed ha he e is a
signi ican ocus on es ing di e en ypes o ML algo i hms and how hei me hods can
inc ease he accu acy o SoH es ima ion. This is he case wi h he s udy p esen ed in [
44
]
wi h long sho - e m memo y (LSTM) neu al ne wo ks, he combina ion o LSTM and
ex eme-lea ning-machine (ELM) neu al ne wo ks p esen ed in [
48
], o he use o ensemble
me hods, such as bagging in decision ees, h ough he andom o es (RF) algo i hm
in [
47
]. In summa y, pape s wi h he highes numbe o ci a ions in he po olio ypically
ocus on inc easing es ima ion accu acy h ough expe imen s ela ed o algo i hms.
Compa ible s udies can also be highligh ed, such as he s udy p esen ed in [
78
],
which compa es he LSTM, FNN, and CNN neu al ne wo k algo i hms, whe e he LSTM
algo i hm achie ed he bes pe o mance, wi h an MAPE (mean absolu e pe cen age e o )
o a ound 0.5%, compa ed o abou 1.5% o FNN and 2% o CNN ne wo ks. In [
79
],
he au ho s compa e di e en p obabilis ic and ime-se ies algo i hms (ARIMAX, linea
quan ile eg ession, boo s ap mul iple linea eg ession, and Bayesian boo s ap mul iple
linea eg ession), wi h he bes esul s ob ained om he Bayesian boo s ap mul iple
linea eg ession algo i hm’s quan ile eg ession, which achie ed MAPE alues anging
om 0.2% o 1%. O he highligh ed compa a i e s udies include [66,80].
The concep o ea u e enginee ing, which includes manipula ion and selec ion, is a
ele an heme p esen ed in he po olio, as much o he algo i hm’s pe o mance lies in
he s ess o c ea ing ea u es ha p o ide disc imina ion o p edic ions. S udy [
81
] in o-
duces an au onomous ea u e selec ion me hod, which is based i s on an ini ial selec ion
conside ing co ela ion coe icien s, ee algo i hms, and a a iance ac o , ollowed by an
i e a i e me hod o ea u e combina ion. In [
82
], an analysis o ea u e enginee ing o SoH
es ima ion is p esen ed, e alua ing di e en echniques o ea u e c ea ion and selec ion,
such as uni a ia e selec ion by Pea son co ela ion, ea u e impo ance, ea u e clus e ing,
gene ic algo i hms, and sequen ial ea u e selec ion. The au ho s concluded ha he use o
sequen ial selec ion p esen ed a good balance be ween pe o mance and compu a ional
cos . Fea u e es s we e e alua ed using SVM and Ex aT ee-based algo i hms. O he
s udies ocusing on ea u es can be consul ed in [81,83–96].
The ques o au oma ing modeling p ocesses was ound in he de elopmen o an
au oML app oach in [
97
]. The amewo k buil is capable o pe o ming he en i e modeling
cycle using Bayesian op imiza ion, elimina ing he need o esea che s om o he ields
o spend ime on labo ious s eps, such as ea u e ex ac ion, cons uc ion, and selec ion.
The esul s ob ained yielded MAEs (median absolu e e o s) anging om abou 0.02% o
0.05% o SoH es ima ion.
In e p e abili y model analyses we e ound in s udies [
98
,
99
], based on he use o a
echnique known as SHAP (Shapley Addi i e Explana ions). SHAP is based on a heo e ical
game app oach ha seeks o explain he ou pu o any ML model by quan i ying how each
ea u e impac s he model’s p edic ion [
100
,
101
]. O he model in e p e abili y app oaches
we e also add essed in [102–105].
App oaches ela ed o hype pa ame e uning we e also ound in he po olio. In [
106
],
he au ho s explo e he Bayesian op imiza ion o hype pa ame e s in a combina ion o
DCNN and LSTM neu al ne wo ks, achie ing an RMSE ( oo -mean-squa e e o ) o 0.0061
Ene gies 2025,18, 746 19 o 77
o SoH es ima ion. In s udy [
107
], a pipeline op imiza ion based on a ee and gene ic
algo i hm is p esen ed. S udy [
108
] also uses a gene ic algo i hm as a means o pa ame e
op imiza ion, p esen ing a amewo k o SoH es ima ion, wi h e o s o abou 2%.
The use o senso s o cap u ing ba e y condi ions implies abula da a; howe e ,
s udies we e ound in he po olio, which analyze he implemen a ion o image-based
algo i hms o SoH p edic ion. In [
109
], he au ho s p opose a me hod capable o using
only one cha ge and discha ge cycle o SoH p edic ion, using image p ocessing o cu en
and ol age cu es. Using ans e lea ning, he au ho s achie ed an MAPE in he ange o
10%. T ans e lea ning is also used in he algo i hm based on ba e y cu e image analysis
in [
110
]. The au ho s analyze images o one cycle, i e cycles, and en cycles, wi h MAEs o
abou i y, i y- i e, and six y cycles, espec i ely, using eigh p e ained ne wo ks, such
as ResNe and GoogleNe . O he s udies using algo i hms om he compu e ision a ea
we e ound in [111–113].
Rega ding he cell echnologies employed, almos all he publica ions co espond
o li hium-ion echnology, among which we can highligh LFP (li hium i on phospha e),
LCO (li hium cobal oxide), NCA (li hium nickel cobal aluminum oxide), and NMC
(li hium nickel manganese cobal oxide) ba e y ypes. Only h ee s udies made use o
ba e y echnologies di e en om li hium. S udy [
114
] analyzes he es ima ion o he
SoHs o emo ed lead–acid ba e ies, aiming o euse. In [
115
], lead–acid ba e ies a e also
analyzed, and he au ho s de elop an SoH p edic ion model using LSTM ne wo ks based
on cha ge cu e da a. In [
116
], he au ho s use a neu al ne wo k o p edic he emaining
li espan o a zinc-ion ba e y.
3.2. Li e a u e Re iew
Wi hin he a icle selec ion p ocess, 38 pape s co espond o e iew pape s, and hey
a e p esen ed in Table 4. The pape s a e essen ially di ided in o e iews wi h quali a i e
analyses (e.g., ends, challenges, and gene al o e iews), as well as pape s mo e ocused on
a speci ic se o echniques and su eying pe o mances and he ex ac ion o deg ada ion
ea u es and heal h indica o s. In all he analyzed pape s, he e was no indica ion o he
use o a me hodological p ocess o selec ing he bibliog aphic po olio, highligh ing he
impo ance o his wo k as a poin o e olu ion wi hin he esea ch ield.
Table 4. Re iew pape s in he bibliog aphic po olio.
Ti le Yea Ci ed Re .
Da a-d i en heal h es ima ion and li e ime
p edic ion o li hium-ion ba e ies: A e iew 2019 749 [10]
Machine lea ning applied o
elec i ied- ehicle-ba e ies’ s a e-o -cha ge and
s a e-o -heal h es ima ion: S a e o he a
2020 267 [11]
A e iew o second-li e Li-ion ba e ies: p ospec s,
challenges, and issues 2022 213 [12]
A e iew o s a e-o -heal h es ima ions and
emaining-use ul-li e p ognos ics o
li hium-ion ba e ies
2021 200 [13]
A e iew o non-p obabilis ic
machine-lea ning-based s a e-o -heal h es ima ion
echniques o li hium-ion ba e ies
2021 180 [67]
Ene gies 2025,18, 746 20 o 77
Table 4. Con .
Ti le Yea Ci ed Re .
A c i ical e iew o imp o ed deep-lea ning me hods
o he emaining-use ul-li e p edic ion o
li hium-ion ba e ies
2021 159 [5]
So ing, eg ouping, and echelon u iliza ion o
la ge-scale e i ed li hium ba e ies: A c i ical e iew
2021 117 [9]
Big aining da a o a i icial-in elligence-based
Li-ion diagnoses and p ognoses 2020 100 [117]
Machine lea ning in s a e-o -heal h and
emaining-use ul-li e es ima ion: Theo e ical and
echnological de elopmen s in ba e y
deg ada ion modeling
2022 88 [118]
S a e-o -heal h p edic ion o li hium-ion ba e ies
based on machine lea ning: Ad ances
and pe spec i es
2021 81 [119]
A c i ical e iew o imp o ed deep con olu ional
neu al ne wo ks o mul i- imescale s a e p edic ion
o li hium-ion ba e ies
2022 75 [30]
A e iew o deep-lea ning app oaches o p edic he
s a es o heal h and s a es o cha ge o
li hium-ion ba e ies
2022 69 [26]
A c i ical e iew o online
ba e y- emaining-use ul-li e ime
p edic ion me hods
2021 62 [120]
A i icial neu al ne wo ks, g adien boos ing, and
suppo ec o machines o elec ic- ehicle-ba e ies’
s a e es ima ion: A e iew
2022 57 [31]
S a e-o -heal h es ima ion and emaining-use ul-li e
assessmen o li hium-ion ba e ies: A
compa a i e s udy
2022 43 [121]
A e iew o mode n machine-lea ning echniques in
he p edic ion o he emaining use ul li e o
li hium-ion ba e ies
2023 34 [122]
O e iew o machine-lea ning me hods o
li hium-ion-ba e ies’ emaining-use ul-li e ime
p edic ion
2021 33 [123]
A e iew o machine-lea ning-based s a e-o -cha ge
and s a e-o -heal h es ima ion algo i hms o
li hium-ion ba e ies
2023 33 [124]
T ans e lea ning o ba e ies’ sma e -s a e
es ima ion and aging p ognos ics: Recen p og ess,
challenges, and p ospec s
2023 32 [27]
Re iew o “g ay box” li e ime modeling o
li hium-ion ba e ies: Combining physics and
da a-d i en me hods
2022 31 [125]
Ene gies 2025,18, 746 21 o 77
Table 4. Con .
Ti le Yea Ci ed Re .
Deep-lea ning-enabled s a e-o -cha ge,
s a e-o -heal h, and emaining-use ul-li e es ima ions
o sma ba e y managemen sys ems: Me hods,
implemen a ions, issues, and p ospec s
2022 26 [24]
Explainabili y-d i en model imp o emen o SOH
es ima ion o li hium-ion ba e ies 2023 20 [126]
S a e es ima ion models o li hium-ion ba e ies o
ba e y managemen sys ems: S a us, challenges, and
u u e ends
2023 20 [127]
S a e-o -cha ge, emaining-use ul-li e, and
knee-poin es ima ions based on a i icial in elligence
and machine lea ning o li hium-ion EV ba e ies: A
comp ehensi e e iew
2022 19 [128]
The de elopmen o machine-lea ning-based
emaining-use ul-li e p edic ions o
li hium-ion ba e ies
2023 17 [129]
Comp ehensi e e iew o ba e y s a e es ima ion
s a egies using machine lea ning o ba e y
managemen sys ems o ai c a p opulsion ba e ies
2023 16 [130]
A comp ehensi e e iew o li hium-ion-ba e ies’
s a e-o -heal h p ognosis me hods combining aging
mechanism analysis
2023 11 [131]
Resea ch p og ess and applica ion o deep lea ning
in emaining-use ul-li e, s a e-o -heal h, and ba e y
he mal managemen o li hium ba e ies
2023 11 [132]
A e iew o he p edic ion o he heal h s a e and
se ing li e o li hium-ion ba e ies 2022 7 [6]
Specialized deep neu al ne wo ks o ba e y heal h
p ognos ics: Oppo uni ies and challenges 2023 7 [25]
Machine-lea ning echniques’ sui abili y o es ima e
he e ained capaci y in li hium-ion ba e ies om
pa ial cha ge/discha ge cu es
2023 7 [133]
Deep ea u e ex ac ion in li e ime p ognos ics o
li hium-ion ba e ies: Ad ances, challenges,
and pe spec i es
2023 6 [28]
Compa ing deep-lea ning me hods o p edic he
emaining use ul li e o li hium-ion ba e ies 2022 4 [134]
Machine-lea ning-based emaining-use ul-li e
p edic ion echniques o li hium-ion-ba e y
managemen sys ems: A comp ehensi e e iew
2023 2 [29]
Fea u e– a ge pai ing in machine lea ning o
ba e y heal h diagnosis and p ognosis: A
c i ical e iew
2023 2 [135]
Ene gies 2025,18, 746 22 o 77
Table 4. Con .
Ti le Yea Ci ed Re .
Resea ch on me hods o ex ac ing aging
cha ac e is ics and he heal h s a us o li hium-ion
ba e ies based on small samples
2022 1 [136]
Elec ic- ehicle-ba e ies’ capaci y deg ada ion and
heal h es ima ion using machine-lea ning echniques:
A e iew
2023 0 [137]
Open access da ase , code lib a y, and benchma king
deep-lea ning app oaches o s a e-o -heal h
es ima ions o li hium-ion ba e ies
2024 0 [138]
Figu e 12 shows ha he numbe o e iew a icle publica ions wi hin he po olio
has been inc easing o e he yea s, albei wi h a conside ably lowe coe icien compa ed
o ha o he o e all olume ic analysis. The e appea s o be a di e ence be ween 2022
and 2023, sugges ing a po en ial end owa d s abili y in he coming yea s.
Figu e 12. Volume ic analysis o he publica ion yea o e iew pape s in he BP.
The mos ci ed e iew a icle in he po olio is he s udy p esen ed in [
10
], which
examines big-da a echniques ega ding hei easibili y and cos e ec i eness in dealing
wi h ba e y heal h in eal-wo ld applica ions. The me hods a e ca ego ized, and ad an-
ages and limi a ions a e iden i ied. The au ho s begin by p esen ing me hods ha do no
in ol e model aining, such as he di e en ial analysis o cha ge and discha ge p o iles,
s ess es s, and he mal analyses. Then, hey e iew he use o ML o SoH es ima ion,
highligh ing he undamen al s ep o ea u e ex ac ion. They ca ego ize hese ea u es
in o h ee main g oups: (i) model- i ed ea u es, which depend on es s like in e nal e-
sis ance and a e no easily accessible by senso s in a BMS (ba e y managemen sys em);
(ii) p ocessed ex e nal ea u es, which a e he esul s o di e en ial analyses; and (iii) di ec
ex e nal ea u es, which a e all he a iables ha a senso can collec wi hin he ba e y
sys em and can gene a e a la ge numbe o a iables. The au ho s also b ie ly e iew
non-p obabilis ic ML me hods, such as a i icial neu al ne wo ks, SVMs, and p obabilis ic
models, like Gaussian eg ession.
The non-p obabilis ic me hods a e he cen al heme o he s udy p esen ed in [
67
],
whe e i e ypes o ML algo i hms o ba e ies’ SoH es ima ion a e e iewed: linea
eg ession, SVM, KNN, neu al ne wo ks, and ensemble me hods. The s udy compa a i ely
ou lines he ad an ages and applicabili y o he di e en me hods om a heo e ical
s andpoin . Th ee aspec s a e conside ed o compa ing he me hods: he algo i hm’s
Ene gies 2025,18, 746 23 o 77
pe o mance based on i e pe o mance me ics (RMSE, MAE, AE, APE, and MaxE), he
publica ion end ob ained by coun ing he numbe o publica ions in he las en yea s, and
he aining modes conside ing ea u e ex ac ion and selec ion. The s udy used 144 pape s
conside ed as ele an and published up o 10 yea s be o e (wi h he e e ence yea being
2021), howe e , wi hou e ealing he c i e ia used o ob aining he po olio. The au ho s
conclude ha neu al-ne wo k-based me hods and SVMs a e s ill unde esea ch and ha
DL me hods ha e shown g ea po en ial in SoH es ima ion unde complex ba e y-aging
condi ions, especially when big da a a e a ailable, and ha ensemble me hods, like andom
o es , can be conside ed as an eme ging al e na i e o balancing da a size and accu acy.
Rega ding he use o ML echniques in second-li e ba e ies, he s udy p esen ed
in [
9
] e iewed he s a us and challenges o la ge-scale second-li e applica ions. The
au ho s discuss me hodologies o classi ying and eg ouping e i ed ba e ies. They
p opose a apid, mul ile el, and mul idimensional classi ica ion me hod o la ge-scale use.
The classi ica ion me hod in ol es i s sol ing a one-dimensional classi ica ion p oblem
o ob ain simila ba e ies in e ms o hei eac ion s age. Then, a mul idimensional
classi ica ion is pe o med based on capaci y and in e nal esis ance, whe e usage scena ios
a e e alua ed, o example, o de e mine whe he he p io i y use is o ene gy o powe
supply. The second li e is also discussed in he e iew p esen ed in [
12
], which analyzes
economic, echnical, and en i onmen al ac o s ela ed o he use o second-li e li hium-ion
ba e ies, including SoH es ima ion me hods.
Rega ding he e iews om his yea , i is wo h highligh ing he s udy in [
27
], which
p esen s he i s sys ema ic e iew o ans e -lea ning applica ions in he ield o ba e y
managemen , ocusing on ba e ies’ s a e es ima ions and aging p ognoses. The au ho s
p o ide he s a e o he a in e ms o p inciples, algo i hmic s uc u es, ad an ages, and
disad an ages. Fo SoH es ima ion, a su ey o pape s in he ield showed ha ans e
s a egies ocus on p oblem domain adap a ion and he ine- uning o he inal model. The
di icul ies poin ed ou by he au ho s in using ans e lea ning lie in he low labeling
deg ee o he da a, which depends on he da a acquisi ion capabili y a sho e in e als in
a BMS. This is exace ba ed by he low equency o ac ual ba e y capaci y es ing du ing
usage, especially o SoH es ima ion pu poses.
3.3. Public Da abases
Analyzing he non- e iew pape s p esen in he bibliog aphic po olio, i was ound
ha abou 41% make use o p op ie a y and closed da ase s, wi hou sha ing eposi o ies
o use in o he s udies. On he o he hand, a signi ican and inc easingly g owing po ion
o pape s conduc in es iga ions using public da ase s, comp ising 59% o he non- e iew
pape s in he po olio. As emphasized in [
11
], ad ancemen s in he ield o ML o
es ima ing ba e ies’ SoHs ely on in o ma ion sha ing so ha new esea ch can de elop
and esul compa isons can occu , he eby allowing in e ences abou echniques ha may
enhance es ima ion accu acy. This scena io demons a es his sha ing end, leading o
as e and mo e oluminous de elopmen s in he esea ch ield. I is wo h no ing ha ai
compa a i e analyses o models/app oaches also equi e he sha ing o da a spli s used o
aining and es ing/ alida ion; only hen can compa isons be made when dealing wi h
he same popula ion.
Figu e 13 demons a es ha au ho -p o ided da ase s ha e he highes equency
o use. Howe e , he majo i y o hese da ase s a e complemen a y o public da ase s.
Among hese, he highligh goes o he use o da a p o ided by he P ognos ics Cen e
o Excellence Da ase Reposi o y [
139
], om NASA, which accoun s o 51% o he open
da ase s used in he su eyed po olio. The da ase p esen ed in [
1
], de eloped a he
Ene gies 2025,18, 746 24 o 77
Massachuse s Ins i u e o Technology (MIT), also cons i u es an impo an da a sou ce in
he su eyed pape s.
Figu e 13. Volume ic analysis o da ase s used in he pape s o he bibliog aphic po olio.
The e olu ion o he p opo ion o closed and public da ase s is p esen ed in Figu e 14.
I is no iceable ha he olume o applica ions using public da ase s s a s o become
p edominan om 2022, wi h he use o public da ase s being abou 3.2 imes highe in 2023.
This inc ease could be because o he esea ch end o using mul iple da ase s, and because
mo e da a sou ces a e a ailable, he applica ion o public da ase s would end o inc ease.
The e o e, o mi iga e his e ec , Figu e 14 conside s only he Boolean condi ion o whe he
a public da ase was used o no , and he esul s a e simila , wi h he numbe o pape s
using public da ase s in 2023 being abou 2.3 imes highe han hose using
closed da ase s.
Figu e 14. Annual e olu ion in he BP o pape s using public da ase s e sus closed da ase s.
Ene gies 2025,18, 746 25 o 77
Th ough an e alua ion acco ding o he da ase o igin, Figu e 15 illus a es he e olu-
ion o he da ase usage o e he yea s in he bibliog aphic po olio (BP). The inc easing
use o NASA da ase s is no iceable, ollowed by he usage o he MIT [
1
], Ox o d, and
CALCE da ase s. O he public da ase s wi h e en lowe le els o usage a e also iden i ied
in he po olio: he Beijing Ins i u e o Technology (BIT), Ca negie Mellon Uni e si y,
S an o d Uni e si y, Camb idge Uni e si y, he Uni e si y o Hawaii, Pu due Uni e si y
(UL-PUR), he Uni e si y o Bologna (UNIBO), and he Cen e o Elec ochemical Ene gy
S o age Ulm–Ka ls uhe (CELEST).
Figu e 15. Annual e olu ion, in he BP, o he o igin o public and au ho da ase s.
Table 5p o ides a summa y o each o he public da ase s ound in he po olio,
as well as hei cha ac e is ics and in which pape s hey we e used. In o al, 12 sou ces
o public da a we e e ealed, co esponding o 20 di e en da ase s, all using li hium
echnology as he main sou ce o he analyzed ba e ies. The syn hesis o hese da abases
cons i u es impo an in o ma ion o u u e s udies, as i acili a es he selec ion and design
o new s udies on SoHs.
In he NASA eposi o y, wo da ase s a e a ailable o de eloping models aimed a
es ima ing SoHs. The i s da ase con ains 34 li hium-ion 18,650 cells wi h a capaci y
o 2 Ah, unde going p ocesses o cha ging, discha ging, and impedance measu emen s.
Va ious empe a u es a e used, including 4
◦
C, 24
◦
C, and 44
◦
C, wi h he cha ging p ocess
consis ing o cons an cu en un il 4.2 V, ollowed by cons an ol age un il eaching he
cu o cu en . Di e en discha ge egimes a e adop ed. The second da ase co esponds
o 28 li hium-ion 18,650 cells wi h a capaci y o 2.2 Ah ha a e con inuously cycled wi h
andomly gene a ed cu en p o iles. Re e ence cha ge and discha ge cycles a e also
pe o med a e a andom ixed in e al. In o al, he cells a e di ided in o se en equal
g oups, wi h he cycles occu ing a a empe a u e o 40
◦
C. In i e g oups, he cha ging
cycle ollows he adi ional cons an cu en –cons an ol age (CC-CV) pa e n, ollowed
by andomly selec ed discha ges. In wo g oups, bo h he cha ging and discha ging
p ocesses a e selec ed andomly. The cycling p ocesses o he cells a e e mina ed when
hei capaci y eached ei he 80% o 50% o he ini ial capaci y, depending on he ype o
es de ined. Bo h NASA da ase s a e p o ided in “.ma ” ex ension iles.
Ene gies 2025,18, 746 32 o 77
Figu e 17. F equency o g oups o ML algo i hms p esen ed in he BP.
Excluding he use o neu al ne wo ks, ee-based me hods ha e gained conside able
ep esen a ion in he po olio. No ably, ensemble me hods, such as boos ing, we e em-
ployed in [
83
,
272
,
399
], along wi h implemen a ions o popula boos ing algo i hms, like
XGBoos in [
262
,
400
,
401
], Ligh GBM in [
402
–
404
], and Ca Boos in [
405
,
406
], which ha e
gained p ominence in he ield o abula da a p edic ion in ecen s udies. The use o
bagging can be iden i ied in he implemen a ion o decision- ee ensembles, such as andom
o es , as explo ed in s udies [
47
,
161
,
407
]. Ke nel-based me hods, such as SVMs, can be
classi ied as algo i hms belonging o a classical and da ed app oach [
408
], ye hey we e
conside ably analyzed in he po olio in s udies [
323
,
409
,
410
]. The use o classical and
highly in e p e able linea eg ession was explo ed in 45 s udies, among which, no able
s udies include hose p esen ed in [1,80,146,238,321,411,412].
Ano he app oach o ela i e impo ance co esponds o algo i hms ha a e a pa
o s a is ical me hodologies, whe e, ou o he 43 implemen a ions in he po olio,
37 co esponded
o he use o he GPR algo i hm, wi h examples o implemen a ions and
analyses ound in [
329
,
369
,
413
–
415
]. As p esen ed below, he GPR algo i hm demons a ed
signi ican usage in hyb id me hodologies, anking six h in usage wi hin he po olio
when conside ing hyb idized algo i hms. Ano he poin wo h no ing is he in e p e a ion
conduc ed by s udies ha sough o analyze deg ada ion h ough classical ime-se ies
app oaches, as p esen ed in [
79
], which implemen s he ARIMAX (Au oReg essi e In e-
g a ed Mo ing A e age Model wi h eXogenous inpu ) me hod, and in s udy [
416
] using he
ARIMA (Au oReg essi e In eg a ed Mo ing A e age Model) me hod. The NAR (Nonlinea
Au o eg essi e) model is explo ed in [64].
The e olu ion o algo i hmic ca ego ies h oughou he ho izon comp ising he biblio-
g aphic po olio is p esen ed in Figu e 18. I is wo h no ing ha he au ho s consis en ly
ocused on explo ing neu al ne wo k implemen a ions h oughou he en i e ime ho izon,
wi h he di e ence om o he ca ego ies main aining a g owing p o ile. I is possible o
obse e a signi ican inc ease in he implemen a ion o decision- ee-based algo i hms om
2022 o 2023. The implemen a ion o linea models has also been gaining momen um,
mainly because o he compa isons ha hese simple models can o e compa ed o mo e
complex algo i hms. Addi ionally, hey p esen g ea e in e p e abili y o a iables and,
he e o e, o he modeling [
6
,
104
,
417
]. The use o ime-se ies echniques has emained
ela i ely cons an in he po olio, which, in con as o he inc easing olume o publica-
ions pe yea , indica es ha he pe cen age o implemen a ion compa ed o ha o o he
Ene gies 2025,18, 746 33 o 77
ca ego ies has been dec easing. Algo i hms ela ed o clus e ing, quan ile eg ession, he
neighbo hood me hod, and unsupe ised lea ning we e mo e ecen ly implemen ed wi hin
he po olio, be ween 2022 and 2023.
Figu e 18. E olu ion o algo i hmic implemen a ion in he BP by ca ego y.
Going deepe in o he analysis o he wo main ca ego ies o algo i hms implemen ed
in he po olio, he g aphs in Figu e 19 depic he dis ibu ions o neu al ne wo k and
decision- ee algo i hms. In he neu al ne wo k ca ego y, ollowing he obse a ions om
he o e all analysis, he e is a dominance o LSTM and CNN ne wo ks, ollowed by simple
neu al ne wo ks, algo i hms based on well-known ne wo ks, such as ex eme-lea ning
machines and RNNs. In he decision- ee algo i hms, he e is a p edominance o ensemble
bagging using he andom o es algo i hm, accoun ing o 35% o he ee implemen a ions
in he po olio, ollowed by boos ing algo i hms, such as XGBoos , GBT, LGBM, and
Adaboos . A de ailed explo a ion o his ype o ensemble can be obse ed in ba e y
deg ada ion s udies, wi h app oxima ely 60% o he ee implemen a ions in he po olio.
(a)
(b)
Figu e 19. F equencies o implemen ed algo i hms: (a) decision ees; (b) neu al ne wo ks.
When analyzing he e olu ion o echniques implemen ed in he po olio, as depic ed
in Figu e 20, i is no iceable ha he use o LSTM ne wo ks p edomina es h oughou almos
he en i e analyzed ime ho izon. The use o CNN ne wo ks began o gain p ominence om
publica ions in 2021. The use o he andom o es became he hi d mos implemen ed
echnique in he po olio’s wo ks in 2023; howe e , he use o simple ANNs showed a
sha p decline in he las yea . Implemen a ions o he SVM me hod seem o be decele a ing,
wi h a dec ease in usage in 2021 and main aining he numbe o implemen a ions in 2023
Ene gies 2025,18, 746 34 o 77
compa ed o 2022. The use o GRU ne wo ks also appea s o be ending, becoming he
ou h mos implemen ed algo i hm in 2023. As a baseline and compa a i e algo i hm,
linea eg ession also demons a es an inc ease in he numbe o implemen a ions o e he
ho izon. O he algo i hms ha seem o be expe iencing a g owing explo a ion a e DNN,
GPR, and XGBoos .
Figu e 20. E olu ion o algo i hmic implemen a ion in he bibliog aphic po olio.
The e olu ion o he po olio’s main neu al ne wo k implemen a ions is p esen ed in
he g aph in Figu e 21. As p e iously highligh ed, he use o LSTM and CNN ne wo ks
is a he o e on o au ho s’ esea ch in he ield, wi h LSTM ne wo ks being he main
echnique in his ca ego y om 2020 onwa d, and CNNs gaining p ominence om 2021. I
is no ewo hy o highligh some ecen jumps in implemen a ions in he po olio, om
2022 o 2023, such as he explo a ion o GRU, DNN, ELM, and BPNN echniques. I is
s iking o see he esu gence o he explo a ion o mo e classical ne wo ks, such as BPNNs,
by au ho s in he ield.
Figu e 21. E olu ion o neu al ne wo k algo i hmic implemen a ion in he BP.
Because o i s seconda y p ominence in he po olio, we also p esen he e olu ion o
decision- ee-based algo i hms in Figu e 22. The e olu ion o he andom o es algo i hm’s
usage o e he yea s can be obse ed, wi h a no able inc ease in 2023, and he possible
eplacemen o GBT boos ing by newe e sions, such as XGBoos and Ligh GBM.
Ene gies 2025,18, 746 35 o 77
Figu e 22. E olu ion o decision- ee-based algo i hmic implemen a ion in he po olio.
3.4.1. Deep-Lea ning Models
ML, as a sub ield o a i icial in elligence, employs algo i hms and s a is ical echniques
o cons uc p edic i e models. Neu al ne wo ks ep esen a subse o ML algo i hms ha
ha e seen hei s uc u al complexi y inc ease o e ime, in andem wi h compu a ional
ad ancemen s. This complexi y p ima ily mani es s in he augmen a ion o in e media e
laye s wi hin ne wo ks, enhancing he algo i hm’s abili y o disce n pa e ns and gi ing
ise o a sub ield known as DL algo i hms [
418
,
419
]. T adi ional ML algo i hms o en
ou pe o m DL me hods in scena ios o limi ed da a a ailabili y. Howe e , as da ase s
expand, adi ional ML algo i hms end o each pe o mance pla eaus, while DL algo i hms
demons a e signi ican supe io i y o e o he lea ning s a egies [418].
The expec ed po en ial o DL echniques can be obse ed in he bibliog aphic po olio.
Wi hin he se o echniques belonging o neu al-ne wo k-based algo i hms, 307 pape s
using DL algo i hms we e iden i ied, ep esen ing a signi ican 57.5% o he po olio. This
demons a es a s ong end wi hin his esea ch ield. DL algo i hms we e conside ed
as hose wi h mo e han h ee hidden laye s. Al hough he e is no consensus among
au ho s and esea che s in he ield ega ding he exac numbe o laye s equi ed o
cha ac e ize a ne wo k as DL, some e e ences conside his numbe o laye s o indica e
“ligh ” DL ne wo ks, while “hea y” DL ne wo ks can ha e om ens o hund eds o
hidden laye s [420,421].
A compa a i e analysis o he e olu ion o he p opo ion o DL usage in neu al
ne wo k algo i hms is shown in Figu e 23. The e is no iceable s abili y in he p opo ion,
discoun ing he ac o o publica ion olume in he ea ly yea s, which se les be ween 80%
and 90% in he las 3 yea s o he po olio.
Ene gies 2025,18, 746 36 o 77
Figu e 23. P opo ion o DL implemen a ion in neu al ne wo k echniques in he po olio.
The olume o DL echnique implemen a ions in he po olio is p esen ed in Table 7,
wi h a isual p opo ion o e iew shown in Figu e 24. In Table 7, he e m “F equency”
e e s o he numbe o implemen a ions eco ded o each echnique. Pape s in he
po olio may p esen mo e han one implemen a ion wi hin he same s udy. Toge he ,
LSTM and CNN echniques accoun o 60% o he implemen a ions, while RNN, DNN,
and GRU echniques s and ou , wi h implemen a ions in mo e han 20 pape s each. In o al,
23 echniques we e implemen ed only once.
Table 7. Su ey o DL echniques implemen ed by au ho s in he bibliog aphic po olio.
Algo i hm F equency Algo i hm F equency Algo i hm F equency
LSTM 161 RESNET 2 PKNN 1
CNN 86 ELM 2 DBNN 1
RNN 29 BPNN 2 DSMTNET 1
DNN 27
EFFICIENTNET
1 DCN 1
GRU 26 CRNN 1 DDAN 1
ANN 10
VISION
TRANS-
FORMER
NETWORK
1
DEEP REIN-
FORCE-
MENT
LEARN-
ING
1
MLP 8 VGG11 1 DELM 1
TCN 6
TRANSFORMER
NEURAL
NETWORK
1
GOOGLENET
1
ENN 5 TDNN 1
DENSENET
1
FFNN 5 BNN 1 ALEXNET 1
GRAPH
NN 4 CAPSNET 1 EDFM 1
DBN 3
REGRESSIVE
MATCH-
ING
NETWORK
1
DILATED
RESIDUAL
NETWORK
1
DCNN 3
CDTSGANN
1 FCNN 1
Ene gies 2025,18, 746 37 o 77
Figu e 24. Dis ibu ion o echniques in DL implemen a ions in he po olio.
The e olu ion o he main DL algo i hms implemen ed in he po olio is p esen ed
in Figu e 25, which shows ha he mos implemen ed algo i hm is he LSTM ne wo k.
In addi ion o he ising ends o LSTM and CNN ne wo ks, we again highligh he
ecen implemen a ion ends o DNN and GRU algo i hms, as well as he i s ele an
implemen a ions o he DCNN algo i hm, ound in 2023, which combines cha ac e is ics o
DNN and CNN ne wo ks.
Figu e 25. E olu ion o DL algo i hmic implemen a ion in he BP.
In [
44
], which is he main DL publica ion acco ding o he ci a ion coun , he au ho s
employ a hyb id LSTM-RNN model o cap u e long- e m in o ma ion ega ding he ela-
ionship be ween a ba e y’s capaci y and i s deg ada ion, emphasizing ha such a dual
app oach is ecommended o a oid o e i ing issues. Ano he wo k u ilizing a hyb id
echnique based on LSTM is p esen ed in [
48
], using an Elman neu al ne wo k (ENN).
The concep o ans e lea ning, which in ol es he use o neu al ne wo ks ained and
ine- uned in la ge da ase s and hen ine- uned on hei inal laye s in a speci ic da ase o
ans e knowledge o ano he p oblem, is discussed in [
65
], which implemen s an LSTM
ne wo k. O he examples o s udies using LSTMs can be ound in [51,59,422,423].
Rega ding ele ance by ci a ion coun , he main s udies using DL a e p esen ed in
Table 8, whe e i is no able ha he use o LSTM is p esen in six ou o he en s udies.
Ano he in e es ing poin is he use o a hyb id app oach by he publica ions, using wo
DL algo i hms in his case. The able also highligh s he da ase s used by he au ho s, wi h
hal he publica ions u ilizing public da a Addi ionally, he able includes a ma king o
Ene gies 2025,18, 746 38 o 77
indica e whe he he implemen a ion was hyb id, whe e mo e han one algo i hm was
used o de e mine he same p edic ion.
Table 8. Main publica ions in he po olio wi h DL implemen a ion.
Algo i hm Hyb id Da ase Ti le Yea Ci ed Re .
LSTM,
RNN Yes Au ho
Long sho - e m memo y ecu en neu al
ne wo k o emaining-use ul-li e p edic ion
o li hium-ion ba e ies
2018 880 [44]
LSTM,
GPR Yes Au ho
A da a-d i en app oach wi h unce ain y
quan i ica ion o p edic ing u u e
capaci ies and emaining use ul li e o
li hium-ion ba e ies
2021 434 [45]
LSTM,
ENN Yes Au ho
Remaining-use ul-li e p edic ion o
li hium-ion ba e ies based on a hyb id
model combining he long sho - e m
memo y and Elman neu al ne wo ks
2019 316 [48]
DNN No NASA
Remaining-use ul-li e p edic ion o
li hium-ion ba e ies: A deep-lea ning
app oach
2018 313 [49]
CNN,
LSTM Yes NASA
A da a-d i en au o-CNN-LSTM p edic ion
model o li hium-ion-ba e ies’ emaining
use ul li e
2021 291 [50]
LSTM No NASA
S a e-o -heal h es ima ion and
emaining-use ul-li e p edic ion o
li hium-ion ba e ies based on a a ian long
sho - e m memo y neu al ne wo k
2020 284 [51]
DCNN No Au ho A deep-lea ning me hod o online capaci y
es ima ion o li hium-ion ba e ies 2019 260 [53]
DNN No
CALCE,
NASA,
MIT,
OXFORD
Machine-lea ning pipeline o ba e ies’
s a e-o -heal h es ima ions 2021 246 [54]
LSTM No NASA
A neu al-ne wo k-based me hod o RUL
p edic ion and SOH moni o ing o
li hium-ion ba e ies
2019 245 [55]
PKNN No Au ho
A no el es ima ion me hod o he s a es o
heal h o li hium-ion ba e ies using a
p io -knowledge-based neu al ne wo k and
a Ma ko chain
2019 239 [56]
The i e mos ecen s udies wi h DL implemen a ion in he bibliog aphic po olio
a e p esen ed in Table 9. The publica ions co espond o he yea 2024, which, in o al,
had 17 publica ions on he subjec in he i s wo weeks o he yea ( he o al numbe o
po olio publica ions in 2024 was 21). The i e highligh ed pape s make use o public
da ase s (16 ou o 17 in o al o he yea ), wi h wo publica ions implemen ing a hyb id
app oach wi h DL (se en ou o se en een in o al o he yea ), including he use o a
decision- ee-based algo i hm.
Ene gies 2025,18, 746 39 o 77
Table 9. Recen publica ions in he po olio wi h DL implemen a ion.
Algo i hm Hyb id Da ase Ti le Yea Ci ed Re .
GCN No NASA,
OXFORD
S a e-o -heal h and
emaining-use ul-li e p edic ions o
li hium-ion ba e ies wi h a condi ional
g aph con olu ional ne wo k
2024 2 [179]
RNN No MIT
Jelly ish-op imized ecu en neu al
ne wo k o s a e-o -heal h es ima ions
o li hium-ion ba e ies
2024 2 [336]
LSTM No NASA,
CALCE
Remaining-use ul-li e p edic ions o
li hium Ba e ies based on a
CNN–Mog i ie LSTM-MMD
2024 1 [192]
MLP, GRU Yes NASA,
CALCE
An MLP–mixe and mix u e o expe
models o emaining-use ul-li e
p edic ions o li hium-ion ba e ies
2024 0 [220]
RF, GRU Yes NASA
S a e-o -heal h es ima ions o
li hium-ion ba e ies using a andom
o es and a ga ed ecu en uni
2024 0 [221]
3.4.2. Hyb id Models
Hyb id ML models combine di e en ML echniques and algo i hms o enhance
p edic ion pe o mance by le e aging he s eng hs o each me hod while compensa ing
o hei indi idual weaknesses [
48
,
424
]. Wi hin he bibliog aphic po olio, a o al o
135 publica ions
ha e implemen ed his app oach. The e olu ion o hyb id model usage in
he po olio is depic ed in Figu e 26. As shown, i can be in e ed ha he use o hyb id
app oaches in SoH es ima ion is a ecen ield o explo a ion, wi h a signi ican inc ease
in implemen a ions in 2023. Conside ing he olume o po olio publica ions, he use o
hyb id app oaches ep esen s app oxima ely 30% o he pape s su eyed, a jump o nea ly
50% compa ed o 2022, whe e i was p esen in abou 18% o publica ions. In he i s wo
weeks o 2024, a o al o nine pape s wi h hyb id app oaches we e published.
Figu e 26. E olu ion o hyb id algo i hmic implemen a ion in he BP.
Ene gies 2025,18, 746 40 o 77
The mos u ilized echniques in he hyb id modeling app oach also co espond o
he use o DL ne wo ks, such as LSTM and CNN, wi h a conside able ad an age, wi h
74 and 51 implemen a ions, espec i ely, as indica ed in Figu e 27. O he DL algo i hms,
such as GRU and RNN, a e also no able, o example, in [
232
,
233
,
381
,
425
]. O he classical
algo i hms, such as SVM and GPR, ound in [
149
,
426
], and decision- ee-based algo i hms,
like RF, XGBoos , and Ligh GBM, p esen in [80,221,427], a e also no ewo hy.
Figu e 27. F equency o echniques add essed in pape s wi h hyb id algo i hms in he BP.
The e olu ion o he implemen a ions o he main algo i hms is p esen ed in Figu e 28.
I is possible o obse e he inc eases in he implemen a ions o LSTM and CNN ne wo ks,
in line wi h p e ious esul s, and he ecen e olu ion o he use o GRU and RF algo i hms,
wi h a peak in usage in 2023.
Figu e 28. E olu ion o hyb id algo i hmic implemen a ion in he bibliog aphic po olio.
Ene gies 2025,18, 746 41 o 77
Figu e 29 p esen s he ound combina ions esul ing om he analysis o hyb id
algo i hms in he po olio. The p ima y combina ion occu s wi h he LSTM and CNN
ne wo ks, wi h 27 implemen a ions in he po olio. Algo i hms ha appea indi idually in
he su ey e lec ei he a hyb id app oach (e.g., he in eg a ion o di e en con igu a ions
o he same algo i hm, such as combining a 2-dimensional CNN wi h a 3-dimensional
CNN), o me hodologies ha inco po a e il e s (e.g., he Kalman il e ) and op imiza ion
algo i hms as a pa o hei design. In o al, 65 pape s p esen ed combina ions o algo i hms
ha we e implemen ed only once in he po olio, indica ing ha many esea che s s ill
e alua e di e en app oaches o hyb id models.
Figu e 29. Combina ions o algo i hms ound in pape s wi h a hyb id app oach in he BP.
The combina ions o each algo i hm in he po olio a e p esen ed in Table 10, allowing
o he iden i ica ion o hyb id app oaches, e alua ed by he au ho s, wi hin he po olio,
which can se e as a s a ing poin o es ing hyb id models in new esea ch. A isualiza-
ion o hese combina ions is shown in Figu e 30, whe e cen e s o algo i hmic connec ions
can be obse ed, e ol ing a ound LSTM, CNN, RF, and GPR echniques. As demons a ed
in Table 10 and Figu e 30, LSTM ne wo ks exhibi a conside able ange o combina ions wi h
o he algo i hms, including decision- ee, s a is ical, ke nel, neighbo hood, and clus e ing
algo i hms, as well as o he neu al ne wo k a chi ec u es.
Table 10. Connec ions be ween algo i hms in pape s wi h a hyb id app oach in he BP.
Algo i hm Algo i hmic Connec ions
LSTM
DCNN, FFNN, CNN, ANN, ENN, DNN, TCN, RNN,
XGBOOST, BMA, GRAPH NEURAL NETWORK, SVM,
GPR, DBN, GRU, RANDOM FOREST, FUZZY
CLUSTERING, BPNN, LINEAR QUANTILE
REGRESSION, MLP, RESNET, ADABOOST
RANDOM FOREST
ANN, NAR, LINEAR REGRESSION,
GRADIENT-BOOSTING DECISION TREE, GPR,
LIGHTGBM, XGBOOST, LSTM, SVM, RBFNN, RIDGE
REGRESSION, KNN, GRU, EXTRATREES, ELM
SVM
ARIMA, DECISION TREE, ELM, LSTM, GPR, RBFNN,
RANDOM FOREST, RIDGE REGRESSION, LINEAR
REGRESSION, GRU, RNN
Ene gies 2025,18, 746 48 o 77
di icul . In hese cases, he compa isons a e empi ical, and he e is an associa ed p obabili y
o a pa icula app oach being be e han ano he . Among he a gumen s used a e conce ns
ela ed o he quali y o he da a om some cells, as well as supposed bias in he aining
spli based on di e ences in he dis ibu ion o he cycles used in [
1
]. Ano he poin ha
d ew a en ion was cases whe e au ho s pe o med spli s o aining, es ing, and alida ion
while keeping da a samples om all he cells in each se , which impac s he eliabili y o
he esul s p esen ed, as pe he pe o mances in he a icles highligh ed below.
Table 15 summa izes he RUL ( emaining-use ul-li e) p edic ion pe o mances in
publica ions ha had he same alida ion se , o aling i e pape s. The alida ion se s a e
e e ed o as he 1
◦
Tes and 2
◦
Tes by he au ho s in [
1
]. The 1
◦
Tes includes ba e ies
unde he same cycling condi ions as hose o he aining se , while he 2
◦
Tes co esponds
o ba e ies wi h a di e en usage p o ile. No ably, he pe o mance gain achie ed in [
335
]
is highligh ed, whe e he MAPE e o s in bo h es se s a e educed by o e 50% compa ed
o ha o he baseline s udy [
1
]. This imp o emen was achie ed using he same
100 cycles
o in o ma ion o he p edic ion and implemen ing a hyb id model using he LSTM DL
echnique along wi h he GPR algo i hm. Howe e , such signi ican esul s we e no ound
in he use o he LSTM-CNN combina ion in [
333
], whe e a ange o e o s simila o ha o
he baseline s udy [
1
] was obse ed despi e employing mo e complex echniques. In [
317
],
an inc ease in pe o mance is e iden wi h a hyb id app oach in ol ing neu al ne wo ks,
linea eg ession, and RF, using a educed se o 80 cycles. These indings sugges ha
e o s in algo i hmic selec ion do no necessa ily gua an ee highe pe o mance, and s eps
such as ea u e cons uc ion and selec ion may ep esen an e en mo e ele an s age
in esea ch.
In Table 16, he es o he pe o mance su ey wi h he MIT da ase , conduc ed in
he bibliog aphic po olio, is p esen ed. He e, he au ho s did no main ain he same
modeling and alida ion spli s, and he e a e a ia ions in he a ge a iables. The e o e,
all he compa isons made may exhibi signi ican bias. The a ge s desc ibed in he able a e
p esen ed o main ain he nomencla u e adop ed by he au ho s. The a ge ’s “ea ly ba e y
li e ime”, also e e ed o in s udies as he “ea ly cycle li e”, aims o de e mine he o al
numbe o cycles a ba e y will p esen based on da a om he i s cycles o a ba e y. The
“end o li e” in he analyzed s udies is ela ed o de e mining he o al numbe o cycles
conside ing da a om he las cycles, wi hou necessa ily knowing he en i e ba e y his o y.
The e o e, i is gene ally accompanied by models ha de e mine he emaining numbe
o cycles (RULs) and/o he cu en cycle. The “capaci y” a ge was linked o s udies
ha used eg ession models in p edic ing ime se ies (“capaci y ajec o ies”), whe e he
e olu ion o he ba e y capaci y o e ime is ob ained o he p edic ion on a sho ho izon,
such as he discha ge capaci y in he nex cycle, can be used o SoH upda es.
Ene gies 2025,18, 746 49 o 77
Table 15. P edic ion pe o mance o s udies using he same samples o alida ion wi h MIT da ase .
Algo i hm 1◦Tes 2◦Tes Cycles Ti le Yea Ci ed Re .
RMSE MAPE RMSE MAPE
Linea
eg ession 118 14.1 214 10.7 100 Da a-d i en p edic ion o he ba e y cycle li e be o e
capaci y deg ada ion 2019 811 [1]
RF, linea
eg ession,
ANN
80 9.8 174 7.5 80 P ognos ics o he ba e y cycle li e in he ea ly-cycle s age
based on a hyb id model 2021 41 [317]
Ridge Reg 125
-
188
-100 S a is ical lea ning o accu a e and in e p e able ba e y
li e ime p edic ions 2021 30 [102]
Ene Reg 132 196
RF 141 197
MLP 140 218
CNN 72 204
CNN, MLP 114 8.54 178 11.31 100 A hyb id ensemble deep-lea ning app oach o he ea ly
p edic ion o ba e ies’ emaining use ul li e 2023 9 [333]
GPR, LSTM 30 5.52 51 5.35 100 Join modeling o ea ly p edic ions o Li-ion-ba e ies’
cycle li e and deg ada ion ajec o y 2023 3 [335]
Table 16. P edic ion pe o mance o s udies in he po olio using di e en samples o alida ion wi h he MIT da ase .
Algo i hm Ta ge MAPE RMSE RMSPE MAE MRE R2Obse a ions Re . Yea
Bayesian idge
Capaci y
(Ah)
0.45 0.76
- P edic ing capaci y conside ing only a sho
po ion o pa ial cha ge/discha ge da a
- Requi es a 15 min sample o ope a ion
- U ilizes cha ging and discha ging s eps
- 63 cells o aining, 10 o calib a ion, 51 o
es ing (spli based on he dis ibu ion o cycle
numbe s in he da ase , main aining he same
dis ibu ion ac oss all he se s)
[54]2021
GPR 1.00 1.91
RF 0.11 0.14
DNN 0.23 0.45
CNN RUL 10.6 76
- U iliza ion o ou cycles
- Inco po a ion o cha ging and discha ging s eps
- 86 cells o aining, 19 o alida ion, 19 o es ing [8] 2020
Ene gies 2025,18, 746 50 o 77
Table 16. Con .
Algo i hm Ta ge MAPE RMSE RMSPE MAE MRE R2Obse a ions Re . Yea
GBT RUL 7.5 84.9 58.6 0.94
- Usage o 250 cycles
- Inco po a ion o he discha ge s age
-
Da a spli in o 2/3 o aining and 1/3 o es ing;
no speci ica ion i cells om he aining se a e
excluded om he es se ; he spli ing p ocess is
epea ed ou imes, and he pe o mances
a e analyzed
- Pe o mance co esponds o he a e age o he
ou cases
[83] 2020
CNN Ea ly ba e y
li e ime
3.80 (1) 42 (1) 33 (1) - Tes ing o he use o he i s 20 (1), 40 (2), 60 (3),
80 (4), 100 (5) cycles o ba e y li e p edic ion
- U iliza ion o he i s i e cycles and he las
i een cycles o RUL p edic ion
- Inco po a ion o he cha ging s age
- 94 cells o aining, 30 o es ing
[315]2021
1.30 (2) 19 (2) 13 (2)
1.12 (3) 13 (3) 11 (3)
1.21 (4) 13 (4) 10 (4)
1.12 (5) 11 (5) 9 (5)
RUL 3.55 11 9
DNN
End o li e 7.78 (1) 57 (1)
-
Tes ing o he use o he las 1 (1) o 100 (2) cycles.
- Inco po a ion o he discha ging s age
- EoL = cu en cycle + cycles used o da a
collec ion + RUL.
- 65 cells o aining, 16 o es ing (disca ding
43 cells)
- Majo i y o RMSE o RUL < 50 cycles, la ge
e o s o cells wi h ewe han 100 cycles
- * E o s o p edic ing he cu en li e cycle
inc ease in ini ely o cells wi h o e 750 cycles
(au ho ’s jus i ica ion based on he low sample
quan i y o his scena io)
[318]2022
3.97 (2) 33 (2)
Cycle li e <65 (1)
<40 (2)
>90 *
RUL <65 (1)
<40 (2)
Ene gies 2025,18, 746 51 o 77
Table 16. Con .
Algo i hm Ta ge MAPE RMSE RMSPE MAE MRE R2Obse a ions Re . Yea
SVR Capaci y
ajec o y
(Ah)
1.61 3.22
- Use o he las 20 cycles
-
Es ima es he e olu ion o capaci y ajec o y o e
ime un il EoL ( ime se ies using eg ession)
- Inco po a ion o bo h cha ging and
discha ging s ages.
- 84 cells o aining, 40 o es ing
[253]2022
RF 0.93 2.12
GPR 1.35 2.58
ANN 1.13 1.92
CNN RUL 4.15 27.47 16.09
- Use o 10 cycles
- Inco po a ion o he cha ging s age
-
70% o he da a o aining, 30% o es ing (does
no speci y i cells om he aining se we e
excluded om es ing)
[319] 2022
Linea eg, (1)
RUL
90 53.81 *
- Use o 10 cycles
- Does no exclude cells om aining du ing
es ing; 60% o he da a o aining, 20% o
alida ion, and 20% o es ing
- Inco po a ion o he cha ging s age
-
Classi ica ion model o p edic i a ba e y has less
han 150 cycles o RUL o 150 cycles o mo e
o RUL
- RUL App oaches:
- (1): Does no conside he classi ica ion model
- (2): Reg ession model o each p edic ed RUL
class in he classi ica ion model
- * Conside ing cases whe e RUL > 150 cycles:
18.51, 10.51, 9.79, espec i ely
- Fo capaci y, he au ho e alua ed 100 **, 150 ***,
and 200 **** cycles ahead
[321]2021
MLP (1) 52 23.03 *
Logis ic eg. +
MLP (2) 49 15.2 *
MLP
Discha ge
capaci y a e
“x” cycles.
0.24 **
0.45
***
0.64
****
Ene gies 2025,18, 746 52 o 77
Table 16. Con .
Algo i hm Ta ge MAPE RMSE RMSPE MAE MRE R2Obse a ions Re . Yea
T ans e
Lea ning
(CNN + RNN +
“ ully
connec ed”)
Capaci y
(Ah)
0.176 * 2.57 * 0.999 *
- Use o he las 30 cycles o p edic he capaci y o
he nex cycle and RUL
- The au ho does no assess he pe o mance o
es ima ing he capaci y ajec o y o ho izons
longe han one cycle
- Use o he cha ging s age
- Use o he MIT da ase o ain a model and
e alua e he pe o mance o he model wi h
ans e lea ning on a da ase cons uc ed by
he au ho
- Au ho ’s da ase con ains in o ma ion om
77 LFP/g aphi e cells o 1.1 Ah.
- 22 cells sepa a ed o es ing
- * Pe o mance conside ing aining wi h he
au ho ’s da ase .
-
** Pe o mance conside ing ans e lea ning om
a model p e ained wi h he MIT da ase
[385]2022
0.328 ** 4.65 ** 0.997
**
RUL 8.72 * 186 * 0.804 *
9.80 ** 240 ** 0.770
**
Elas ic ne
RUL
5.21 43.38 0.98
- Use o 100 cycles
- Use o bo h cha ging and discha ging s ages
-
No exclusion o cells om aining in es ing; 70%
o he da a o aining, 30% o es ing
[329]2022
GPR 5.26 43.71 0.98
SVM 5.88 53.04 0.97
RF 8.17 84.69 0.92
DT ensemble 7.93 88.74 0.91
XGBoos 7.92 91.13 0.92
RVM 10.32 96.21 0.89
DT 9.59 106.62 0.87
Ene gies 2025,18, 746 53 o 77
Table 16. Con .
Algo i hm Ta ge MAPE RMSE RMSPE MAE MRE R2Obse a ions Re . Yea
CNN, LSTM
Cycle li e
2.28 (1) 19 (1) 14 (1) 0.9980
(1)
- Tes ing o usage om he las 50 (1), 60 (2), 70 (3),
80 (4), 90 (5), 100 (6) cycles
- Usage o bo h cha ging and discha ging s ages
-
93 cells o aining and 31 o es ing (spli based
on he dis ibu ion o cycle numbe s in he
da ase , main aining he same dis ibu ion in all
he se s)
[327]2022
4.59 (2) 50 (2) 33 (2) 0.9869
(2)
3.02 (3) 25 (3) 18 (3) 0.9967
(3)
3.43 (4) 25 (4) 19 (4) 0.9967
(4)
1.84 (5) 16 (5) 13 (5) 0.9985
(5)
1.47 (6) 11 (6) 9 (6) 0.9993
(6)
RUL
2.16 (1) 12 (1) 8 (1) 0.9993
(1)
3.17 (2) 15 (2) 12 (2) 0.9989
(2)
1.93 (3) 11 (3) 8 (3) 0.9994
(3)
1.85 (4) 14 (4) 10 (4) 0.9990
(4)
1.72 (5) 13 (5) 9 (5) 0.9992
(5)
1.25 (6) 8 (6) 6 (6) 0.9997
(6)
G aph Neu al
Ne wo k
Capaci y
ajec o y
(Ah)
0.009 * 0.0377
*
0.9399
*
- Using 350 measu emen poin s as inpu .
- Usage o he cha ging s age.
- 70% o he cells used o aining and 30%
o es ing.
-
Es ima es he e olu ion o capaci y ajec o y o e
ime un il End o Li e ( ime se ies using
eg ession).
- * Pe o mance based on he wo s p edic ed cell.
- ** Pe o mance based on he bes p edic ed cell.
[92]2023
0.004 ** 0.0025
**
0.9894
**
Ene gies 2025,18, 746 54 o 77
Table 16. Con .
Algo i hm Ta ge MAPE RMSE RMSPE MAE MRE R2Obse a ions Re . Yea
Ligh GBM
SoH (%)
1.751 - Es ima ion o SoH based on 300 s measu emen s
- Usage o he discha ge s age
- Cells 91 and 100 om he MIT da ase a e used
o aining, cell 124 used o es ing
- * Model conside ing Ligh GBM, XGBoos ,
Random Fo es (RF), SVR, GPR as base models,
and linea eg ession as a me a-model
[99]2023
XGBoos 1.616
RF 1.721
SVR 1.926
GPR 1.539
S acking 1.489 *
LSTM
SoH (%) a e
“x” cycles
0.016 (1)
1.81 (1) 0.0098
(1)
- Tes ing p edic ion ho izons o 25 (1), 50 (2), 100
(3), 150 (4), 200 (5), 250 (6), 300 (7), 350 (8), 400 (9)
cycles ahead
- Usage o cha ge and discha ge s ages
- 64% o he cells a e used o aining, 20% o
alida ion, and 16% o es ing (cells canno be in
mo e han one se )
- A e age o pe o mances pe cell
[144]2023
0.021 (2)
2.30 (2) 0.0130
(2)
0.024 (3)
2.80 (3) 0.0140
(3)
0.024 (4)
2.86 (4) 0.0120
(4)
0.031 (5)
3.60 (5) 0.0180
(5)
0.026 (6)
3.00 (6) 0.0150
(6)
0.030 (7)
3.49 (7) 0.0200
(7)
0.032 (8)
3.70 (8) 0.0200
(8)
0.033 (9)
3.80 (9) 0.0201
(9)
Ene gies 2025,18, 746 55 o 77
Table 16. Con .
Algo i hm Ta ge MAPE RMSE RMSPE MAE MRE R2Obse a ions Re . Yea
RF Cycle li e 0.57 4.65
- Usage o 100 cycles.
- U iliza ion o cha ge and discha ge s ages.
- 75% o he da a used o aining, 25% used o
es ing (cells om he aining se we e no
excluded om es ing).
- Disc epan esul s compa ed o he li e a u e,
po en ial model alida ion e o by he au ho .
[334] 2022
ResNe 50
Ea ly
li e ime
119.98 0.8501 - Use o he i s 100 cycles o p edic ing he o al
li espan.
- U iliza ion o images ob ained om plo s wi h
ol age and capaci y in o ma ion as ea u es.
- 80 cells o aining and 43 cells o es ing,
p ocess epea ed i e imes.
[111]2024
CNN 115.85 0.8557
LeNe 129.77 0.8197
AlexNe 91.51 0.9121
VGG16 122.19 0.8466
Ene gies 2025,18, 746 56 o 77
Th ough a compa ison analysis, i is possible o obse e pe o mance imp o emen s
compa ed o he baseline s udy in [
315
], wi h a MAPE o a ound 3.5% using a CNN. In [
8
],
he au ho s achie ed signi ican pe o mance wi h a CNN and only ou cycles o da a
as inpu o he algo i hm, a educ ion ha p o ides new pe spec i es o he use and
condi ioning o ba e ies. Using TL, he au ho s in [
385
] achie ed esul s wi hin he e o
magni udes o he s udies ha use only one da ase , demons a ing ha TL can be a use ul
ool o agg ega ing he olume o in o ma ion om o he ypes o expe imen al es s and
cell echnologies o o e come da a limi a ions. In [
327
], he bes esul s o RUL p edic ion
we e achie ed, wi h MAPE alues below 2%. The au ho s conduc ed es s conside ing
di e en da a usage in e als, anging om 50 o 100 cycles, and using a hyb id LSTM-
CNN app oach. Con e sely, in [
334
], he au ho s claim e o s a ound hal a cycle, well
below hose p esen ed in a ious consul ed s udies. The use o da a om all he cells
du ing aining ends up b inging a possible leakage when alida ing he algo i hm because
he pa e n o all he cells was passed o he model, which, consequen ly, did no de elop
p ope lea ning bu possible “ o e memo iza ion”, associa ing le els o a iable alues wi h
ela ed li e cycles. This poin highligh s he impo ance o co ec ly analyzing he esul s
o he dissemina ion o esea ch in he ield.
This analysis e eals signi ican challenges in compa ing RUL p edic ion models
because o inconsis encies in da a spli ing, a ge a iables, and e alua ion me ics ac oss
di e en s udies. Al hough ad ancemen s ha e been obse ed, pa icula ly wi h hyb id
models, like LSTM-GPR and he applica ion o TL, he lack o s anda dized me hodologies
hinde s di ec compa isons and hinde s he iden i ica ion o uly supe io app oaches.
The use o limi ed da a cycles in aining, as demons a ed in [
8
], and he explo a ion
o ea u e enginee ing, as sugges ed by he esul s in [
317
], p esen p omising a enues
o u u e esea ch. Howe e , i is c ucial o emphasize he impo ance o igo ous da a
spli ing p ocedu es, a oiding da a leakage, as obse ed in [
334
], o ensu e he eliabili y
and gene alizabili y o he ob ained esul s. No ably, o mos o he analyzed models,
MAPE e o s o a ound 10% ha e become achie able wi h he de elopmen o algo i hms
and open da ase s. Mo ing o wa d, es ablishing s anda dized da ase s and e alua ion
p o ocols will be essen ial o acili a e p og ess in he ield and enable mo e meaning ul
compa isons be ween di e en RUL p edic ion models.
3.6. The Impo ance o SoH in Sma Sys ems, Ene gy In o ma ics, and Sma G ids
Accu a e ba e y SoH es ima es, de i ed om ML algo i hms and analyses based on
la ge da ase s, ha e signi ican implica ions o ene gy in o ma ics and in elligen sys ems,
such as sma g ids. This s udy explo es some o he key applica ions connec ing SoH
p edic ion o imp o emen s in ene gy e iciency and sus ainabili y.
Ene gy In o ma ics and Ene gy Managemen in Sma G ids: Ene gy in o ma ics, he
in eg a ion o in o ma ion sys ems and ene gy, plays a undamen al ole in he e icien
managemen o sma g ids. Accu a e SoH es ima ion enables mo e e ec i e managemen
o second-li e ba e ies by in eg a ing hem in o s o age and dis ibu ion ne wo ks. This
app oach no only educes was e bu also enhances he eliabili y and esilience o elec ical
g ids, especially in con ex s in ol ing enewable ene gy sou ces [1,10,12].
IoT De ices and Sus ainabili y: SoH p edic ion models based on DL echniques, such
as LSTM and CNN ne wo ks, acili a e he p e en i e main enance o IoT de ices ha ely
on ba e ies. These models suppo a mo e sus ainable economy by op imizing eplacemen
cycles and ex ending he li espans o connec ed sma de ices [
48
,
54
,
64
]. The use o hese
de ices in sma g ids also educes eliance on manual in e en ions, p omo ing g ea e
au oma ion and e iciency [9,12].
Ene gies 2025,18, 746 57 o 77
Real-Time Moni o ing and Con ol: Wi eless senso ne wo ks (WSNs) in eg a ed
wi h SoH algo i hms o e eal- ime moni o ing capabili ies, essen ial o dynamic sys em
adjus men s. In sma g ids, his enables load balancing and he op imiza ion o he ene gy
dis ibu ion, imp o ing he o e all sys em pe o mance [54].
En i onmen al Impac and Sus ainabili y: The euse o ba e ies, unde pinned by
eliable SoH es ima es, con ibu es o a ci cula economy by educing he en i onmen al
impac and he ca bon oo p in associa ed wi h he p oduc ion o new ba e ies [
9
,
12
].
Hyb id models, such as he combina ion o LSTM wi h Elman neu al ne wo ks (ENNs),
ha e al eady demons a ed accu acy gains o up o 50% compa ed o classical app oaches,
inc easing con idence in ba e y euse o s o age sys ems and sma g ids [48].
Th ough hese applica ions, SoH p edic ion no only enhances he managemen and
e iciency o sma g ids bu also ein o ces he connec ions among ene gy in o ma ics,
sus ainabili y, and echnological inno a ion. This highligh s he impo ance o obus
p edic ion me hods o he u u e o in elligen ene gy sys ems.
Addi ionally, i is essen ial o emphasize ha accu a e SoH p edic ion signi ican ly
con ibu es o he e olu ion o in elligen sys ems by educing ope a ional unce ain ies
and enabling he seamless in eg a ion o eme ging echnologies. The p ecise o ecas ing o
SoH enhances sys em eliabili y by enabling he op imized alloca ion o ene gy esou ces,
such as second-li e ba e ies, ac oss di e se use cases. These ad ancemen s also suppo
he adop ion o p edic i e main enance sys ems, which educe ope a ional cos s while max-
imizing ene gy e iciency and long- e m sus ainabili y. By con e ging machine-lea ning
echniques, such as deep neu al ne wo ks, wi h ad anced da a managemen pla o ms,
SoH becomes a c i ical me ic o decision-making in sma g ids and he IoT, d i ing
esilience and sus ainabili y in ene gy in as uc u e.
4. Conclusions
This s udy highligh s he g owing impo ance o ML echniques in es ima ing he SoHs
o ba e ies, as e idenced by a sys ema ic bibliog aphic po olio analysis. The applica ion
o P oKnow-C enabled he objec i e selec ion o 534 ele an pape s om an ini ial pool o
6032 publica ions, p o iding a s uc u ed and eplicable me hodology o cha ac e izing
esea ch wi hin his domain.
The esul s e eal se e al key ends. Fi s , he e has been a signi ican inc ease in
scien i ic p oduc ion in his a ea, pa icula ly since 2022, wi h 40% o he selec ed pape s
published in 2023. The inc easing ele ance o ba e y euse, d i en by he expansion o he
elec ic ehicle ma ke , is expec ed o u he boos esea ch in SoH es ima ion. Second, he
analysis highligh s he impo ance o open da ase s, wi h 60% o he e iewed s udies using
publicly a ailable da a. The NASA P ognos ics Cen e o Excellence eposi o y emains he
mos ci ed sou ce, accoun ing o o e hal o he open da a usage. O e all, he po olio
analysis e ealed he p esence o 12 a ailable open da a sou ces, wi h 6 o hese sou ces
published in he yea s 2022 and 2023.
F om a me hodological pe spec i e, DL echniques, especially LSTM ne wo ks and
CNNs, domina e he ield, wi h DL accoun ing o 58% o he implemen a ions. Hyb id
app oaches, including hose combining LSTM and CNNs, a e inc easingly p ominen , ep-
esen ing app oxima ely 25% o he e iewed s udies. The eme gence o TL in publica ions
since 2022 also highligh s a p omising a enue o le e aging di e se da ase s o add ess
da a sca ci y and he e ogenei y in SoH modeling.
Pe o mance e alua ions based on he MIT da ase indica e ha classical app oaches
achie e mean absolu e pe cen age e o s o app oxima ely 10%, whe eas DL echniques
ha e educed e o s by 50% in some cases. Some s udies epo p edic ion e o s as low as
1–4% using CNNs, emphasizing he po en ial o ad anced algo i hms in his ield.
Ene gies 2025,18, 746 64 o 77
80.
Li, F.; Zhang, L.; Chen, B.; Gao, D.; Cheng, Y.; Zhang, X.; Yang, Y.; Gao, K.; Huang, Z. An Op imal S acking Ensemble o
Remaining Use ul Li e Es ima ion o Sys ems Unde Mul i-Ope a ing Condi ions. IEEE Access 2020,8, 31854–31868. [C ossRe ]
81.
Liu, G.; Zhang, X.; Liu, Z. S a e o Heal h Es ima ion o Powe Ba e ies Based on Mul i-Fea u e Fusion Models Using S acking
Algo i hm. Ene gy 2022,259, 124851. [C ossRe ]
82.
Paulson, N.H.; Kubal, J.; Wa d, L.; Saxena, S.; Lu, W.; Babinec, S.J. Fea u e Enginee ing o Machine Lea ning Enabled Ea ly
P edic ion o Ba e y Li e ime. J. Powe Sou ces 2022,527, 231127. [C ossRe ]
83.
Yang, F.; Wang, D.; Xu, F.; Huang, Z.; Tsui, K.-L. Li espan P edic ion o Li hium-Ion Ba e ies Based on Va ious Ex ac ed Fea u es
and G adien Boos ing Reg ession T ee Model. J. Powe Sou ces 2020,476, 228654. [C ossRe ]
84.
G eenbank, S.; Howey, D.A. Piecewise-Linea Modelling wi h Au oma ed Fea u e Selec ion o Li-Ion Ba e y End-o -Li e
P ognosis. Mech. Sys . Signal P ocess. 2023,184, 109612. [C ossRe ]
85.
Wu, J.; Su, H.; Meng, J.; Lin, M. S a e o Heal h Es ima ion o Li hium-Ion Ba e y ia Recu si e Fea u e Elimina ion on Pa ial
Cha ging Cu es. IEEE J. Eme g. Sel. Top. Powe Elec on. 2023,11, 131–142. [C ossRe ]
86.
G eenbank, S.; Howey, D. Au oma ed Fea u e Ex ac ion and Selec ion o Da a-D i en Models o Rapid Ba e y Capaci y Fade
and End o Li e. IEEE T ans. Ind. In o m. 2022,18, 2965–2973. [C ossRe ]
87.
Peng, J.; Zheng, Z.; Zhang, X.; Deng, K.; Gao, K.; Li, H.; Chen, B.; Yang, Y.; Huang, Z. A Da a-D i en Me hod wi h Fea u e
Enhancemen and Adap i e Op imiza ion o Li hium-Ion Ba e y Remaining Use ul Li e P edic ion. Ene gies 2020,13, 752.
[C ossRe ]
88.
Ren, Z.; Du, C.; Ren, W. S a e o Heal h Es ima ion o Li hium-Ion Ba e ies Using a Mul i-Fea u e-Ex ac ion S a egy and
PSO-NARXNN. Ba e ies 2022,9, 7. [C ossRe ]
89.
Fei, Z.; Zhang, Z.; Yang, F.; Tsui, K.-L.; Li, L. Ea ly-S age Li e ime P edic ion o Li hium-Ion Ba e ies: A Deep Lea ning
F amewo k Join ly Conside ing Machine-Lea ned and Handc a ed Da a Fea u es. J. Ene gy S o age 2022,52, 104936. [C ossRe ]
90.
Tang, T.; Yuan, H. The Capaci y P edic ion o Li-Ion Ba e ies Based on a New Fea u e Ex ac ion Technique and an Imp o ed
Ex eme Lea ning Machine Algo i hm. J. Powe Sou ces 2021,514, 230572. [C ossRe ]
91.
Wang, Z.; Liu, N.; Guo, Y. Adap i e Sliding Window LSTM NN Based RUL P edic ion o Li hium-Ion Ba e ies In eg a ing LTSA
Fea u e Recons uc ion. Neu ocompu ing 2021,466, 178–189. [C ossRe ]
92.
Wang, Z.; Yang, F.; Xu, Q.; Wang, Y.; Yan, H.; Xie, M. Capaci y Es ima ion o Li hium-Ion Ba e ies Based on Da a Agg ega ion
and Fea u e Fusion ia G aph Neu al Ne wo k. Appl. Ene gy 2023,336, 120808. [C ossRe ]
93.
Cui, Z.; Gao, X.; Mao, J.; Wang, C. Remaining Capaci y P edic ion o Li hium-Ion Ba e y Based on he Fea u e T ans o ma ion
P ocess Neu al Ne wo k. Expe Sys . Appl. 2022,190, 116075. [C ossRe ]
94.
Yao, X.-Y.; Chen, G.; Hu, L.; Pech , M. A Mul i-Model Fea u e Fusion Model o Li hium-Ion Ba e y S a e o Heal h P edic ion. J.
Ene gy S o age 2022,56, 106051. [C ossRe ]
95.
Fu, P.; Chu, L.; Hou, Z.; Guo, Z.; Lin, Y.; Hu, J. S a e-o -Heal h P edic ion Using T ans e Lea ning and a Mul i-Fea u e Fusion
Model. Senso s 2022,22, 8530. [C ossRe ]
96.
Himad i Sekha , B.; Sindhu, S.; Amalendu Bikash, C.; Chandan Kuma , C. S a e-o -Heal h Es ima ion and End o Li e P edic ion
o he Li hium-Ion Ba e y by Co ela able Fea u e-Based Machine Lea ning App oach. In . J. Pe o m. Eng. 2021,17, 825.
[C ossRe ]
97.
Luo, C.; Zhang, Z.; Qiao, D.; Lai, X.; Li, Y.; Wang, S. Li e P edic ion unde Cha ging P ocess o Li hium-Ion Ba e ies Based on
Au oML. Ene gies 2022,15, 4594. [C ossRe ]
98.
Lee, G.; Kim, J.; Lee, C. S a e-o -Heal h Es ima ion o Li-Ion Ba e ies in he Ea ly Phases o Quali ica ion Tes s: An In e p e able
Machine Lea ning App oach. Expe Sys . Appl. 2022,197, 116817. [C ossRe ]
99.
Li, G.; Li, B.; Li, C.; Wang, S. S a e-o -Heal h Rapid Es ima ion o Li hium-Ion Ba e y Based on an In e p e able S acking
Ensemble Model wi h Sho -Te m Vol age P o iles. Ene gy 2023,263, 126064. [C ossRe ]
100.
Lundbe g, S.M.; Lee, S.-I. A Uni ied App oach o In e p e ing Model P edic ions. In P oceedings o he 31s In e na ional
Con e ence on Neu al In o ma ion P ocessing Sys ems, Long Beach, CA, USA, 4–9 Decembe 2017; Guyon, I., Von Luxbu g,
U., Bengio, S., Wallach, H., Fe gus, R., Vishwana han, S., Ga ne , R., Eds.; Cu an Associa es, Inc.: B ooklyn, NY, USA, 2017;
Volume 30.
101.
Lundbe g, S.M.; E ion, G.; Chen, H.; DeG a e, A.; P u kin, J.M.; Nai , B.; Ka z, R.; Himmel a b, J.; Bansal, N.; Lee, S.-I. F om Local
Explana ions o Global Unde s anding wi h Explainable AI o T ees. Na . Mach. In ell. 2020,2, 56–67. [C ossRe ]
102.
A ia, P.M.; Se e son, K.A.; Wi me , J.D. S a is ical Lea ning o Accu a e and In e p e able Ba e y Li e ime P edic ion. J.
Elec ochem. Soc. 2021,168, 090547. [C ossRe ]
103.
Lee, G.; Kwon, D.; Lee, C. A Con olu ional Neu al Ne wo k Model o SOH Es ima ion o Li-Ion Ba e ies wi h Physical
In e p e abili y. Mech. Sys . Signal P ocess. 2023,188, 110004. [C ossRe ]
104.
Zhang, H.; Su, Y.; Al a , F.; Wik, T.; G os, S. In e p e able Ba e y Cycle Li e Range P edic ion Using Ea ly Cell Deg ada ion Da a.
IEEE T ans. T ansp. Elec i . 2023,9, 2669–2682. [C ossRe ]
Ene gies 2025,18, 746 65 o 77
105.
Pang, X.; Yang, W.; Wang, C.; Fan, H.; Wang, L.; Li, J.; Zhong, S.; Zheng, W.; Zou, H.; Chen, S.; e al. A No el Hyb id Model o
Li hium-Ion Ba e ies Li espan P edic ion wi h High Accu acy and In e p e abili y. J. Ene gy S o age 2023,61, 106728. [C ossRe ]
106.
Kong, D.; Wang, S.; Ping, P. S a e-o -Heal h Es ima ion and Remaining Use ul Li e o Li hium-Ion Ba e y Based on Deep
Lea ning wi h Bayesian Hype pa ame e Op imiza ion. In . J. Ene gy Res. 2022,46, 6081–6098. [C ossRe ]
107.
Reddy, M.C.; Viswana h, M.N.; Va ma, P.S. Remaining Use ul Li e P edic ion o Li hium Ion Ba e y Using T ee Based Pipe-Line
Op imiza ion Tool. J. Ad . Res. Dyn. Con ol Sys . 2020,12, 1955–1960. [C ossRe ]
108.
Shu, X.; Li, G.; Shen, J.; Lei, Z.; Chen, Z.; Liu, Y. A Uni o m Es ima ion F amewo k o S a e o Heal h o Li hium-Ion Ba e ies
Conside ing Fea u e Ex ac ion and Pa ame e s Op imiza ion. Ene gy 2020,204, 117957. [C ossRe ]
109.
Cou u e, J.; Lin, X. No el Image-Based Rapid RUL P edic ion o Li-Ion Ba e ies Using a Capsule Ne wo k and T ans e Lea ning.
IEEE T ans. T ansp. Elec i . 2023,9, 958–967. [C ossRe ]
110.
Cou u e, J.; Lin, X. Image- and Heal h Indica o -Based T ans e Lea ning Hyb idiza ion o Ba e y RUL P edic ion. Eng. Appl.
A i . In ell. 2022,114, 105120. [C ossRe ]
111.
He, N.; Wang, Q.; Lu, Z.; Chai, Y.; Yang, F. Ea ly P edic ion o Ba e y Li e ime Based on G aphical Fea u es and Con olu ional
Neu al Ne wo ks. Appl. Ene gy 2024,353, 122048. [C ossRe ]
112.
Zhao, X.; He, H.; Li, J.; Wei, Z.; Huang, R.; Shi, M. F om G ayscale Image o Ba e y Aging Awa eness—A New Ba e y Capaci y
Es ima ion Model wi h Compu e Vision App oach. IEEE T ans. Ind. In o m. 2023,19, 8965–8975. [C ossRe ]
113.
Li, Y.; Li, K.; Liu, X.; Zhang, L. Fas Ba e y Capaci y Es ima ion Using Con olu ional Neu al Ne wo ks. T ans. Ins . Meas. Con ol
2020, 0142331220966425. [C ossRe ]
114.
Teng, J.-H.; Chen, R.-J.; Lee, P.-T.; Hsu, C.-W. Accu a e and E icien SOH Es ima ion o Re i ed Ba e ies. Ene gies 2023,16, 1240.
[C ossRe ]
115.
Huang, C.; Li, N. Fas Heal h S a e Es ima ion o Lead–Acid Ba e ies Based on Mul i-Time Cons an Cu en Cha ging Cu e.
Elec onics 2023,12, 4552. [C ossRe ]
116.
Lu, S.; Yin, Z.; Liao, S.; Yang, B.; Liu, S.; Liu, M.; Yin, L.; Zheng, W. An Asymme ic Encode –Decode Model o Zn-Ion Ba e y
Li e ime P edic ion. Ene gy Rep. 2022,8, 33–50. [C ossRe ]
117.
Duba y, M.; Beck, D. Big Da a T aining Da a o A i icial In elligence-Based Li-Ion Diagnosis and P ognosis. J. Powe Sou ces
2020,479, 228806. [C ossRe ]
118.
Rau , H.; Khalid, M.; A shad, N. Machine Lea ning in S a e o Heal h and Remaining Use ul Li e Es ima ion: Theo e ical and
Technological De elopmen in Ba e y Deg ada ion Modelling. Renew. Sus ain. Ene gy Re . 2022,156, 111903. [C ossRe ]
119.
Shu, X.; Shen, S.; Shen, J.; Zhang, Y.; Li, G.; Chen, Z.; Liu, Y. S a e o Heal h P edic ion o Li hium-Ion Ba e ies Based on Machine
Lea ning: Ad ances and Pe spec i es. iScience 2021,24, 103265. [C ossRe ]
120.
Wang, S.; Jin, S.; Deng, D.; Fe nandez, C. A C i ical Re iew o Online Ba e y Remaining Use ul Li e ime P edic ion Me hods.
F on . Mech. Eng. 2021,7, 719718. [C ossRe ]
121.
Toughzaoui, Y.; Toosi, S.B.; Chaoui, H.; Louahlia, H.; Pe one, R.; Le Masson, S.; Gualous, H. S a e o Heal h Es ima ion and
Remaining Use ul Li e Assessmen o Li hium-Ion Ba e ies: A Compa a i e S udy. J. Ene gy S o age 2022,51, 104520. [C ossRe ]
122.
Sha ma, P.; Bo a, B.J. A Re iew o Mode n Machine Lea ning Techniques in he P edic ion o Remaining Use ul Li e o Li hium-Ion
Ba e ies. Ba e ies 2023,9, 13. [C ossRe ]
123.
Jin, S.; Sui, X.; Huang, X.; Wang, S.; Teodo escu, R.; S oe, D.-I. O e iew o Machine Lea ning Me hods o Li hium-Ion Ba e y
Remaining Use ul Li e ime P edic ion. Elec onics 2021,10, 3126. [C ossRe ]
124.
Ren, Z.; Du, C. A Re iew o Machine Lea ning S a e-o -Cha ge and S a e-o -Heal h Es ima ion Algo i hms o Li hium-Ion
Ba e ies. Ene gy Rep. 2023,9, 2993–3021. [C ossRe ]
125.
Guo, W.; Sun, Z.; Vilsen, S.B.; Meng, J.; S oe, D.I. Re iew o “G ey Box” Li e ime Modeling o Li hium-Ion Ba e y: Combining
Physics and Da a-D i en Me hods. J. Ene gy S o age 2022,56, 105992. [C ossRe ]
126.
Wang, F.; Zhao, Z.; Zhai, Z.; Shang, Z.; Yan, R.; Chen, X. Explainabili y-D i en Model Imp o emen o SOH Es ima ion o
Li hium-Ion Ba e y. Reliab. Eng. Sys . Sa . 2023,232, 109046. [C ossRe ]
127.
Zhou, L.; Lai, X.; Li, B.; Yao, Y.; Yuan, M.; Weng, J.; Zheng, Y. S a e Es ima ion Models o Li hium-Ion Ba e ies o Ba e y
Managemen Sys em: S a us, Challenges, and Fu u e T ends. Ba e ies 2023,9, 131. [C ossRe ]
128.
Shah, A.; Shah, K.; Shah, C.; Shah, M. S a e o Cha ge, Remaining Use ul Li e and Knee Poin Es ima ion Based on A i icial
In elligence and Machine Lea ning in Li hium-Ion EV Ba e ies: A Comp ehensi e Re iew. Renew. Ene gy Focus 2022,42, 146–164.
[C ossRe ]
129.
Li, X.; Yu, D.; Sø en Byg, V.; Daniel Ioan, S. The De elopmen o Machine Lea ning-Based Remaining Use ul Li e P edic ion o
Li hium-Ion Ba e ies. J. Ene gy Chem. 2023,82, 103–121. [C ossRe ]
130.
Raoo i, T.; Yildiz, M. Comp ehensi e Re iew o Ba e y S a e Es ima ion S a egies Using Machine Lea ning o Ba e y Manage-
men Sys ems o Ai c a P opulsion Ba e ies. J. Ene gy S o age 2023,59, 106486. [C ossRe ]
131.
Xiao, Y.; Wen, J.; Yao, L.; Zheng, J.; Fang, Z.; Shen, Y. A Comp ehensi e Re iew o he Li hium-Ion Ba e y S a e o Heal h
P ognosis Me hods Combining Aging Mechanism Analysis. J. Ene gy S o age 2023,65, 107347. [C ossRe ]
Ene gies 2025,18, 746 66 o 77
132.
He, W.; Li, Z.; Liu, T.; Liu, Z.; Guo, X.; Du, J.; Li, X.; Sun, P.; Ming, W. Resea ch P og ess and Applica ion o Deep Lea ning in
Remaining Use ul Li e, S a e o Heal h and Ba e y The mal Managemen o Li hium Ba e ies. J. Ene gy S o age 2023,70, 107868.
[C ossRe ]
133.
Bel an, H.; Sansano, E.; Pech , M. Machine Lea ning Techniques Sui abili y o Es ima e he Re ained Capaci y in Li hium-Ion
Ba e ies om Pa ial Cha ge/Discha ge Cu es. J. Ene gy S o age 2023,59, 106346. [C ossRe ]
134.
Z aibi, B.; Mansou i, M.; Loukili, S.E. Compa ing Deep Lea ning Me hods o P edic he Remaining Use ul Li e o Li hium-Ion
Ba e ies. Ma e . Today P oc. 2022,62, 6298–6304. [C ossRe ]
135.
Huang, Z.; Sugia o, L.; Lu, Y.-C. Fea u e–Ta ge Pai ing in Machine Lea ning o Ba e y Heal h Diagnosis and P ognosis: A
C i ical Re iew. EcoMa 2023,5, e12345. [C ossRe ]
136.
Zhou, Y.; Chang, Y.; Wang, Y.; Li, R. Resea ch on Me hods o Ex ac ing Aging Cha ac e is ics and Heal h S a us o Li hium-Ion
Ba e ies Based on Small Samples. J. Renew. Sus ain. Ene gy 2022,14, 24101. [C ossRe ]
137.
Das, K.; Kuma , R. Elec ic Vehicle Ba e y Capaci y Deg ada ion and Heal h Es ima ion Using Machine-Lea ning Techniques: A
Re iew. Clean Ene gy 2023,7, 1268–1281. [C ossRe ]
138.
Wang, F.; Zhai, Z.; Liu, B.; Zheng, S.; Zhao, Z.; Chen, X. Open Access Da ase , Code Lib a y and Benchma king Deep Lea ning
App oaches o S a e-o -Heal h Es ima ion o Li hium-Ion Ba e ies. J. Ene gy S o age 2024,77, 109884. [C ossRe ]
139. O zech, G. P ognos ics Cen e o Excellence Da a Se Reposi o y; NASA: Washing on, DC, USA, 2022.
140.
Duan, W.; Song, S.; Xiao, F.; Chen, Y.; Peng, S.; Song, C. Ba e y SOH Es ima ion and RUL P edic ion F amewo k Based on
Va iable Fo ge ing Fac o Online Sequen ial Ex eme Lea ning Machine and Pa icle Fil e . J. Ene gy S o age 2023,65, 107322.
[C ossRe ]
141.
Yao, L.; Wen, J.; Xu, S.; Zheng, J.; Hou, J.; Fang, Z.; Xiao, Y. S a e o Heal h Es ima ion Based on he Long Sho -Te m Memo y
Ne wo k Using Inc emen al Capaci y and T ans e Lea ning. Senso s 2022,22, 7835. [C ossRe ] [PubMed]
142.
Yang, X.; Ma, B.; Xie, H.; Wang, W.; Zou, B.; Liang, F.; Hua, X.; Liu, X.; Chen, S. Li hium-Ion Ba e y S a e o Heal h Es ima ion
wi h Mul i-Fea u e Collabo a i e Analysis and Deep Lea ning Me hod. Ba e ies 2023,9, 120. [C ossRe ]
143.
Peng, S.; Sun, Y.; Liu, D.; Yu, Q.; Kan, J.; Pech , M. S a e o Heal h Es ima ion o Li hium-Ion Ba e ies Based on Mul i-Heal h
Fea u es Ex ac ion and Imp o ed Long Sho -Te m Memo y Neu al Ne wo k. Ene gy 2023,282, 128956. [C ossRe ]
144.
Jo ge, I.; Mesbahi, T.; Same , A.; Boné, R. Time Se ies Fea u e Ex ac ion o Li hium-Ion Ba e ies S a e-O -Heal h P edic ion. J.
Ene gy S o age 2023,59, 106436. [C ossRe ]
145.
Zhang, C.; Wang, S.; Yu, C.; Xie, Y.; Fe nandez, C. Imp o ed Pa icle Swa m Op imiza ion-Ex eme Lea ning Machine Modeling
S a egies o he Accu a e Li hium-Ion Ba e y S a e o Heal h Es ima ion and High-Adap abili y Remaining Use ul Li e
P edic ion. J. Elec ochem. Soc. 2022,169, 080520. [C ossRe ]
146.
Maleki, S.; Ray, B.; Hagh, M.T. Hyb id F amewo k o P edic ing and Fo ecas ing S a e o Heal h o Li hium-Ion Ba e ies in
Elec ic Vehicles. Sus ain. Ene gy G ids Ne w. 2022,30, 100603. [C ossRe ]
147.
Wu, W.; Lu, S. Remaining Use ul Li e P edic ion o Li hium-Ion Ba e ies Based on Da a P ep ocessing and Imp o ed ELM. IEEE
T ans. Ins um. Meas. 2023,72, 1–14. [C ossRe ]
148.
Zhang, Y.; Ma, H.; Wang, S.; Li, S.; Guo, R. Indi ec P edic ion o Remaining Use ul Li e o Li hium-Ion Ba e ies Based on
Imp o ed Mul iple Ke nel Ex eme Lea ning Machine. J. Ene gy S o age 2023,64, 107181. [C ossRe ]
149.
Sun, C.; Qu, A.; Zhang, J.; Shi, Q.; Jia, Z. Remaining Use ul Li e P edic ion o Li hium-Ion Ba e ies Based on Imp o ed Va ia ional
Mode Decomposi ion and Machine Lea ning Algo i hm. Ene gies 2022,16, 313. [C ossRe ]
150.
Tang, X.; Wan, H.; Wang, W.; Gu, M.; Wang, L.; Gan, L. Li hium-Ion Ba e y Remaining Use ul Li e P edic ion Based on Hyb id
Model. Sus ainabili y 2023,15, 6261. [C ossRe ]
151.
Xue, K.; Yang, J.; Yang, M.; Wang, D. An Imp o ed Gene ic Hyb id P ognos ic Me hod o RUL P edic ion Based on PF-LSTM
Lea ning. IEEE T ans. Ins um. Meas. 2023,72, 1–21. [C ossRe ]
152.
Tian, A.; Yang, C.; Gao, Y.; Li, T.; Wang, L.; Chang, C.; Jiang, J. A S a e o Heal h Es ima ion Me hod o Li hium-Ion Ba e ies
Based on DT-IC-V Heal h Fea u es Ex ac ed om Pa ial Cha ging Segmen . In . J. G een Ene gy 2023,20, 997–1011. [C ossRe ]
153.
Guo, Y.; Yang, D.; Zhao, K.; Wang, K. S a e o Heal h Es ima ion o Li hium-Ion Ba e y Based on Bi-Di ec ional Long Sho -Te m
Memo y Neu al Ne wo k and A en ion Mechanism. Ene gy Rep. 2022,8, 208–215. [C ossRe ]
154.
Pugalen hi, K.; Pa k, H.; Hussain, S.; Ragha an, N. Remaining Use ul Li e P edic ion o Li hium-Ion Ba e ies Using Neu al
Ne wo ks wi h Adap i e Bayesian Lea ning. Senso s 2022,22, 3803. [C ossRe ] [PubMed]
155.
Hell, S.M.; Kim, C.D. De elopmen o a Da a-D i en Me hod o Online Ba e y Remaining-Use ul-Li e P edic ion. Ba e ies 2022,
8, 192. [C ossRe ]
156.
Zhu, C.; Gao, M.; He, Z.; Wu, H.; Sun, C.; Zhang, Z.; Bao, Z. S a e o Heal h P edic ion o Li-Ion Ba e ies wi h End- o-End Deep
Lea ning. J. Ene gy S o age 2023,65, 107218. [C ossRe ]
157.
Bao, X.; Chen, L.; Lopes, A.M.; Li, X.; Xie, S.; Li, P.; Chen, Y. Hyb id Deep Neu al Ne wo k wi h Dimension A en ion o
S a e-o -Heal h Es ima ion o Li hium-Ion Ba e ies. Ene gy 2023,278, 127734. [C ossRe ]
Ene gies 2025,18, 746 67 o 77
158.
Guo, X.; Wang, K.; Yao, S.; Fu, G.; Ning, Y. RUL P edic ion o Li hium Ion Ba e y Based on CEEMDAN-CNN BiLSTM Model.
Ene gy Rep. 2023,9, 1299–1306. [C ossRe ]
159.
Sheng, H.; Zhou, Y.; Bai, L.; Shi, L. T ans e S a e o Heal h Es ima ion Based on C oss-Mani old Embedding. J. Ene gy S o age
2022,47, 103555. [C ossRe ]
160.
Liu, R. Remaining Use ul Li e P edic ion o Li hium-Ion Ba e ies Using Mul iple Ke nel Ex eme Lea ning Machine. Recen Ad .
Compu . Sci. Commun. 2022,15, 715–721. [C ossRe ]
161.
Wu, J.; Cheng, X.; Huang, H.; Fang, C.; Zhang, L.; Zhao, X.; Zhang, L.; Xing, J. Remaining Use ul Li e P edic ion o Li hium-Ion
Ba e ies Based on PSO-RF Algo i hm. F on . Ene gy Res. 2023,10, 937035. [C ossRe ]
162.
Sahoo, S.; Ha iha an, K.S.; Aga wal, S.; Swe na h, S.B.; Bha i, R.; Han, S.; Lee, S. T ans e Lea ning Based Gene alized F amewo k
o S a e o Heal h Es ima ion o Li-Ion Cells. Sci. Rep. 2022,12, 13173. [C ossRe ] [PubMed]
163.
Li, N.; He, F.; Ma, W.; Wang, R.; Jiang, L.; Zhang, X. An Indi ec S a e-o -Heal h Es ima ion Me hod Based on Imp o ed Gene ic
and Back P opaga ion o Online Li hium-Ion Ba e y Used in Elec ic Vehicles. IEEE T ans. Veh. Technol. 2022,71, 12682–12690.
[C ossRe ]
164.
Jiang, J.; Zhao, S.; Zhang, C. S a e-o -Heal h Es ima e o he Li hium-Ion Ba e y Using Chi-Squa e and ELM-LSTM. Wo ld Elec .
Veh. J. 2021,12, 228. [C ossRe ]
165.
El-Dalahmeh, M.; Al-G ee , M.; El-Dalahmeh, M.; Bashi , I. Capaci y Es ima ion o Li hium-Ion Ba e ies Based on Adap i e
Empi ical Wa ele T ans o m and Long Sho -Te m Memo y Neu al Ne wo k. J. Ene gy S o age 2023,70, 108046. [C ossRe ]
166.
Tian, Y.; Wen, J.; Yang, Y.; Shi, Y.; Zeng, J. S a e-o -Heal h P edic ion o Li hium-Ion Ba e ies Based on CNN-BiLSTM-AM. Ba e ies
2022,8, 155. [C ossRe ]
167.
Mei, P.; Ka imi, H.R.; Chen, F.; Yang, S.; Huang, C.; Qiu, S. A Lea ning-Based Vehicle-Cloud Collabo a ion App oach o Join
Es ima ion o S a e-o -Ene gy and S a e-o -Heal h. Senso s 2022,22, 9474. [C ossRe ]
168.
Song, R.; Yang, L.; Chen, L.; Dong, Z. Capaci y Es ima ion Me hod o Li hium-Ion Ba e ies Based on Deep Con olu ion Neu al
Ne wo k. In . J. Bio-Inspi ed Compu . 2022,20, 119–125. [C ossRe ]
169.
Wang, J.; Feng, X.; Zhang, X.; Xiang, Y. Imp o ed Modeling o Li hium-Ion Ba e y Capaci y Deg ada ion Using an Indi idual-
S a e T aining Me hod and Recu en So plus Neu al Ne wo k. IEEE Access 2021,9, 7845–7855. [C ossRe ]
170.
Wen, L.; Bo, N.; Ye, X.; Li, X. A No el Au o-LSTM-Based S a e o Heal h Es ima ion Me hod o Li hium-Ion Ba e ies. J.
Elec ochem. Ene gy Con e s. S o age 2021,18, 030902. [C ossRe ]
171.
Wei, Y. P edic ion o S a e o Heal h o Li hium-Ion Ba e y Using Heal h Index In o med A en ion Model. Senso s 2023,23, 2587.
[C ossRe ] [PubMed]
172.
Chen, C.; Wei, J.; Li, Z. Remaining Use ul Li e P edic ion o Li hium-Ion Ba e ies Based on a Hyb id Deep Lea ning Model.
P ocesses 2023,11, 2333. [C ossRe ]
173.
Zhu, T.; Wang, W.; Yu, M. A No el Hyb id Scheme o Remaining Use ul Li e P ognos ic Based on Seconda y Decomposi ion,
BiGRU and E o Co ec ion. Ene gy 2023,276, 127565. [C ossRe ]
174.
Pham, T.; Le, T.; Dang, D.; Bui, H.; Pham, H.; T uong, L.; Nguyen, M.; Vo, H.; Tho, Q.T. ARNS: A Da a-D i en App oach o
SoH Es ima ion o Li hium-Ion Ba e y Using Nes ed Sequence Models wi h Conside ing Relaxa ion E ec . IEEE Access 2022,10,
117067–117083. [C ossRe ]
175.
Ang, E.Y.M.; Paw, Y.C. Linea Model o Online S a e o Heal h Es ima ion o Li hium-Ion Ba e ies Using Segmen ed Discha ge
P o iles. IEEE T ans. T ansp. Elec i . 2023,9, 2464–2471. [C ossRe ]
176.
Hong, S.; Kang, M.; Kim, J.; Baek, J. Sequen ial Applica ion o Denoising Au oencode and Long-Sho Recu en Con olu ional
Ne wo k o Noise-Robus Remaining-Use ul-Li e P edic ion F amewo k o Li hium-Ion Ba e ies. Compu . Ind. Eng. 2023,
179, 109231. [C ossRe ]
177.
Chen, D.; Zheng, X.; Chen, C.; Zhao, W.; Chen, D.; Zheng, X.; Chen, C.; Zhao, W. Remaining Use ul Li e P edic ion o he
Li hium-Ion Ba e y Based on CNN-LSTM Fusion Model and G ey Rela ional Analysis. Elec on. Res. A ch. 2023,31, 633–655.
[C ossRe ]
178.
Zhu, C.; Zheng, B.; He, Z.; Gao, M.; Sun, C.; Bao, Z. S a e o Heal h Es ima ion o Li hium-Ion Ba e y Using Time Con olu ion
Memo y Neu al Ne wo k. Mob. In . Sys . 2021,2021, 4826409. [C ossRe ]
179.
Wei, Y.; Wu, D. S a e o Heal h and Remaining Use ul Li e P edic ion o Li hium-Ion Ba e ies wi h Condi ional G aph Con olu-
ional Ne wo k. Expe Sys . Appl. 2024,238, 122041. [C ossRe ]
180.
Huang, K.; Yao, K.; Guo, Y.; L , Z. S a e o Heal h Es ima ion o Li hium-Ion Ba e ies Based on Fine-Tuning o Rebuilding
T ans e Lea ning S a egies Combined wi h New Fea u es Mining. Ene gy 2023,282, 128739. [C ossRe ]
181.
Zou, B.; Xiong, M.; Wang, H.; Ding, W.; Jiang, P.; Hua, W.; Zhang, Y.; Zhang, L.; Wang, W.; Tan, R. A Deep Lea ning App oach o
S a e-o -Heal h Es ima ion o Li hium-Ion Ba e ies Based on a Mul i-Fea u e and A en ion Mechanism Collabo a ion. Ba e ies
2023,9, 329. [C ossRe ]
182.
Liu, H.; Li, Y.; Luo, L.; Zhang, C. A Li hium-Ion Ba e y Capaci y and RUL P edic ion Fusion Me hod Based on Decomposi ion
S a egy and GRU. Ba e ies 2023,9, 323. [C ossRe ]
Ene gies 2025,18, 746 68 o 77
183.
Yu, Z.; Liu, N.; Zhang, Y.; Qi, L.; Li, R. Ba e y SOH P edic ion Based on Mul i-Dimensional Heal h Indica o s. Ba e ies 2023,9, 80.
[C ossRe ]
184.
Ma, W.; Cai, P.; Sun, F.; Wang, X.; Gong, J. S a e o Heal h Es ima ion o Li hium-Ion Ba e ies Based on Ex eme Lea ning
Machine wi h Imp o ed Blinex Loss. In . J. Elec ochem. Sci. 2022,17, 221170. [C ossRe ]
185.
Yuan, J.; Qin, Z.; Huang, H.; Gan, X.; Li, S.; Li, B. S a e o Heal h Es ima ion and Remaining Use ul Li e P edic ion o a
Li hium-Ion Ba e y wi h a Two-Laye S acking Reg esso . Ene gies 2023,16, 2313. [C ossRe ]
186.
Gao, D.; Liu, X.; Zhu, Z.; Yang, Q. A Hyb id CNN-BiLSTM App oach o Remaining Use ul Li e P edic ion o EVs Li hium-Ion
Ba e y. Meas. Con ol 2023,56, 371–383. [C ossRe ]
187.
Chen, L.; Xie, S.; Lopes, A.M.; Bao, X. A Vision T ans o me -Based Deep Neu al Ne wo k o S a e o Heal h Es ima ion o
Li hium-Ion Ba e ies. In . J. Elec . Powe Ene gy Sys . 2023,152, 109233. [C ossRe ]
188.
Han, Y.; Li, C.; Zheng, L.; Lei, G.; Li, L. Remaining Use ul Li e P edic ion o Li hium-Ion Ba e ies by Using a Denoising
T ans o me -Based Neu al Ne wo k. Ene gies 2023,16, 6328. [C ossRe ]
189.
Wang, M.; Xiang, G.; Cui, L.; Zhang, Q.; Chen, J. Remaining Use ul Li e Dis ibu ion P edic ion F amewo k o Li hium-Ion
Ba e y Fused P io Knowledge and Moni o ing Da a. Meas. Sci. Technol. 2023,34, 125108. [C ossRe ]
190.
Ma i, I.; Pe ko ski, E.; C is aldi, L.; Fai e , M. Compa ing Machine Lea ning S a egies o SoH Es ima ion o Li hium-Ion
Ba e ies Using a Fea u e-Based App oach. Ene gies 2023,16, 4423. [C ossRe ]
191.
Xie, Q.; Liu, R.; Huang, J.; Su, J. Residual Li e P edic ion o Li hium-Ion Ba e ies Based on Da a P ep ocessing and a P io i
Knowledge-Assis ed CNN-LSTM. Ene gy 2023,281, 128232. [C ossRe ]
192.
Li, Z.; Li, A.; Bai, F.; Zuo, H.; Zhang, Y. Remaining Use ul Li e P edic ion o Li hium Ba e y Based on ACNN-Mog i ie
LSTM-MMD. Meas. Sci. Technol. 2024,35, 016101. [C ossRe ]
193.
Ta a , M.O.; Naq i, I.H.; Khalid, Z.; Pech , M. Accu a e P edic ion o Remaining Use ul Li e o Li hium-Ion Ba e y Using Deep
Neu al Ne wo ks wi h Memo y Fea u es. F on . Ene gy Res. 2023,11, 1059701. [C ossRe ]
194.
Wu, J.; Liu, Z.; Zhang, Y.; Lei, D.; Zhang, B.; Cao, W. Da a-D i en S a e o Heal h Es ima ion o Li hium-Ion Ba e y Based on
Vol age Va ia ion Cu es. J. Ene gy S o age 2023,73, 109191. [C ossRe ]
195.
Wang, Z.; Liu, Y.; Wang, F.; Wang, H.; Su, M. Capaci y and Remaining Use ul Li e P edic ion o Li hium-Ion Ba e ies Based on
Sequence Decomposi ion and a Deep-Lea ning Ne wo k. J. Ene gy S o age 2023,72, 108085. [C ossRe ]
196.
Chang, Y.-H.; Hsieh, Y.-C.; Chai, Y.-H.; Lin, H.-W. Remaining-Use ul-Li e P edic ion o Li-Ion Ba e ies. Ene gies 2023,16, 3096.
[C ossRe ]
197.
Wang, J.; Zhu, L.; Dai, H. An E icien S a e-o -Heal h Es ima ion Me hod o Li hium-Ion Ba e ies Based on Fea u e-Impo ance
Ranking S a egy and PSO-GRNN Algo i hm. J. Ene gy S o age 2023,72, 108638. [C ossRe ]
198.
Nai , P.; Vakha ia, V.; Bo ade, H.; Shah, M.; Wankhede, V. P edic ing Li-Ion Ba e y Remaining Use ul Li e: An XDFM-D i en
App oach wi h Explainable AI. Ene gies 2023,16, 5725. [C ossRe ]
199.
Zhang, F.; Shen, Z.; Xu, M.; Xie, Q.; Fu, Q.; Ma, R. Remaining Use ul Li e P edic ion o Li hium-Ion Ba e ies Based on TCN-DCN
Fusion Model Combined wi h IRRS Fil e ing. J. Ene gy S o age 2023,72, 108586. [C ossRe ]
200.
Pan, R.; Liu, T.; Huang, W.; Wang, Y.; Yang, D.; Chen, J. S a e o Heal h Es ima ion o Li hium-Ion Ba e ies Based on Two-S age
Fea u es Ex ac ion and G adien Boos ing Decision T ee. Ene gy 2023,285, 129460. [C ossRe ]
201.
Zhang, Y.; Zhang, Y.; Wu, T. Es ima ion o S a e o Heal h Based on Cha ging Cha ac e is ics and Back-P opaga ion Neu al
Ne wo ks wi h Imp o ed A om Sea ch Op imiza ion Algo i hm. Glob. Ene gy In e connec . 2023,6, 228–237. [C ossRe ]
202.
Ja a i, S.; Byun, Y.-C. A CNN-GRU App oach o he Accu a e P edic ion o Ba e ies’ Remaining Use ul Li e om Cha ging
P o iles. Compu e s 2023,12, 219. [C ossRe ]
203.
Wang, Y.; He, Q.; Zhang, D.; Lu, S.; Yuan, C. Imp o ing Li-Ion Ba e y Heal h: P edic ing Remaining Use ul Li e Using
IWBOA-ELM Algo i hm. J. Ene gy S o age 2023,72, 108547. [C ossRe ]
204.
Wang, F.; Yang, Y.; Huang, T.; Xu, Y. Li e ime P edic ion o Elec onic De ices Based on he P-S acking Machine Lea ning Model.
Mic oelec on. Reliab. 2023,146, 115027. [C ossRe ]
205. Ansa i, S.; Ayob, A.; Lipu, M.S.H.; Hussain, A.; Abdol asol, M.G.M.; Zainu i, M.A.A.M.; Saad, M.H.M. Op imized Da a-D i en
App oach o Remaining Use ul Li e P edic ion o Li hium-Ion Ba e ies Based on Sliding Window and Sys ema ic Sampling. J.
Ene gy S o age 2023,73, 109198. [C ossRe ]
206.
Guo, L.; He, H.; Ren, Y.; Li, R.; Jiang, B.; Gong, J. P ognos ics o Li hium-Ion Ba e ies Heal h S a e Based on Adap i e Mode
Decomposi ion and Long Sho -Te m Memo y Neu al Ne wo k. Eng. Appl. A i . In ell. 2024,127, 107317. [C ossRe ]
207.
Zhang, Y.; Wang, Y.; Zhang, C.; Qiao, X.; Ge, Y.; Li, X.; Peng, T.; Nazi , M.S. S a e-o -Heal h Es ima ion o Li hium-Ion Ba e y
ia an E olu iona y S acking Ensemble Lea ning Pa adigm o Random Vec o Func ional Link and Ac i e-S a e-T acking
Long–Sho -Te m Memo y Neu al Ne wo k. Appl. Ene gy 2024,356, 122417. [C ossRe ]
208.
Li, Z.; Bai, F.; Zuo, H.; Zhang, Y. Remaining Use ul Li e P edic ion o Li hium-Ion Ba e ies Based on I e a i e T ans e Lea ning
and Mog i ie LSTM. Ba e ies 2023,9, 448. [C ossRe ]
Ene gies 2025,18, 746 69 o 77
209.
Ren, Z.; Du, C.; Zhao, Y. A No el Me hod o S a e o Heal h Es ima ion o Li hium-Ion Ba e ies Based on Deep Lea ning Neu al
Ne wo k and T ans e Lea ning. Ba e ies 2023,9, 585. [C ossRe ]
210.
Zheng, W.; Zhou, X.; Bai, C.; Zhou, D.; Fu, P. Adap a ion o Deep Ne wo k in T ans e Lea ning o Es ima ing S a e o Heal h in
Elec ic Vehicles du ing Ope a ion. Ba e ies 2023,9, 547. [C ossRe ]
211.
Kuo, P.-H.; Tseng, Y.-R.; Luan, P.-C.; Yau, H.-T. B oad T ans e Lea ning Ne wo k Based Li-Ion Ba e y Li e ime P edic ion Model.
Ene gy Rep. 2023,10, 881–893. [C ossRe ]
212.
Bosello, M.; Falcome , C.; Rossi, C.; Pau, G. To Cha ge o o Sell? EV Pack Use ul Li e Es ima ion ia LSTMs, CNNs, and
Au oencode s. Ene gies 2023,16, 2837. [C ossRe ]
213.
Be ghou , T.; Benbouzid, M.; Ami a , Y.; Yao, G. Li hium-Ion Ba e y S a e o Heal h P edic ion wi h a Robus Collabo a i e
Augmen ed Hidden Laye Feed o wa d Neu al Ne wo k App oach. IEEE T ans. T ansp. Elec i . 2023,9, 4492–4502. [C ossRe ]
214.
Hong, J.; Chen, Y.; Chai, Q.; Lin, Q.; Wang, W. S a e-o -Heal h Es ima ion o Li hium-Ion Ba e ies Using a No el Dual-S age
A en ion Mechanism Based Recu en Neu al Ne wo k. J. Ene gy S o age 2023,72, 109297. [C ossRe ]
215.
Wang, S.; Ma, H.; Zhang, Y.; Li, S.; He, W. Remaining Use ul Li e P edic ion Me hod o Li hium-Ion Ba e ies Is Based on
Va ia ional Modal Decomposi ion and Deep Lea ning In eg a ed App oach. Ene gy 2023,282, 128984. [C ossRe ]
216.
Zhang, L.; Zhang, J.; Gao, T.; Lyu, L.; Wang, L.; Shi, W.; Jiang, L.; Cai, G. Imp o ed LSTM Based S a e o Heal h Es ima ion Using
Random Segmen s o he Cha ging Cu es o Li hium-Ion Ba e ies. J. Ene gy S o age 2023,74, 109370. [C ossRe ]
217.
Guo, F.; Huang, G.; Zhang, W.; Wen, A.; Li, T.; He, H.; Huang, H.; Zhu, S. Li hium Ba e y S a e-o -Heal h Es ima ion Based on
Sample Da a Gene a ion and Tempo al Con olu ional Neu al Ne wo k. Ene gies 2023,16, 8010. [C ossRe ]
218.
Liu, Y.; Sun, G.; Liu, X. Remaining Use ul Li e P edic ion o Li hium-Ion Ba e ies Based on Peak In e al Fea u es and Deep
Lea ning. J. Ene gy S o age 2023,73, 109308. [C ossRe ]
219.
Jin, Z.; Fang, C.; Wu, J.; Li, J.; Zeng, W.; Zhao, X. Remaining Use ul Li e P edic ion o Li hium-Ion Ba e y Using a Da a-D i en
Me hod. In . J. Wi el. Mob. Compu . 2022,23, 239–249. [C ossRe ]
220.
Zhao, L.; Song, S.; Wang, P.; Wang, C.; Wang, J.; Guo, M. A MLP-Mixe and Mix u e o Expe Model o Remaining Use ul Li e
P edic ion o Li hium-Ion Ba e ies. F on . Compu . Sci. 2023,18, 185329. [C ossRe ]
221.
Wang, X.; Hu, B.; Su, X.; Xu, L.; Zhu, D. S a e o Heal h Es ima ion o Li hium-Ion Ba e ies Using Random Fo es and Ga ed
Recu en Uni . J. Ene gy S o age 2024,76, 109796. [C ossRe ]
222.
Chen, L.; Xie, S.; Lopes, A.M.; Li, H.; Bao, X.; Zhang, C.; Li, P. A New SOH Es ima ion Me hod o Li hium-Ion Ba e ies Based on
Model-Da a-Fusion. Ene gy 2024,286, 129597. [C ossRe ]
223.
Mazzi, Y.; Ben Sassi, H.; E ahimi, F. Li hium-Ion Ba e y S a e o Heal h Es ima ion Using a Hyb id Model Based on a
Con olu ional Neu al Ne wo k and Bidi ec ional Ga ed Recu en Uni . Eng. Appl. A i . In ell. 2024,127, 107199. [C ossRe ]
224.
Zhigang, L.; Meng, Z.; Ruohai, D.; Peng, W.; Hui, G.; Hongxi, W. Remaining Use ul Li e P edic ion o Li hium-Ion Ba e ies Based
on KS Agglome a ion Func ion In eg a ing Mul i-Expe Knowledge. Mic oelec on. Reliab. 2023,145, 114985. [C ossRe ]
225.
Li, C.; Han, X.; Zhang, Q.; Li, M.; Rao, Z.; Liao, W.; Liu, X.; Liu, X.; Li, G. S a e-o -Heal h and Remaining-Use ul-Li e Es ima ions
o Li hium-Ion Ba e y Based on Tempo al Con olu ional Ne wo k-Long Sho -Te m Memo y. J. Ene gy S o age 2023,74, 109498.
[C ossRe ]
226.
Rincón-Maya, C.; Gue a a-Ca azas, F.; He nández-Ba ajas, F.; Pa ino-Rod iguez, C.; Usuga-Manco, O. Remaining Use ul Li e
P edic ion o Li hium-Ion Ba e y Using ICC-CNN-LSTM Me hodology. Ene gies 2023,16, 7081. [C ossRe ]
227.
Zhang, C.; Wang, S.; Yu, C.; Wang, Y.; Fe nandez, C. A Comple e Ensemble Empi ical Mode Decomposi ion wi h Adap i e Noise
Deep Au o eg essi e Recu en Neu al Ne wo k Me hod o he Whole Li e Remaining Use ul Li e P edic ion o Li hium-Ion
Ba e ies. Ionics 2023,29, 4337–4349. [C ossRe ]
228.
Cai, Y.; Li, W.; Zahid, T.; Zheng, C.; Zhang, Q.; Xu, K. Ea ly P edic ion o Remaining Use ul Li e o Li hium-Ion Ba e ies Based
on CEEMDAN-T ans o me -DNN Hyb id Model. Heliyon 2023,9, e17754. [C ossRe ]
229.
Feng, J.; Cai, F.; Li, H.; Huang, K.; Yin, H. A Da a-D i en P edic ion Model o he Remaining Use ul Li e P edic ion o Li hium-Ion
Ba e ies. P ocess Sa . En i on. P o . 2023,180, 601–615. [C ossRe ]
230.
Xia, T.; Zhang, X.; Zhu, H.; Zhang, X.; Shen, J. An Accu a e Denoising Li hium-Ion Ba e y Remaining Use ul Li e P edic ion
Model Based on CNN and LSTM wi h Sel -A en ion. Ionics 2023,29, 5315–5328. [C ossRe ]
231.
Ding, G.; Chen, H. A No el Li hium-Ion Ba e y Capaci y P edic ion F amewo k Based on SVMD-AO-DELM. Signal Image Video
P ocess. 2023,17, 3793–3801. [C ossRe ]
232.
Xue, J.; Ma, W.; Feng, X.; Guo, P.; Guo, Y.; Hu, X.; Chen, B. S acking In eg a ed Lea ning Model ia ELM and GRU wi h Mix u e
Co en opy Loss o Robus S a e o Heal h Es ima ion o Li hium-Ion Ba e ies. Ene gy 2023,284, 129279. [C ossRe ]
233.
Zheng, X.; Chen, D.; Wang, Y.; Zhuang, L.; Zheng, X.; Chen, D.; Wang, Y.; Zhuang, L. Remaining Use ul Li e Indi ec P edic ion o
Li hium-Ion Ba e ies Using CNN-BiGRU Fusion Model and TPE Op imiza ion. AIMS Ene gy 2023,11, 896–917. [C ossRe ]
234.
Geng, C.; Zhang, T.; Chen, B.; Zhou, Q. Ba e y S a e o Heal h Es ima ion Using GA-BP Neu al Ne wo k on Da a Fea u e Mining.
IEICE Elec on. Exp ess 2023,20, 20230370. [C ossRe ]
Ene gies 2025,18, 746 70 o 77
235.
Sun, S.; Zhang, H.; Ge, J.; Che, L. S a e-o -Heal h Es ima ion o Li hium-Ion Ba e y Using Model-Based Fea u e Op imiza ion
and Deep Ex eme Lea ning Machine. J. Ene gy S o age 2023,72, 108732. [C ossRe ]
236.
Z aibi, B.; Oka , C.; Chaoui, H.; Mansou i, M. Remaining Use ul Li e Assessmen o Li hium-Ion Ba e ies Using CNN-LSTM-
DNN Hyb id Me hod. IEEE T ans. Veh. Technol. 2021,70, 4252–4261. [C ossRe ]
237.
Zhao, S.; Zhang, C.; Wang, Y. Li hium-Ion Ba e y Capaci y and Remaining Use ul Li e P edic ion Using Boa d Lea ning Sys em
and Long Sho -Te m Memo y Neu al Ne wo k. J. Ene gy S o age 2022,52, 104901. [C ossRe ]
238.
Li, Y.; Sheng, H.; Cheng, Y.; S oe, D.-I.; Teodo escu, R. S a e-o -Heal h Es ima ion o Li hium-Ion Ba e ies Based on Semi-
Supe ised T ans e Componen Analysis. Appl. Ene gy 2020,277, 115504. [C ossRe ]
239.
Guo, Y.; Yang, D.; Zhang, Y.; Wang, L.; Wang, K. Online Es ima ion o SOH o Li hium-Ion Ba e y Based on SSA-Elman Neu al
Ne wo k. P o . Con ol Mod. Powe Sys . 2022,7, 40. [C ossRe ]
240.
Chen, D.; Hong, W.; Zhou, X. T ans o me Ne wo k o Remaining Use ul Li e P edic ion o Li hium-Ion Ba e ies. IEEE Access
2022,10, 19621–19628. [C ossRe ]
241.
Nascimen o, R.G.; Co be a, M.; Kulka ni, C.S.; Viana, F.A.C. Hyb id Physics-In o med Neu al Ne wo ks o Li hium-Ion Ba e y
Modeling and P ognosis. J. Powe Sou ces 2021,513, 230526. [C ossRe ]
242.
Li, Y.; Sheng, H.; Cheng, Y.; Kuang, H. Li hium-Ion Ba e y S a e o Heal h Moni o ing Based on Ensemble Lea ning. In
P oceedings o he 2019 IEEE In e na ional Ins umen a ion and Measu emen Technology Con e ence, I2MTC 2019, Auckland,
New Zealand, 20–23 May 2019.
243.
Li, P.; Zhang, Z.; G osu, R.; Deng, Z.; Hou, J.; Rong, Y.; Wu, R. An End- o-End Neu al Ne wo k F amewo k o S a e-o -Heal h
Es ima ion and Remaining Use ul Li e P edic ion o Elec ic Vehicle Li hium Ba e ies. Renew. Sus ain. Ene gy Re . 2022,
156, 111843. [C ossRe ]
244.
Gou, B.; Xu, Y.; Feng, X. An Ensemble Lea ning-Based Da a-D i en Me hod o Online S a e-o -Heal h Es ima ion o Li hium-Ion
Ba e ies. IEEE T ans. T ansp. Elec i . 2021,7, 422–436. [C ossRe ]
245.
Wang, S.; Fan, Y.; Jin, S.; Takyi-Aninakwa, P.; Fe nandez, C. Imp o ed An i-Noise Adap i e Long Sho -Te m Memo y Neu al
Ne wo k Modeling o he Robus Remaining Use ul Li e P edic ion o Li hium-Ion Ba e ies. Reliab. Eng. Sys . Sa . 2023,
230, 108920. [C ossRe ]
246.
Ca elani, M.; Ciani, L.; Fan acci, R.; Pa izi, G.; Picano, B. Remaining Use ul Li e Es ima ion o P ognos ics o Li hium-Ion
Ba e ies Based on Recu en Neu al Ne wo k. IEEE T ans. Ins um. Meas. 2021,70, 1–11. [C ossRe ]
247.
Khan, N.; Ullah, F.U.M.; A nan; Ullah, A.; Lee, M.Y.; Baik, S.W. Ba e ies S a e o Heal h Es ima ion ia E icien Neu al Ne wo ks
wi h Mul iple Channel Cha ging P o iles. IEEE Access 2021,9, 7797–7813. [C ossRe ]
248.
Sun, H.; Sun, J.; Zhao, K.; Wang, L.; Wang, K. Da a-D i en ICA-Bi-LSTM-Combined Li hium Ba e y SOH Es ima ion. Ma h.
P obl. Eng. 2022,2022, 9645892. [C ossRe ]
249.
Kim, S.; Choi, Y.Y.; Kim, K.J.; Choi, J.-I. Fo ecas ing S a e-o -Heal h o Li hium-Ion Ba e ies Using Va ia ional Long Sho -Te m
Memo y wi h T ans e Lea ning. J. Ene gy S o age 2021,41, 102893. [C ossRe ]
250.
Tang, T.; Yuan, H. A Hyb id App oach Based on Decomposi ion Algo i hm and Neu al Ne wo k o Remaining Use ul Li e
P edic ion o Li hium-Ion Ba e y. Reliab. Eng. Sys . Sa . 2022,217, 108082. [C ossRe ]
251.
Han, T.; Wang, Z.; Meng, H. End- o-End Capaci y Es ima ion o Li hium-Ion Ba e ies wi h an Enhanced Long Sho -Te m
Memo y Ne wo k Conside ing Domain Adap a ion. J. Powe Sou ces 2022,520, 230823. [C ossRe ]
252.
Ding, P.; Liu, X.; Li, H.; Huang, Z.; Zhang, K.; Shao, L.; Abedinia, O. Use ul Li e P edic ion Based on Wa ele Packe Decomposi ion
and Two-Dimensional Con olu ional Neu al Ne wo k o Li hium-Ion Ba e ies. Renew. Sus ain. Ene gy Re . 2021,148, 111287.
[C ossRe ]
253.
Zhang, Y.; Wik, T.; Be gs öm, J.; Pech , M.; Zou, C. A Machine Lea ning-Based F amewo k o Online P edic ion o Ba e y
Ageing T ajec o y and Li e ime Using His og am Da a. J. Powe Sou ces 2022,526, 231110. [C ossRe ]
254.
Chinomona, B.; Chung, C.; Chang, L.-K.; Su, W.-C.; Tsai, M.-C. Long Sho -Te m Memo y App oach o Es ima e Ba e y Remaining
Use ul Li e Using Pa ial Da a. IEEE Access 2020,8, 165419–165431. [C ossRe ]
255.
Goh, H.H.; Lan, Z.; Zhang, D.; Dai, W.; Ku niawan, T.A.; Goh, K.C. Es ima ion o he S a e o Heal h (SOH) o Ba e ies Using
Disc e e Cu a u e Fea u e Ex ac ion. J. Ene gy S o age 2022,50, 104646. [C ossRe ]
256.
Wei, M.; Gu, H.; Ye, M.; Wang, Q.; Xu, X.; Wu, C. Remaining Use ul Li e P edic ion o Li hium-Ion Ba e ies Based on Mon e
Ca lo D opou and Ga ed Recu en Uni . Ene gy Rep. 2021,7, 2862–2871. [C ossRe ]
257.
Wang, Z.; Zeng, S.; Guo, J.; Qin, T. Remaining Capaci y Es ima ion o Li hium-Ion Ba e ies Based on he Cons an Vol age
Cha ging P o ile. PLoS ONE 2018,13, e0200169. [C ossRe ]
258.
Guo, Y.; Yu, P.; Zhu, C.; Zhao, K.; Wang, L.; Wang, K. A S a e-o -Heal h Es ima ion Me hod Conside ing Capaci y Reco e y o
Li hium Ba e ies. In . J. Ene gy Res. 2022,46, 23730–23745. [C ossRe ]
259.
Gu, X.; See, K.W.; Li, P.; Shan, K.; Wang, Y.; Zhao, L.; Lim, K.C.; Zhang, N. A No el S a e-o -Heal h Es ima ion o he Li hium-Ion
Ba e y Using a Con olu ional Neu al Ne wo k and T ans o me Model. Ene gy 2023,262, 125501. [C ossRe ]
Ene gies 2025,18, 746 71 o 77
260.
Yang, Z.; Wang, Y.; Kong, C. Remaining Use ul Li e P edic ion o Li hium-Ion Ba e ies Based on a Mix u e o Ensemble Empi ical
Mode Decomposi ion and GWO-SVR Model. IEEE T ans. Ins um. Meas. 2021,70, 1–11. [C ossRe ]
261.
Ka a, A. A Da a-D i en App oach Based on Deep Neu al Ne wo ks o Li hium-Ion Ba e y P ognos ics. Neu al Compu . Appl.
2021,33, 13525–13538. [C ossRe ]
262.
Song, S.; Fei, C.; Xia, H. Li hium-Ion Ba e y SOH Es ima ion Based on XGBoos Algo i hm wi h Accu acy Co ec ion. Ene gies
2020,13, 812. [C ossRe ]
263.
Liu, W.; Xu, Y.; Feng, X. A Hie a chical and Flexible Da a-D i en Me hod o Online S a e-o -Heal h Es ima ion o Li-Ion Ba e y.
IEEE T ans. Veh. Technol. 2020,69, 14739–14748. [C ossRe ]
264.
Wu, J.; Fang, L.; Dong, G.; Lin, M. S a e o Heal h Es ima ion o Li hium-Ion Ba e y wi h Imp o ed Radial Basis Func ion Neu al
Ne wo k. Ene gy 2023,262, 125380. [C ossRe ]
265.
Cheng, Y.; Song, D.; Wang, Z.; Lu, C.; Ze houni, N. An Ensemble P ognos ic Me hod o Li hium-Ion Ba e y Capaci y Es ima ion
Based on Time-Va ying Weigh Alloca ion. Appl. Ene gy 2020,266, 114817. [C ossRe ]
266.
Mao, L.; Xu, J.; Chen, J.; Zhao, J.; Wu, Y.; Yao, F. A LSTM-STW and GS-LM Fusion Me hod o Li hium-Ion Ba e y RUL P edic ion
Based on EEMD. Ene gies 2020,13, 2380. [C ossRe ]
267.
Koh z, S.; Xu, Y.; Zheng, Z.; Wang, P. Physics-In o med Machine Lea ning Model o Ba e y S a e o Heal h P ognos ics Using
Pa ial Cha ging Segmen s. Mech. Sys . Signal P ocess. 2022,172, 109002. [C ossRe ]
268.
A deshi i, R.R.; Liu, M.; Ma, C. Mul i a ia e S acked Bidi ec ional Long Sho Te m Memo y o Li hium-Ion Ba e y Heal h
Managemen . Reliab. Eng. Sys . Sa . 2022,224, 108481. [C ossRe ]
269.
Gong, D.; Gao, Y.; Kou, Y.; Wang, Y. S a e o Heal h Es ima ion o Li hium-Ion Ba e y Based on Ene gy Fea u es. Ene gy 2022,
257, 124812. [C ossRe ]
270.
Wu, J.; Cui, X.; Meng, J.; Peng, J.; Lin, M. Da a-D i en T ans e -S acking-Based S a e o Heal h Es ima ion o Li hium-Ion
Ba e ies. IEEE T ans. Ind. Elec on. 2024,71, 604–614. [C ossRe ]
271.
D iscoll, L.; de la To e, S.; Gomez-Ruiz, J.A. Fea u e-Based Li hium-Ion Ba e y S a e o Heal h Es ima ion wi h A i icial Neu al
Ne wo ks. J. Ene gy S o age 2022,50, 104584. [C ossRe ]
272.
Qin, P.; Zhao, L.; Liu, Z. S a e o Heal h P edic ion o Li hium-Ion Ba e y Using a G adien Boos ing-Based Da a-D i en Me hod.
J. Ene gy S o age 2022,47, 103644. [C ossRe ]
273.
Cao, M.; Zhang, T.; Wang, J.; Liu, Y. A Deep Belie Ne wo k App oach o Remaining Capaci y Es ima ion o Li hium-Ion Ba e ies
Based on Cha ging P ocess Fea u es. J. Ene gy S o age 2022,48, 103825. [C ossRe ]
274.
Wei, Y.; Wu, D. P edic ion o S a e o Heal h and Remaining Use ul Li e o Li hium-Ion Ba e y Using G aph Con olu ional
Ne wo k wi h Dual A en ion Mechanisms. Reliab. Eng. Sys . Sa . 2023,230, 108947. [C ossRe ]
275.
Chen, Z.; Chen, L.; Shen, W.; Xu, K. Remaining Use ul Li e P edic ion o Li hium-Ion Ba e y ia a Sequence Decomposi ion and
Deep Lea ning In eg a ed App oach. IEEE T ans. Veh. Technol. 2022,71, 1466–1479. [C ossRe ]
276.
Ma, B.; Yang, S.; Zhang, L.; Wang, W.; Chen, S.; Yang, X.; Xie, H.; Yu, H.; Wang, H.; Liu, X. Remaining Use ul Li e and S a e o
Heal h P edic ion o Li hium Ba e ies Based on Di e en ial The mal Vol amme y and a Deep-Lea ning Model. J. Powe Sou ces
2022,548, 232030. [C ossRe ]
277.
Ruan, H.; Wei, Z.; Shang, W.; Wang, X.; He, H. A i icial In elligence-Based Heal h Diagnos ic o Li hium-Ion Ba e y Le e aging
T ansien S age o Cons an Cu en and Cons an Vol age Cha ging. Appl. Ene gy 2023,336, 120751. [C ossRe ]
278.
Tang, A.; Jiang, Y.; Yu, Q.; Zhang, Z. A Hyb id Neu al Ne wo k Model wi h A en ion Mechanism o S a e o Heal h Es ima ion
o Li hium-Ion Ba e ies. J. Ene gy S o age 2023,68, 107734. [C ossRe ]
279.
Lin, M.; You, Y.; Wang, W.; Wu, J. Ba e y Heal h P ognosis wi h Ga ed Recu en Uni Neu al Ne wo ks and Hidden Ma ko
Model Conside ing Unce ain y Quan i ica ion. Reliab. Eng. Sys . Sa . 2023,230, 108978. [C ossRe ]
280.
Bock a h, S.; Lo en z, V.; P uckne , M. S a e o Heal h Es ima ion o Li hium-Ion Ba e ies wi h a Tempo al Con olu ional Neu al
Ne wo k Using Pa ial Load P o iles. Appl. Ene gy 2023,329, 120307. [C ossRe ]
281.
Xu, H.; Wu, L.; Xiong, S.; Li, W.; Ga g, A.; Gao, L. An Imp o ed CNN-LSTM Model-Based S a e-o -Heal h Es ima ion App oach
o Li hium-Ion Ba e ies. Ene gy 2023,276, 127585. [C ossRe ]
282.
Liu, Y.; Sun, J.; Shang, Y.; Zhang, X.; Ren, S.; Wang, D. A No el Remaining Use ul Li e P edic ion Me hod o Li hium-Ion
Ba e y Based on Long Sho -Te m Memo y Ne wo k Op imized by Imp o ed Spa ow Sea ch Algo i hm. J. Ene gy S o age 2023,
61, 106645. [C ossRe ]
283.
Bama i, S.; Chaoui, H. Li hium-Ion Ba e ies Long Ho izon Heal h P ognos ic Using Machine Lea ning. IEEE T ans. Ene gy
Con e s. 2022,37, 1176–1186. [C ossRe ]
284. Cao, M.; Zhang, T.; Yu, B.; Liu, Y. A Me hod o In e al P edic ion o Sa elli e Ba e y S a e o Heal h Based on Sample En opy.
IEEE Access 2019,7, 141549–141561. [C ossRe ]
285.
Zhang, Z.; Li, L.; Li, X.; Hu, Y.; Huang, K.; Xue, B.; Wang, Y.; Yu, Y. S a e-o -Heal h Es ima ion o he Li hium-Ion Ba e y Based
on G adien Boos ing Decision T ee wi h Au onomous Selec ion o Excellen Fea u es. In . J. Ene gy Res. 2022,46, 1756–1765.
[C ossRe ]
Ene gies 2025,18, 746 72 o 77
286.
Su, C.; Chen, H.; Wen, Z. P edic ion o Remaining Use ul Li e o Li hium-Ion Ba e y wi h Mul iple Heal h Indica o s. Eksploa . i
Niezawodn. 2021,23, 176–183. [C ossRe ]
287.
Chen, Z.; Xue, Q.; Wu, Y.; Shen, S.; Zhang, Y.; Shen, J. Capaci y P edic ion and Valida ion o Li hium-Ion Ba e ies Based on Long
Sho -Te m Memo y Recu en Neu al Ne wo k. IEEE Access 2020,8, 172783–172798. [C ossRe ]
288.
He, J.; Tian, Y.; Wu, L. A Hyb id Da a-D i en Me hod o Rapid P edic ion o Li hium-Ion Ba e y Capaci y. Reliab. Eng. Sys . Sa .
2022,226, 108674. [C ossRe ]
289.
Lin, C.; Xu, J.; Hou, J.; Liang, Y.; Mei, X. Ensemble Me hod wi h He e ogeneous Models o Ba e y S a e-o -Heal h Es ima ion.
IEEE T ans. Ind. In o m. 2023,19, 10160–10169. [C ossRe ]
290.
Jia, J.; Wang, K.; Shi, Y.; Wen, J.; Pang, X.; Zeng, J. A Mul i-Scale S a e o Heal h P edic ion F amewo k o Li hium-Ion Ba e ies
Conside ing he Tempe a u e Va ia ion du ing Ba e y Discha ge. J. Ene gy S o age 2021,42, 103076. [C ossRe ]
291.
Vakha ia, V.; Shah, M.; Nai , P.; Bo ade, H.; Sahlo , P.; Wankhede, V. Es ima ion o Li hium-Ion Ba e y Discha ge Capaci y by
In eg a ing Op imized Explainable-AI and S acked LSTM Model. Ba e ies 2023,9, 125. [C ossRe ]
292.
Li, D.; Yang, L. Remaining Use ul Li e P edic ion o Li hium Ba e y Based on Sequen ial CNN–LSTM Me hod. J. Elec ochem.
Ene gy Con e s. S o age 2021,18, 041005. [C ossRe ]
293.
Fan, Z.; Zi-xuan, X.; Ming-hu, W. S a e o Heal h Es ima ion o Li-Ion Ba e y Using Cha ac e is ic Vol age In e als and Gene ic
Algo i hm Op imized Back P opaga ion Neu al Ne wo k. J. Ene gy S o age 2023,57, 106277. [C ossRe ]
294.
Wei, Z.; Han, X.; Li, J. S a e o Heal h Assessmen o Echelon U iliza ion Ba e ies Based on Deep Neu al Ne wo k Lea ning wi h
E o Co ec ion. J. Ene gy S o age 2022,51, 104428. [C ossRe ]
295.
Tang, T.; Yuan, H. An Indi ec Remaining Use ul Li e P ognosis o Li-Ion Ba e ies Based on Heal h Indica o and No el A i icial
Neu al Ne wo k. J. Ene gy S o age 2022,52, 104701. [C ossRe ]
296.
Chen, X.; Liu, Z.; Sheng, H.; Wu, K.; Mi, J.; Li, Q. T ans e Lea ning Based Remaining Use ul Li e P edic ion o Li hium-Ion
Ba e y Conside ing Capaci y Regene a ion Phenomenon. J. Ene gy S o age 2024,76, 109798. [C ossRe ]
297.
Bao, Z.; Jiang, J.; Zhu, C.; Gao, M. A New Hyb id Neu al Ne wo k Me hod o S a e-o -Heal h Es ima ion o Li hium-Ion Ba e y.
Ene gies 2022,15, 4399. [C ossRe ]
298.
Wang, Z.; Liu, N.; Chen, C.; Guo, Y. Adap i e Sel -A en ion LSTM o RUL P edic ion o Li hium-Ion Ba e ies. In . Sci. 2023,635,
398–413. [C ossRe ]
299.
Yao, X.-Y.; Chen, G.; Pech , M.; Chen, B. A No el G aph-Based F amewo k o S a e o Heal h P edic ion o Li hium-Ion Ba e y. J.
Ene gy S o age 2023,58, 106437. [C ossRe ]
300.
Mao, L.; Hu, H.; Chen, J.; Zhao, J.; Qu, K.; Jiang, L. Online S a e-o -Heal h Es ima ion Me hod o Li hium-Ion Ba e y Based on
CEEMDAN o Fea u e Analysis and RBF Neu al Ne wo k. IEEE J. Eme g. Sel. Top. Powe Elec on. 2023,11, 187–200. [C ossRe ]
301.
Bi, J.; Lee, J.-C.; Liu, H. Pe o mance Compa ison o Long Sho -Te m Memo y and a Tempo al Con olu ional Ne wo k o S a e
o Heal h Es ima ion o a Li hium-Ion Ba e y Using I s Cha ging Cha ac e is ics. Ene gies 2022,15, 2448. [C ossRe ]
302.
Cui, Y.; Chen, Y. P ognos ics o Li hium-Ion Ba e ies Based on Capaci y Regene a ion Analysis and Long Sho -Te m Memo y
Ne wo k. IEEE T ans. Ins um. Meas. 2022,71, 1–13. [C ossRe ]
303.
Chen, Z.; Zhang, S.; Shi, N.; Li, F.; Wang, Y.; Cui, J. Online S a e-o -Heal h Es ima ion o Li hium-Ion Ba e y Based on Rele ance
Vec o Machine wi h Dynamic In eg a ion. Appl. So Compu . 2022,129, 109615. [C ossRe ]
304.
Zhao, J.; Zhu, Y.; Zhang, B.; Liu, M.; Wang, J.; Liu, C.; Zhang, Y. Me hod o P edic ing SOH and RUL o Li hium-Ion Ba e y Based
on he Combina ion o LSTM and GPR. Sus ainabili y 2022,14, 11865. [C ossRe ]
305.
Zhang, D.; Li, W.; Han, X.; Lu, B.; Zhang, Q.; Bo, C. E ol ing Elman Neu al Ne wo ks Based S a e-o -Heal h Es ima ion o
Sa elli e Li hium-Ion Ba e ies. J. Ene gy S o age 2023,59, 106571. [C ossRe ]
306.
Zhu, M.; Ouyang, Q.; Wan, Y.; Wang, Z. Remaining Use ul Li e P edic ion o Li hium-Ion Ba e ies: A Hyb id App oach o
G ey–Ma ko Chain Model and Imp o ed Gaussian P ocess. IEEE J. Eme g. Sel. Top. Powe Elec on. 2023,11, 143–153. [C ossRe ]
307.
Ang, E.Y.M.; Paw, Y.C. E icien Linea P edic i e Model wi h Sho Te m Fea u es o Li hium-Ion Ba e ies S a e o Heal h
Es ima ion. J. Ene gy S o age 2021,44, 103409. [C ossRe ]
308.
Ouyang, M.; Shen, P. P edic ion o Remaining Use ul Li e o Li hium Ba e ies Based on WOA-VMD and LSTM. Ene gies 2022,15,
8918. [C ossRe ]
309.
Zhang, Y.; Wang, Y.; Xia, Y.; Chen, W. A Deep Lea ning App oach o Es ima e he S a e o Heal h o Li hium-Ion Ba e ies unde
Va ied and Incomple e Wo king Condi ions. J. Ene gy S o age 2023,58, 106323. [C ossRe ]
310.
Ma, B.; Zhang, L.; Yu, H.; Zou, B.; Wang, W.; Zhang, C.; Yang, S.; Liu, X. End-Cloud Collabo a ion Me hod Enables Accu a e S a e
o Heal h and Remaining Use ul Li e Online Es ima ion in Li hium-Ion Ba e ies. J. Ene gy Chem. 2023,82, 1–17. [C ossRe ]
311.
Gao, M.; Bao, Z.; Zhu, C.; Jiang, J.; He, Z.; Dong, Z.; Song, Y. HFCM-LSTM: A No el Hyb id F amewo k o S a e-o -Heal h
Es ima ion o Li hium-Ion Ba e y. Ene gy Rep. 2023,9, 2577–2590. [C ossRe ]
312.
Li, L.; Li, Y.; Mao, R.; Li, L.; Hua, W.; Zhang, J. Remaining Use ul Li e P edic ion o Li hium-Ion Ba e ies wi h a Hyb id Model
Based on TCN-GRU-DNN and Dual A en ion Mechanism. IEEE T ans. T ansp. Elec i . 2023,9, 4726–4740. [C ossRe ]
Ene gies 2025,18, 746 73 o 77
313.
Xu, J.; Liu, B.; Zhang, G.; Zhu, J. S a e-o -Heal h Es ima ion o Li hium-Ion Ba e ies Based on Pa ial Cha ging Segmen and
S acking Model Fusion. Ene gy Sci. Eng. 2023,11, 383–397. [C ossRe ]
314.
Zou, L.; Wen, B.; Wei, Y.; Zhang, Y.; Yang, J.; Zhang, H. Online P edic ion o Remaining Use ul Li e o Li-Ion Ba e ies Based on
Discha ge Vol age Da a. Ene gies 2022,15, 2237. [C ossRe ]
315.
Yang, Y. A Machine-Lea ning P edic ion Me hod o Li hium-Ion Ba e y Li e Based on Cha ge P ocess o Di e en Applica ions.
Appl. Ene gy 2021,292, 116897. [C ossRe ]
316.
Deng, Z.; Lin, X.; Cai, J.; Hu, X. Ba e y Heal h Es ima ion wi h Deg ada ion Pa e n Recogni ion and T ans e Lea ning. J. Powe
Sou ces 2022,525, 231027. [C ossRe ]
317.
Zhang, Y.; Peng, Z.; Guan, Y.; Wu, L. P ognos ics o Ba e y Cycle Li e in he Ea ly-Cycle S age Based on Hyb id Model. Ene gy
2021,221, 119901. [C ossRe ]
318.
Hsu, C.-W.; Xiong, R.; Chen, N.-Y.; Li, J.; Tsou, N.-T. Deep Neu al Ne wo k Ba e y Li e and Vol age P edic ion by Using Da a o
One Cycle Only. Appl. Ene gy 2022,306, 118134. [C ossRe ]
319.
Zhang, Q.; Yang, L.; Guo, W.; Qiang, J.; Peng, C.; Li, Q.; Deng, Z. A Deep Lea ning Me hod o Li hium-Ion Ba e y Remaining
Use ul Li e P edic ion Based on Spa se Segmen Da a ia Cloud Compu ing Sys em. Ene gy 2022,241, 122716. [C ossRe ]
320.
He ing, P.; Balaji Gopal, C.; Aykol, M.; Mon oya, J.H.; Anapolsky, A.; A ia, P.M.; Gen , W.; Hummelshøj, J.S.; Hung, L.;
Kwon, H.-K.; e al. BEEP: A Py hon Lib a y o Ba e y E alua ion and Ea ly P edic ion. So wa eX 2020,11, 100506. [C ossRe ]
321.
Sanz-Go acha egui, I.; Pas o -Flo es, P.; Pajo ic, M.; Wang, Y.; O lik, P.V.; Be nal-Ruiz, C.; Bono-Nuez, A.; A al-Se il, J.S.
Remaining Use ul Li e Es ima ion o LFP Cells in Second-Li e Applica ions. IEEE T ans. Ins um. Meas. 2021,70, 1–10. [C ossRe ]
322.
Ma, G.; Xu, S.; Yang, T.; Du, Z.; Zhu, L.; Ding, H.; Yuan, Y. A T ans e Lea ning-Based Me hod o Pe sonalized S a e o Heal h
Es ima ion o Li hium-Ion Ba e ies. IEEE T ans. Neu al Ne w. Lea n. Sys . 2024,35, 759–769. [C ossRe ]
323.
Zhang, Y.; Zhao, M. Cloud-Based in-Si u Ba e y Li e P edic ion and Classi ica ion Using Machine Lea ning. Ene gy S o age Ma e .
2023,57, 346–359. [C ossRe ]
324.
Guo, F.; Wu, X.; Liu, L.; Ye, J.; Wang, T.; Fu, L.; Wu, Y. P edic ion o Remaining Use ul Li e and S a e o Heal h o Li hium Ba e ies
Based on Time Se ies Fea u e and Sa i zky-Golay Fil e Combined wi h Ga ed Recu en Uni Neu al Ne wo k. Ene gy 2023,
270, 126880. [C ossRe ]
325.
Che, Y.; S oe, D.-I.; Hu, X.; Teodo escu, R. Semi-Supe ised Sel -Lea ning-Based Li e ime P edic ion o Ba e ies. IEEE T ans. Ind.
In o m. 2023,19, 6471–6481. [C ossRe ]
326.
Xu, L.; Deng, Z.; Xie, Y.; Lin, X.; Hu, X. A No el Hyb id Physics-Based and Da a-D i en App oach o Deg ada ion T ajec o y
P edic ion in Li-Ion Ba e ies. IEEE T ans. T ansp. Elec i . 2023,9, 2628–2644. [C ossRe ]
327.
Chen, D.; Zhang, W.; Zhang, C.; Sun, B.; Cong, X.; Wei, S.; Jiang, J. A No el Deep Lea ning-Based Li e P edic ion Me hod o
Li hium-Ion Ba e ies wi h S ong Gene aliza ion Capabili y unde Mul iple Cycle P o iles. Appl. Ene gy 2022,327, 120114.
[C ossRe ]
328.
Fei, Z.; Zhang, Z.; Yang, F.; Tsui, K.-L. A Deep A en ion-Assis ed and Memo y-Augmen ed Tempo al Con olu ional Ne wo k
Based Model o Rapid Li hium-Ion Ba e y Remaining Use ul Li e P edic ions wi h Limi ed Da a. J. Ene gy S o age 2023,
62, 106903. [C ossRe ]
329.
Gong, D.; Gao, Y.; Kou, Y.; Wang, Y. Ea ly P edic ion o Cycle Li e o Li hium-Ion Ba e ies Based on E olu iona y Compu a ion
and Machine Lea ning. J. Ene gy S o age 2022,51, 104376. [C ossRe ]
330.
Zhang, S.; Liu, Z.; Su, H. S a e o Heal h Es ima ion o Li hium-Ion Ba e ies on Few-Sho Lea ning. Ene gy 2023,268, 126726.
[C ossRe ]
331.
Zhou, K.Q.; Qin, Y.; Yuen, C. T ans e -Lea ning-Based S a e-o -Heal h Es ima ion o Li hium-Ion Ba e y wi h Cycle Synch o-
niza ion. IEEE/ASME T ans. Mecha on. 2023,28, 692–702. [C ossRe ]
332.
Kim, S.; Jung, H.; Lee, M.; Choi, Y.Y.; Choi, J.-I. Model-F ee Recons uc ion o Capaci y Deg ada ion T ajec o y o Li hium-Ion
Ba e ies Using Ea ly Cycle Da a. eT anspo a ion 2023,17, 100243. [C ossRe ]
333.
Xu, Q.; Wu, M.; Khoo, E.; Chen, Z.; Li, X. A Hyb id Ensemble Deep Lea ning App oach o Ea ly P edic ion o Ba e y Remaining
Use ul Li e. IEEE/CAA J. Au om. Sin. 2023,10, 177–187. [C ossRe ]
334.
Celik, B.; Sand , R.; dos San os, L.C.P.; Spa schek, R. P edic ion o Ba e y Cycle Li e Using Ea ly-Cycle Da a, Machine Lea ning
and Da a Managemen . Ba e ies 2022,8, 266. [C ossRe ]
335.
Chen, Z.; Chen, L.; Ma, Z.; Xu, K.; Zhou, Y.; Shen, W. Join Modeling o Ea ly P edic ions o Li-Ion Ba e y Cycle Li e and
Deg ada ion T ajec o y. Ene gy 2023,277, 127633. [C ossRe ]
336.
Ansa i, S.; Ayob, A.; Hossain Lipu, M.S.; Hussain, A.; Md Saad, M.H. Jelly ish Op imized Recu en Neu al Ne wo k o S a e o
Heal h Es ima ion o Li hium-Ion Ba e ies. Expe Sys . Appl. 2024,238, 121904. [C ossRe ]
337.
Amogne, Z.E.; Wang, F.-K.; Chou, J.-H. T ans e Lea ning Based on T ans e abili y Measu es o S a e o Heal h P edic ion o
Li hium-Ion Ba e ies. Ba e ies 2023,9, 280. [C ossRe ]
338.
Khei khah-Rad, E.; Pa a eh, A.; Moeini-Agh aie, M.; Dehghanian, P. A Da a-D i en S a e-o -Heal h Es ima ion Model o
Li hium-Ion Ba e ies Using Re e enced-Based Cha ging Time. IEEE T ans. Powe Deli . 2023,38, 3406–3416. [C ossRe ]
