Ci a ion: Velásquez, D.; Vallejo, P.;
To o, M.; Od iozola, J.; Mo eno, A.;
Na e an, G.; Gi aldo, M.; Maiza, M.;
Sie a, B. EDAR 4.0: Machine
Lea ning and Visual Analy ics o
Was ewa e Managemen .
Sus ainabili y 2024,16, 3578. h ps://
doi.o g/10.3390/su16093578
Academic Edi o : And eas
Angelakis
Recei ed: 4 Ma ch 2024
Re ised: 8 Ap il 2024
Accep ed: 18 Ap il 2024
Published: 24 Ap il 2024
Copy igh : © 2024 by he au ho s.
Licensee MDPI, Basel, Swi ze land.
This a icle is an open access a icle
dis ibu ed unde he e ms and
condi ions o he C ea i e Commons
A ibu ion (CC BY) license (h ps://
c ea i ecommons.o g/licenses/by/
4.0/).
sus ainabili y
A icle
EDAR 4.0: Machine Lea ning and Visual Analy ics o
Was ewa e Managemen
Da id Velásquez 1,2,3,4,* , Paola Vallejo 1, Mau icio To o 1, Juan Od iozola 3, Ai o Mo eno 5,
Go ka Na e an 6, Michael Gi aldo 2, Mikel Maiza 3and Basilio Sie a 4
1RID on In o ma ion Technologies and Communica ions Resea ch G oup (GIDITIC), Uni e sidad EAFIT,
Ca e a 49 No. 7 Su -50
, Medellín 050022, Colombia; [email p o ec ed] (P.V.); m o [email p o ec ed] (M.T.)
2Indus y, Ma e ials and Ene gy A ea, Uni e sidad EAFIT, Ca e a 49 No. 7 Su -50,
Medellín 050022, Colombia; [email p o ec ed]
3Depa men o Da a In elligence o Ene gy and Indus ial P ocesses, Vicom ech Founda ion,
Basque Resea ch and Technology Alliance (BRTA), 20014 Donos ia-San Sebas ián, Spain;
[email p o ec ed] (J.O.); [email p o ec ed]g (M.M.)
4Depa men o Compu e Science and A i icial In elligence, Uni e si y o Basque Coun y,
Manuel La dizabal Ibilbidea, 1, 20018 Donos ia-San Sebas ián, Spain; [email p o ec ed]
5Depa men o R&D, Ibe má ica, Ce cas Bajas, 7 in .-O ice 2, 01001 Vi o ia-Gas eiz, Spain;
[email p o ec ed]
6Depa men o R&D, Gi oa-Veolia, Laida Bidea, Building 407, 48170 Zamudio, Spain;
[email p o ec ed]
*Co espondence: [email p o ec ed]
Abs ac : Was ewa e ea men plan (WWTP) ope a ions manage massi e amoun s o da a ha can
be ga he ed wi h new Indus y 4.0 echnologies such as he In e ne o Things and Big Da a. These
da a a e c i ical o allow he was ewa e ea men indus y o imp o e i s ope a ion, con ol, and
main enance. Howe e , he da a a ailable need o be imp o ed and en iched, pa ly due o hei
high dimensionali y and low eliabili y, and he lack o app op ia e da a analysis and p ocessing
ools o such sys ems. This pape p esen s a isual analy ics-based pla o m o WWTP ha allows
use s o iden i y ela ionships among da a h ough da a inspec ion. The esul s show ha he ool
de eloped and implemen ed o a ull-scale WWTP allows ope a o s o cons uc machine lea ning
(ML) models o wa e quali y and o he wa e ea men p ocess a iables. Consequen ly, analyzing
and op imizing plan ope a ion scena ios can enhance key a iables, including ene gy, eagen
consump ion, and wa e quali y. This imp o emen acili a es he de elopmen o a mo e sus ainable
WWTP, con ibu ing o a bene icial en i onmen al impac . Domain expe s alida ed he a iables
in luencing he c ea ed ML models and p o ed hei app op ia eness.
Keywo ds: da a-d i en modeling; machine lea ning; Indus y 4.0; isual analy ics; was ewa e
managemen ; was ewa e ea men plan (WWTP)
1. In oduc ion
Newly connec ed indus y objec s a e gene a ing as amoun s o da a a an inc easing
a e, which mus be s o ed, p ocessed, and moni o ed in eal ime o make in o med
decisions ha op imize p oduc ion in Indus y 4.0 ac o ies. The challenge lies in e ec i ely
isualizing hese newly gene a ed da a, including educing hei dimensionali y and
isualizing mul i a ia e eal- ime da a.
An ad anced app oach o da a p ocessing and isualiza ion ha can be implemen ed
is isual analy ics (VA). Keim e al. [
1
] de ined VA as combining au oma ed analysis
echniques wi h in e ac i e isualiza ions o enable adequa e unde s anding, easoning,
and decision-making based on ex ensi e and complex da a se s. The ocus o VA is o c ea e
new ools ha allow use s o (i) syn hesize in o ma ion and gain new insigh s om la ge
and he e ogeneous da a se s, (ii) de ec he cu en s a e o sys ems and disco e po en ial
Sus ainabili y 2024,16, 3578. h ps://doi.o g/10.3390/su16093578 h ps://www.mdpi.com/jou nal/sus ainabili y
Sus ainabili y 2024,16, 3578 2 o 17
new s a es, and (iii) p o ide eal- ime assessmen s and make in o med ac ions based on
hese assessmen s.
Keim e al. [
1
] p oposed six challenges o VA: (i) scalabili y wi h la ge da a olumes
and high dimensionali y, (ii) g aphical ep esen a ion o da a quali y, (iii) isual ep e-
sen a ion o le els o de ail, (i ) new display in e aces such as la ge-scale powe walls,
( ) e alua ion amewo ks o VA, and ( i) e eshing in e ac ions in eal ime (e.g., wi h
esponse imes less han 100 ms). Many o hese challenges emain un esol ed o his day.
Diez-Oli an e al. [
2
] ecen ly disco e ed ha u ilizing VA o enhance unde s and-
abili y in Indus y 4.0 poses a new challenge. This limi a ion, ac ing as a ba ie o he
widesp ead adop ion o da a-based analysis, lies in he indus ial plan ope a o ’s assimila-
ion o in o ma ion. In da a analysis, he in o ma ion gene a ed by deployed models canno
be eadily p ocessed by non-specialized pe sonnel unless p ep ocessing s a egies a e de-
ised, acili a ing an imp o ed and mo e in ui i e unde s anding o he
cap u ed pa e ns.
A case s udy o VA is was ewa e ea men plan s (WWTPs). WWTPs can be managed
by seeking op imal p ocess condi ions and iden i ying essen ial ac o s, ea u es, o pa e ns
o da a-suppo ed decision-making. Newha e al. [
3
] highligh ed ha WWTP ope a o s
usually s o e a su icien ly la ge amoun o his o ical da a. In addi ion, ecen ad ancemen s
in da a-d i en p ocess con ol and pe o mance analysis and mo e subs an ial compu a ion
powe “could p o ide he was ewa e ea men indus y wi h an oppo uni y o educe
cos s and imp o e ope a ions” [
3
]. One sus ainabili y p oblem his esea ch add esses
conce ns he equi emen o mo e sus ainable ope a ions wi hin WWTPs, ocusing on he
simula ion and op imiza ion o p ocess a iables such as ene gy and eagen consump ion,
and wa e quali y enhancemen . A mo e sus ainable WWTP is en i onmen ally bene icial
and c ucial o ackling signi ican en i onmen al challenges associa ed wi h WWTP e lu-
en wa e quali y. When discha ged in o na u al wa e cou ses like i e s o seas, e luen
con aining high le els o ni a es can lead o inc eased eu ophica ion [
4
], posing se e e
ecological isks. This esea ch highligh s he impo ance o op imizing WWTP p ocesses
h ough da a-suppo ed decision-making o mi iga e hese en i onmen al impac s. How-
e e , he limi ed in es men s in ins umen a ion, con ol, and au oma ion o WWTPs and
he need o a da a-science backg ound o WWTP p o essionals a e limi a ions o making
he bes o he da a.
A key challenge in he decision-making p ocess du ing he Big Da a e a in ol es
iden i ying ele an da a and ex ac ing meaning ul insigh s om hem. To add ess his
p oblem in he con ex o WWTPs, p ojec Es ación Depu ado a de Aguas Residuales
(EDAR 4.0) aims o de elop a se o WWTP ope a ion and managemen sys ems by combin-
ing (i) cloud compu ing, (ii) da a in elligence, and (iii) isual analy ics. EDAR 4.0 aims o
p o ide g ea e da a s o age, p ocessing, compu a ion, and decision-making capabili ies o
WWTP ope a ion [
5
]. EDAR 4.0’s esul s we e es ed and alida ed in a ull-scale municipal
WWTP: La Ca uja (Za agoza, Spain), ope a ed by Veolia.
Fi e a iables ela ed o WWTP’s ope a ion and managemen we e analyzed in EDAR
4.0: biological oxygen demand-5 (
BOD5
), o al chemical oxygen demand (
TCOD
), o al
Kjeldahl ni ogen (
TKN
), o al phospho ous (
TP
), and o al suspended solids (
TSS
), which
a e conside ed as con aminan s. These a iables a e no selec ed andomly bu a e he
a iables ha Eu opean Di ec i e 91/271/EEC [
6
] es ablishes as quali y equi emen s o be
ul illed in he e luen o a WWTP. Likewise, in he case ha applies, as he WWTP is loca ed
in a egion (A agon, Spain) decla ed as an a ea sensi i e o eu ophica ion, speci ic alues
o o al phospho us and o al ni ogen a e applied. Table 1shows he quali y equi emen s
aken as a e e ence in his p ojec based on he p e iously men ioned Eu opean Di ec i e,
whe e columns Absolu e Values and Pe o mances ep esen he maximum concen a ion o
con aminan s pe mi ed and he minimum con aminan s emo al demanded, espec i ely,
by he Eu opean Di ec i e men ioned abo e. I can be no ed ha his di ec i e allows he
WWTP o comply wi h absolu e alues o pe o mances o each con aminan .
Sus ainabili y 2024,16, 3578 3 o 17
Table 1. Wa e quali y equi emen s om Eu opean Di ec i e 91/271/EEC.
Va iable Absolu e Values Pe o mances
BOD525 mgO2/L 70%
TCOD 125 mgO2/L 75%
TKN 10 mg/L 90%
TP 1 mg/L 80%
TSS 35 mg/L 70%
This pape in oduces a pla o m designed o acili a e he c ea ion o da a-d i en
models o simula ing, p edic ing, and op imizing WWTPs. Two modules comp ise his
pla o m: (i) a module dedica ed o he moni o ing and p edic ion o wa e quali y, ensu -
ing compliance wi h en i onmen al s anda ds and enhancing he sus ainabili y o wa e
esou ces, and (ii) a module ocused on he de elopmen o ML models o wa e quali y
and ene gy managemen . This enables he e icien analysis o u u e scena ios and he
op imiza ion o WWTP ope a ions. By signi ican ly educing ene gy and eagen con-
sump ion and imp o ing wa e quali y, his pla o m con ibu es o he en i onmen al
sus ainabili y o WWTPs, he eby minimizing hei ecological oo p in and p omo ing a
posi i e en i onmen al impac .
In wha ollows, a b ie s a e o he a is p esen ed in Sec ion 2. Then, he me hodology
is shown in Sec ion 3, esul s in Sec ion 4, and a discussion in Sec ion 5. Finally, conclusions
and u u e wo k di ec ions a e p oposed in Sec ion 6.
2. S a e o he A
The s a e o he a is di ided in o h ee pa s. The i s pa summa izes di e en
wo ks on VA. The second pa p esen s esea ch on model-based was ewa e managemen .
Finally, he las pa explains esea ch on da a-based was ewa e managemen .
2.1. Visual Analy ics
VA combines in e ac i e isualiza ions wi h da a analysis and machine lea ning (ML)
o empowe people o analyze, explo e, and unde s and ex ensi e da a [
7
]. The amewo k
p oposed by [
8
] can gene alize he VA p ocess (see Figu e 1). The i s s ep is acqui ing da a
s o ed in a da abase o om a da a s eam. These da a a e hen analyzed and p ocessed o
ex ac he mos c i ical ea u es in he isualiza ion s age. An image is gene a ed du ing
he isualiza ion s age, ep esen ing hese p ocessed and selec ed da a, o by he use ’s
speci ica ions. Subsequen ly, he use obse es and comp ehends he image, de i ing
insigh s and knowledge. This s age may be epea ed as long as he use looks h ough
he image. Finally, he use may gene a e hypo heses, which will be de ailed h ough an
explo a ion and analysis s age. Fu he mo e, a new analysis may be equi ed, ansla ing
in o a speci ica ion s age, whe e he use can in e ac wi h he cu en isualiza ion o
gene a e new knowledge.
Acco ding o Diez-Oli an e al. [
2
], VA has eme ged as a p omising discipline o
isually adap he disco e ed insigh s and op imally p esen esul s o di e en human
p o iles. These aspec s a e essen ial in eal-use cases o deploy models o da a analysis in
indus ial plan s wi h minimum usabili y and p ac ical u ili y gua an ee.
As an example o VA, Li, and Ma [
9
] p oposed P6, a decla a i e language o apidly
speci ying he design o VA sys ems ha in eg a e ML and isualiza ion me hods o in e -
ac i e isual analysis. P6 was mo i a ed by h ee goals: in e ac i e ML and isualiza ion
( o acili a e au oma ed analysis), in e ac i e and scalable sys ems ( o p ocess and isualize
la ge da a se s), and decla a i e VA ( o c ea e in e ac i e isualiza ion applica ions). In P6,
he speci ica ion’s basic uni is a pipeline composed o he ollowing speci ica ions: da a,
analysis, iew layou , isualiza ions, and in e ac ions.
Sus ainabili y 2024,16, 3578 4 o 17
Figu e 1. Visual analy ics p ocess amewo k adap ed om [8].
Kalinin e al. [
10
] p esen s a web-based isual analy ics amewo k, enabling easy
in eg a ion o di e en componen s o da a managemen , analysis, and isualiza ion. The
pla o m inco po a es a ious ools o impo ing da a, displaying in o ma ion, s o ing
da a, in e ac i e isualiza ion, s a is ical analysis, and ML.
Nawaz e al. [
11
] de eloped an in elligen human–machine in e ace (HMI) called
ANKSys ha allows ope a ion and decision suppo o he anae obic ammonium oxida-
ion (ANAMMOX) p ocess in WWTPs. This ool in eg a es so sensing, decision-making,
and model simula ion o supe iso y con ol, which consis s o an a i icial neu al ne wo k,
a Kalman il e , and a p incipal componen analysis algo i hm.
Addi ionally, Li and Ma [9] p oposed ha he decla a i e speci ica ion o VA allows
non-specialis s o de elop ad anced da a analy ics and communica ion solu ions ha
combine he bes o human and a i icial in elligence. Acco ding o Ende e al. [
12
], VA
sys ems combine ML (o o he analy ic echniques) wi h in e ac i e da a isualiza ion o
acili a e insigh and analy ical easoning. Ende desc ibed h ee ca ego ies o models and
amewo ks: (i) models mean o desc ibe people’s cogni i e s ages o analyzing da a;
(ii) models and amewo ks ha desc ibe in e ac ion and in o ma ion design o isual ana-
ly ic applica ions; and (iii) ML amewo ks ha emphasize he impo ance o aining da a
and g ound u h o gene a e accu a e and e ec i e compu a ional models.
Keim e al. [13]
men ioned ha he mos common ML algo i hms used wi h VA a e (i) dimension educ ion,
(ii) clus e ing, (iii) classi ica ion, and (i ) eg ession.
As s a ed by Liu e al. [
14
], “in e ac i e model analysis, he p ocess o unde s anding,
diagnosing, and e ining an ML model wi h he help o in e ac i e isualiza ion, is e y
impo an o use s o sol e eal-wo ld a i icial in elligence and da a mining p oblems
e icien ly”. Liu e al.’s pape p esen s a classi ica ion o ele an wo k in VA in o h ee
ca ego ies: (i) unde s anding, (ii) diagnosis, and (iii) e inemen . Liu highligh s ha
many echniques gene a e s a ic images o indica e which pa s o an image a e mos
impo an o he classi ica ion. Howe e , in e ac i e isualiza ion plays a c i ical ole in
model unde s anding and analysis o help people gain insigh in o a ious ML models.
The e o e, ou p oposal add esses he dynamic c ea ion o demand-d i en models, such
as a wa e -quali y model, and how hei esponses con ibu e o he comp ehension o
speci ic a iables.
Sus ainabili y 2024,16, 3578 5 o 17
Massi e da a se s and complex, long- unning analy ics a e common in a ious do-
mains. S olpe e al. [
15
] in oduced he p og essi e isual analy ics (PVA) concep . PVA is
a wo k low ha p o ides he use wi h meaning ul in e media e esul s i he inal esul ’s
compu a ion is oo cos ly. Based on hese in e media e esul s, he use can isualize,
analyze, and in e p e pa ial esul s be o e ob aining he comple e esul s.
VA, in he indus ial con ex , has been used widely. Sun e al. [
16
] p oposed Plan-
ningVis, a VA sys em o suppo he explo a ion and compa ison o p oduc ion plans wi h
h ee le els o de ails: a plan o e iew p esen ing he o e all di e ence be ween plans,
a p oduc iew isualizing a ious p ope ies o indi idual p oduc s, and a p oduc ion
de ail iew displaying he p oduc dependency and he daily p oduc ion de ails in ela ed
ac o ies. Finally, Wu e al. [
17
] epo ed he design and implemen a ion o an in e ac i e VA
sys em, which helps manage s and ope a o s a manu ac u ing si es le e age hei domain
knowledge and apply subs an ial human judgmen s o guide he au oma ed analy ical ap-
p oaches, hus gene a ing unde s andable and us able esul s o eal-wo ld applica ions.
Ou sys em in eg a es ad anced analy ical algo i hms (e.g., Gaussian mix u e model wi h
a Bayesian amewo k) and in ui i e isualiza ion designs o p o ide a comp ehensi e and
adap i e semi-supe ised solu ion o equipmen condi ion moni o ing.
2.2. Model-Based Was ewa e Managemen
A b ie s a e-o - he-a was ewa e ea men plan modeling based on o dina y di e -
en ial equa ions (ODEs) is p esen ed in wha ollows.
The mos common app oach o op imize he p ocess ope a ion agains luc ua ing
in luen wa e quali y is o apply p ocess con ol and simula ion o de i e he op imal
ope a ion me hod. ODEs ha e been widely used o p ocess simula ion. To simula e
WWTPs using ODEs, i is essen ial o i s model he p ocess’s s eady s a e unde a gi en se
o dis u bances and ope a ing condi ions. Howe e , a disad an age is ha he calcula ion
ime is ex ended when analyzing he ODEs. Jong ack e al. [
18
] p oposed an imp o ed
New on–Raphson me hod o sho en he compu a ion ime. The abo e shows ha he e is
s ill ac i e esea ch on he simula ion o was ewa e ea men plan s using ODEs.
In ano he wo k, Flo es-Alsina e al. [
19
] de eloped a plan -wide aqueous phase chem-
is y model desc ibing pH a ia ions in e aced wi h indus y-s anda d models. Flo es-
Alsina e al. o mula ed he gene al equilib ia as a se o di e en ial-algeb aic equa ions
(DAEs) ins ead o ODEs o enhance simula ion speed. Addi ionally, Flo es-Alsina e al.
applied a mul idimensional e sion o he New on–Raphson algo i hm o handle mul iple
algeb aic in e dependencies.
I is impo an o men ion ha he In e na ional Wa e Associa ion (IWA) benchma k
simula ion model has been a ailable o se e al yea s o c ea e pla o ms o con ol
s a egy benchma king o ac i a ed sludge p ocesses. Jeppsson e al. [
20
] ex ended he
IWA benchma k o acili a e con ol-s a egy de elopmen and pe o mance e alua ion a
a plan -wide le el and, consequen ly, i includes bo h p e- ea men o was ewa e and he
p ocesses desc ibing sludge ea men .
Finally, he wo k by Li e al. [
21
] did no in ol e WWTPs bu is wo h men ioning
because i p esen s a combina ion o ODEs wi h ML. Thei pape p esen s a Fou ie neu al
ope a o o modeling u bulen lows wi h ze o-sho supe - esolu ion. This wo k showed
highe speed and be e accu acy compa ed wi h classical sol e s.
2.3. Da a-Based Was ewa e Managemen
In WWTPs, VA acili a es apid and in e ac i e explo a ion o mul iple iews o he
same high-dimensional da a. I is possible o ha e a global iew o da a beha io h ough
di e en colo s, o ien a ions, and da a. In e ac i e isualiza ion o ade-o s in mul iple
dimensions is well-sui ed o si ua ions whe e s akeholde s ha e di e se in e es s [22].
Kim e al. [
23
] p oposed an ope a o decision suppo sys em (ODSS) o suppo
WWTP ope a o s in making app op ia e decisions. Kim e al.’s sys em accoun s o wa e -
quali y a ia ions in he WWTP and comp ises wo diagnosis modules, h ee p edic ion
Sus ainabili y 2024,16, 3578 6 o 17
modules, and a scena io-based suppo ing module. The p edic ion modules a e based on
he k-nea es neighbo s (k-NN) me hod o o ecas wa e quali y h ee days in ad ance.
Simila ly, Heo e al. [
24
] p oposed a hyb id in luen o ecas ing model based on mul imodal
and ensemble-based deep lea ning. This ool p edic s a WWTP’s long- e m (daily) and
sho - e m (hou ly) in luen load.
Ja a e al. [
25
] explo ed he e icacy o a i icial neu al ne wo ks (ANNs) and ML
models, including eed- o wa d neu al ne wo k (FFNN), andom o es (RF), con olu ional
neu al ne wo k (CNN), ecu en neu al ne wo k (RNN), and p e- ained s acked au o-
encode (SAE), o p edic ing WWTP pe o mance. By analyzing da a on pollu ion a iables
o e h ee yea s, he s udy e eals ha simple neu al ne wo ks and RF can accu a ely model
WWTP p ocesses o WWTP managemen , demons a ing high co ela ion coe icien s in
p edic ions o e luen quali y, despi e he limi a ions o deep neu al ne wo ks (DNNs)
due o small da a se sizes. Shao e al. [
26
] explo ed nine machine lea ning algo i hms o
p edic sludge p oduc ion, wi h ex eme g adien boos ing ee (XGBoos ) and andom
o es models showing he highes accu acy. These models iden i ied eal-wo ld in luen
olume, wa e empe a u e, and was ewa e quali y as signi ican ac o s a ec ing sludge
p oduc ion in was ewa e ea men plan s.
Piao e al. [
27
] applied ma hema ical modeling in hei esea ch o de ise six s a egic
imp o emen plans o minimize elec ic powe consump ion in was ewa e ea men
plan s. Thei app oach, which in ica ely u ilized a i icial neu al ne wo ks, no only
es ima ed he elec ic powe sa ings om he p oposed plans bu also unde sco ed he
signi ican po en ial o enhancing sus ainabili y and educing en i onmen al impac s.
By op imizing powe usage, he s udy con ibu es aluable insigh s in o achie ing mo e
eco- iendly ope a ions, demons a ing a pi o al s ep owa ds mi iga ing he ecological
oo p in o was ewa e ea men p ocesses.
3. Me hodology
The me hodology ollowed in his a icle is inspi ed by he p oposal o A Ruskin
e al. [28]
.
This me hodology ollows explo a o y da a analysis (EDA) [29] s eps, as explained below:
1.
Da a collec ion and acquisi ion. I is he p ocess o ga he ing and measu ing in o ma-
ion on a ge ed a iables; i is di ided in o he ollowing ac i i ies:
(a)
Analysis o da a o igin and equency.
(b)
Quan i ica ion o da a unce ain y.
(c)
Compila ion o da a om a ious sou ces.
2.
Da a managemen and da a alida ion. I checks sou ce da a’s accu acy and quali y
be o e using, impo ing, o o he wise p ocessing hem. I is composed o he ollowing
ac i i ies:
(a)
Iden i ica ion o he da a dis ibu ion.
(b)
De ec ion o missing alues.
(c)
De ini ion o e oneous da a.
(d)
De ec ion and emo al o ou lie s based on he a iable analysis.
(e)
De ec ion o ou lie s based on physical p ocesses.
3.
Da a isualiza ion. I is he g aphical ep esen a ion o in o ma ion and da a; i s main
ac i i ies a e:
(a)
Explo a ion and isualiza ion o da a.
(b)
De elopmen o in ui i e, powe ul isualiza ions.
(c)
De elopmen o algo i hms o he p edic ion o u u e condi ions.
A Ruskin e al. [
28
] s a e ha “due o he physical na u e o was ewa e p ocess da a, i
is ecommended ha labo a o y, ope a ions, and enginee ing s a be consul ed a all poin s
in he p ocess o con i m assump ions”. Acco ding o Ande be g [
30
], clus e analysis can
be used o de elop induc i e gene aliza ions. Clus e ing analysis has been used in he
domain o wa e quali y o (i) in es iga e he spa io empo al s uc u e o de e minan s in a
Sus ainabili y 2024,16, 3578 7 o 17
se o 21 Sco ish lakes [
31
], (ii) e alua e he wa e quali y o h ee di e en c oss-sec ions
o he Fen Ri e [32], and (iii) e alua e he quali y o unde g ound wa e [33].
Rada plo s a e a use ul way o p esen mul i a ia e da a. Acco ding o Saa y [
34
],
“ ada plo s ha e g ea u ili y in si ua ions in which he e a e la ge numbe s o independen
a iables, possibly wi h di e en measu emen scales”. In addi ion, Joan Saa y ound ha
“ ada plo s ha e a pa icula ele ance o esea che s who wish o illus a e he deg ee o
mul iple g oup simila i y/consensus o g oup di e ences on mul iple a iables in a single
g aphical display” [34].
4. P oposed EDAR 4.0 Tool
EDAR 4.0’s a chi ec u e has he WWTP p ocess as he base, which includes ac o y-
le el da a acquisi ion o all he p ocesses ha make up a WWTP. This p ocess can be
classi ied in o h ee main s anda d subp ocesses. Fi s , he in luen ep esen s he en y
o he incoming wa e and i s p elimina y and p ima y ea men , usually pe o med in
a p ima y se ling o sedimen a ion ank. Second, he biological ea men p ocess is he
cen al pa o he so-called seconda y ea men . I ep esen s he biological was ewa e
ea men p ocess o di e en ypes o bac e ia and p o ozoa, which can be complemen ed
by addi ional chemical ea men s. Thi d, he e luen p ocess ep esen s he was ewa e
ea men plan ou pu . This ou pu ecei es di ec ly ea ed wa e o wa e ha goes
h ough a seconda y decan a ion o sedimen a ion ank, which can also be conside ed pa
o he plan ’s seconda y ea men .
The p ocesses and subp ocesses o a WWTP a e gene ally con olled by one o mo e
p og ammable logic con olle s (PLCs) in eg a ed wi h di e en senso s and ac ua o s. All
con ol in o ma ion is displayed locally ia human–machine in e aces (HMIs), usually in e-
g a ed in o a SCADA (supe iso y con ol and da a acquisi ion) sys em. All he in o ma ion
on he sys em is gene ally sha ed on a local ne wo k (LAN) based on an indus ial p o ocol.
In EDAR 4.0, his is ex ended o a Fou h Indus ial Re olu ion (4IR) sys em a chi ec u e
by es ablishing an addi ional cloud-based In e ne o Things (IoT) in as uc u e ha can
be eached ia he In e ne , so he o e all WWTP and i s in o ma ion and communica ion
echnology (ICT) in as uc u e mus ha e secu e access. In his cloud, a ious se ices
a e in eg a ed, such as WWTP moni o ing, cloud-based IoT da a acquisi ion and s o age,
in o ma ion isualiza ion, da a analysis, and ela ed se ices, such as isual analysis and
scena io analysis o plan ope a ion op imiza ion h ough machine lea ning models.
An example o accessing he abo e IoT cloud in as uc u e and ela ed se ices is
ia he HTTP REST p o ocol. An example o a da a analy ics se ice is o classi y di e en
ypes o wa e quali y and p edic ( o ecas ) how wa e quali y will change o e ime.
Finally, wi h he abo e IoT cloud pla o m unning, he da a om he sewage ea men
plan can be displayed on a webpage whe e emo e use s can execu e wa e quali y analysis
and o he plan moni o ing unc ionali ies. Figu e 2de ails a iew o he EDAR 4.0, 4IR
sys em a chi ec u e [
35
]. This igu e also explains he so wa e ools used o he IoT cloud
componen s. The Py hon-based Flask lib a y’s API was used in his wo k. A Pos g eSQL
9.6 da abase was used o da a s o age. RapidMine 9.3 was used o da a analy ics and
ML-based model cons uc ion. Finally, he Bokeh 1.4 lib a y was used o he isualiza ion
pa . The ollowing subsec ions de ail each o he ML modules de eloped.
Sus ainabili y 2024,16, 3578 8 o 17
Figu e 2. EDAR a chi ec u e [35].
4.1. Wa e -Quali y Moni o ing
The da a se ob ained om he “La Ca uja” WWTP SCADA sys em was subjec ed o
a se ies o s eps o p ep ocess i and lea e i eady o he da a cleaning p ocess. Once he
da a ha e been cleaned, a p incipal componen analysis (PCA) is applied o ex ac he wo
main componen s ha de ine he da a se . Fu he mo e, a clus e ing p ocess is execu ed
using he K-means algo i hm wi h k = 4, whe e each g oup he algo i hm iden i ies belongs
o a wa e quali y clus e .
The pla o m allows he use o adjus i he wa e quali y moni o ing is displayed
on he on end acco ding o he wa e ea men ’s con aminan s emo al pe o mance o
e luen ’s absolu e wa e quali y alues. The WWTP ope a ion pe iod is ano he pa ame e
he use can se om he pla o m. The abo e was implemen ed because he “La Ca uja”
was ewa e ea men plan had a plan design and equipmen imp o emen o e ime,
so i was essen ial o moni o and sepa a e hese wo pe iods. Wa e quali y p o iles (o
clus e s) a e plo ed using a line p o ile cha and a spide cha . Figu e 3displays he
moni o ing module o he EDAR 4.0 pla o m. In Figu e 3b–d, i can be seen ha he blue
clus e (Clus e 0) has he wo s wa e quali y, whe eas he ed one is he bes (Clus e 3). In
addi ion, i can be no ed ha he WWTP should imp o e he ea men o he NTK chemical
a iable. Addi ionally, Figu e 3a displays he menu ba o he pla o m wi h wo selec o s:
one o he ime pe iod selec ion o he WWTP da a and he o he one o changing he
isualiza ion mode (pe o mance o absolu e alues). Figu e 3b,c shows he wa e quali y
p o ile in clus e s o bo h pe o mance and absolu e alues. The y-axis co esponds o
a no malized (scaled) alue o he a iable’s wa e quali y equi emen s om Eu opean
Di ec i e 91/271/EEC (Table 1), whe e he ed do ed ho izon al line deno es he limi
se by he same Eu opean s anda d. Fo example, i he a iable
BOD5
in absolu e alues
is calcula ed by he clus e ing a 25
mgO2/L
, i will be ep esen ed wi h a alue on he
Sus ainabili y 2024,16, 3578 9 o 17
y-axis o 1. I i is wice his alue (50
mgO2/L
), i will be ep esen ed wi h a alue on he
y-axis o 2, showing ha his a iable does no comply wi h he wa e quali y s anda d.
As ano he example o he case o pe o mance alues, i he a iable
BOD5
is calcula ed
by clus e ing a a pe o mance o 70%, i will be ep esen ed on he y-axis wi h a alue
o 1. Howe e , i i is wice his alue (140%), i will be ep esen ed wi h a alue on he
y-axis o 2, showing ha he a iable complies, as i has a pe o mance wice highe han
wha is equi ed by he wa e quali y s anda d. Figu e 3d,e shows he same wa e quali y
p o ile (clus e s) bu using a spide plo ep esen a ion o bo h pe o mance and absolu e
alues. Figu e 3 ,g displays he a iable impo ance plo o pe o mance and absolu e
alues. Fo ins ance, he a iables “e luen TSS pe ” and “e luen TSS conc” we e he
mos signi ican a iables o compu e he wa e quali y p o iling.
(a)
(b) (c)
(d) (e)
( ) (g)
Figu e 3. Visual analy ics wa e quali y moni o ing pla o m. (a) Moni o ing con igu a ion pa am-
e e s. (b) Wa e quali y line cha (pe o mance). (c) Wa e quali y line cha (absolu e). (d) Wa e
quali y spide cha (pe o mance). (e) Wa e quali y spide cha (absolu e). ( ) Wa e quali y a iable
impo ance (pe o mance). (g) Wa e quali y a iable impo ance (absolu e).
Sus ainabili y 2024,16, 3578 16 o 17
Re e ences
1.
Keim, D.; And ienko, G.; Feke e, J.D.; Gö g, C.; Kohlhamme , J.; Melançon, G. Visual Analy ics: De ini ion, P ocess, and
Challenges. In In o ma ion Visualiza ion: Human-Cen e ed Issues and Pe spec i es; Sp inge : Be lin/Heidelbe g, Ge many, 2008;
pp. 154–175.
2.
Diez-Oli an, A.; Del Se , J.; Gala , D.; Sie a, B. Da a usion and machine lea ning o indus ial p ognosis: T ends and pe spec i es
owa ds Indus y 4.0. In . Fusion 2019,50, 92–111. [C ossRe ]
3.
Newha , K.B.; Holloway, R.W.; He ing, A.S.; Ca h, T.Y. Da a-d i en pe o mance analyses o was ewa e ea men plan s: A
e iew. Wa e Res. 2019,157, 498–513. [C ossRe ] [PubMed]
4.
Luo, L.; Duan, N.; Wang, X.C.; Guo, W.; Ngo, H.H. New he modynamic en opy calcula ion based app oach owa ds quan i ying
he impac o eu ophica ion on wa e en i onmen . Sci. To al En i on. 2017,603, 86–93. [C ossRe ] [PubMed]
5.
Maiza, M.; Od iozola, J.; Gil, A.; Na e an, G.; Basagoi i, R.; Lecuona, I.; Zu u uza, U.; U chegi, G.; Mañas, A. Visual Analy ics o
suppo ing he Managemen o WWTPs. In P oceedings o he Young Wa e P o essionals (YWP) Con e ence, 2017, Bilbao, Spain,
16–18 No embe 2017.
6.
Eu opean Di ec i e 91/271/EEC; Council Di ec i e 91/271/EEC o 21 May 1991 Conce ning U ban Was e-Wa e T ea men .
Eu opean Union Law: B ussels, Belgium, 1991.
7.
Cook, K.A.; Thomas, J.J. Illumina ing he Pa h: The Resea ch and De elopmen Agenda o Visual Analy ics; Technical Repo ; Paci ic
No hwes Na ional Lab. (PNNL): Richland, WA, USA, 2005.
8.
an Wijk, J. The alue o isualiza ion. In P oceedings o he VIS 05. IEEE Visualiza ion, 2005, Minneapolis, MN, USA,
23–28 Oc obe 2005; pp. 79–86.
9.
Li, J.K.; Ma, K.L. P6: A decla a i e language o in eg a ing machine lea ning in isual analy ics. IEEE T ans. Vis. Compu . G aph.
2020,27, 380–389. [C ossRe ] [PubMed]
10.
Kalinin, A.A.; Palanimalai, S.; Zhu, J.; Wu, W.; De aj, N.; Ye, C.; Pona ul, N.; Husain, S.S.; Dino , I.D. SOCRAT: A Dynamic Web
Toolbox o In e ac i e Da a P ocessing, Analysis and Visualiza ion. In o ma ion 2022,13, 547. [C ossRe ] [PubMed]
11.
Nawaz, A.; A o a, A.S.; Ali, W.; Saxena, N.; Khan, M.S.; Yun, C.M.; Lee, M. In elligen Human–Machine In e ace: An Agile
Ope a ion and Decision Suppo o an ANAMMOX SBR Sys em a a Pilo -Scale Was ewa e T ea men Plan . IEEE T ans. Ind.
In o m. 2022,18, 6224–6232. [C ossRe ]
12.
Ende , A.; Riba sky, W.; Tu kay, C.; Wong, B.W.; Nabney, I.; Blanco, I.D.; Rossi, F. The s a e o he a in in eg a ing machine
lea ning in o isual analy ics. In P oceedings o he Compu e G aphics Fo um; Wiley Online Lib a y: Hoboken, NJ, USA, 2017;
Volume 36, pp. 458–486.
13.
Keim, D.A.; Munzne , T.; Rossi, F.; Ve leysen, M. B idging in o ma ion isualiza ion wi h machine lea ning (Dags uhl Semina
15101). In Dags uhl Repo s; Schloss Dags uhl-Leibniz-Zen um ue In o ma ik: Wade n, Ge many, 2015; Volume 5.
14.
Liu, S.; Wang, X.; Liu, M.; Zhu, J. Towa ds be e analysis o machine lea ning models: A isual analy ics pe spec i e. Vis. In o m.
2017,1, 48–56. [C ossRe ]
15.
S olpe , C.D.; Pe e , A.; Go z, D. P og essi e isual analy ics: Use -d i en isual explo a ion o in-p og ess analy ics. IEEE T ans.
Vis. Compu . G aph. 2014,20, 1653–1662. [C ossRe ] [PubMed]
16.
Sun, D.; Huang, R.; Chen, Y.; Wang, Y.; Zeng, J.; Yuan, M.; Pong, T.C.; Qu, H. PlanningVis: A isual analy ics app oach o
p oduc ion planning in sma ac o ies. IEEE T ans. Vis. Compu . G aph. 2019,26, 579–589. [C ossRe ] [PubMed]
17.
Wu, W.; Zheng, Y.; Chen, K.; Wang, X.; Cao, N. A isual analy ics app oach o equipmen condi ion moni o ing in sma ac o ies
o p ocess indus y. In P oceedings o he 2018 IEEE Paci ic Visualiza ion Symposium (Paci icVis), Kobe, Japan, 10–13 Ap il 2018;
pp. 140–149.
18.
Jong ack, K.; Kwang ae, Y.; Wenhua, P.; Yejin, K. Modi ied New on-Raphson Me hod o Minimize Calcula ion Time o Was ewa e
T ea men Plan Simula ion. J. Ko ean Soc. Haza d Mi ig. 2018,18, 319–326.
19.
Flo es-Alsina, X.; Kazadi Mbamba, C.; Solon, K.; V ecko, D.; Tai , S.; Ba s one, D.J.; Jeppsson, U.; Ge naey, K.V. A plan -wide
aqueous phase chemis y module desc ibing pH a ia ions and ion specia ion/pai ing in was ewa e ea men p ocess models.
Wa e Res. 2015,85, 255–265. [C ossRe ] [PubMed]
20.
Jeppsson, U.; Rosen, C.; Alex, J.; Copp, J.; Ge naey, K.V.; Pons, M.N.; Van olleghem, P.A. Towa ds a benchma k simula ion model
o plan -wide con ol s a egy pe o mance e alua ion o WWTPs. Wa e Sci. Technol. 2006,53, 287–295. [C ossRe ] [PubMed]
21.
Li, Z.; Ko achki, N.; Azizzadenesheli, K.; Liu, B.; Bha acha ya, K.; S ua , A.; Anandkuma , A. Fou ie Neu al Ope a o o
Pa ame ic Pa ial Di e en ial Equa ions. a Xi 2020, a Xi :2010.08895.
22.
Ma oso , E.S.; Husko a, I.; Kasp zyk, J.R.; Ha ou, J.J.; Lambe , C.; Reed, P.M. Many-objec i e op imiza ion and isual analy ics
e eal key ade-o s o London’s wa e supply. J. Hyd ol. 2015,531, 1040–1053. [C ossRe ]
23.
Kim, M.; Kim, Y.; Kim, H.; Piao, W.; Kim, C. Ope a o decision suppo sys em o in eg a ed was ewa e managemen including
was ewa e ea men plan s and ecei ing wa e bodies. En i on. Sci. Pollu . Res. In . 2016,23, 10785–10798. [C ossRe ] [PubMed]
24.
Heo, S.; Nam, K.; Loy-Beni ez, J.; Yoo, C. Da a-D i en Hyb id Model o Fo ecas ing Was ewa e In luen Loads Based on
Mul imodal and Ensemble Deep Lea ning. IEEE T ans. Ind. In o m. 2021,17, 6925–6934. [C ossRe ]
25.
Ja a , R.; Awad, A.; Ja a , K.; Shah ou , I. P edic ing E luen Quali y in Full-Scale Was ewa e T ea men Plan s Using Shallow
and Deep A i icial Neu al Ne wo ks. Sus ainabili y 2022,14, 15598. [C ossRe ]
26.
Shao, S.; Fu, D.; Yang, T.; Mu, H.; Gao, Q.; Zhang, Y. Analysis o Machine Lea ning Models o Was ewa e T ea men Plan
Sludge Ou pu P edic ion. Sus ainabili y 2023,15, 13380. [C ossRe ]
Sus ainabili y 2024,16, 3578 17 o 17
27.
Piao, W.; Kim, C.; Cho, S.; Kim, H.; Kim, M.; Kim, Y. De elopmen o a p o ocol o op imize elec ic powe consump ion and li e
cycle en i onmen al impac s o ope a ion o was ewa e ea men plan . En i on. Sci. Pollu . Res. In . 2016,23, 25451–25466.
[C ossRe ] [PubMed]
28.
A Ruskin, G.A.; Jacquez, G.M.; Melike , J.R.; Slo nick, M.J.; Kau mann, A.M.; N iagu, J.O. Visualiza ion and explo a o y analysis
o epidemiologic da a using a no el space ime in o ma ion sys em. In . J. Heal h Geog . 2004,3, 26. [C ossRe ] [PubMed]
29.
Ghosh, A.; Nashaa , M.; Mille , J.; Quade , S.; Ma s on, C. A comp ehensi e e iew o ools o explo a o y analysis o abula
indus ial da ase s. Vis. In o m. 2018,2, 235–253. [C ossRe ]
30.
Ande be g, M. Clus e Analysis o Applica ions: P obabili y and Ma hema ical S a is ics: A Se ies o Monog aphs and Tex books;
P obabili y and Ma hema ical S a is ics; Else ie Science: Ams e dam, The Ne he lands, 2014.
31.
Hagga y, R.; Mille , C.; Sco , E.; Wyllie, F.; Smi h, M. Func ional clus e ing o wa e quali y da a in Sco land. En i onme ics
2012,23, 685–695. [C ossRe ]
32. Wong, H.; Hu, B. Applica ion o in e al clus e ing app oach o wa e quali y e alua ion. J. Hyd ol. 2013,491, 1–12. [C ossRe ]
33.
Vo-Van, T.; Nguyen-Hai, A.; Ta -Hong, M.V.; Nguyen-T ang, T. A New Clus e ing Algo i hm and I s Applica ion in Assessing he
Quali y o Unde g ound Wa e . Sci. P og am. 2020,2020, 6458576. [C ossRe ]
34.
Saa y, M.J. Rada plo s: A use ul way o p esen ing mul i a ia e heal h ca e da a. J. Clin. Epidemiol. 2008,61, 311–317. [C ossRe ]
35.
Velasquez, D.; To o, M.; B use, J.L.; O egui, X.; Maiza, M.; Sie a, B. A No el A chi ec u e De ini ion o AI-D i en Indus y 4.0
Applica ions. In P oceedings o he 2023 In e na ional Con e ence on In elligen Compu ing and Con ol (IC&C), Wuhan, China,
24–26 Feb ua y 2023; pp. 25–31.
36.
Fo una, L.; Rizzo o, G.; La o gna, M.; Nunna i, G.; Xibilia, M.G.; Capone o, R. E olu iona y Op imiza ion Algo i hms. In So
Compu ing: New T ends and Applica ions; Sp inge : London, UK, 2001; pp. 97–116.
Disclaime /Publishe ’s No e: The s a emen s, opinions and da a con ained in all publica ions a e solely hose o he indi idual
au ho (s) and con ibu o (s) and no o MDPI and/o he edi o (s). MDPI and/o he edi o (s) disclaim esponsibili y o any inju y o
people o p ope y esul ing om any ideas, me hods, ins uc ions o p oduc s e e ed o in he con en .
