Ci a ion: Mo cillo-Jimenez, R.;
Gu ié ez-Ba is a, K.; Gómez-Rome o,
J. TSx end: A Tool o Ba ch Analysis
o Tempo al Senso Da a. Ene gies
2023,16, 1581. h ps://doi.o g/
10.3390/en16041581
Academic Edi o : Fe nando
Mo gado-Dias
Recei ed: 24 Decembe 2022
Re ised: 23 Janua y 2023
Accep ed: 1 Feb ua y 2023
Published: 4 Feb ua y 2023
Copy igh : © 2023 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/).
ene gies
A icle
TSx end: A Tool o Ba ch Analysis o Tempo al Senso Da a
Robe o Mo cillo-Jimenez , Ka el Gu ié ez-Ba is a and Juan Gómez-Rome o *
Depa men o Compu e Science and A i icial In elligence, Uni e si y o G anada, 18071 G anada, Spain
*Co espondence: [email p o ec ed].es
Abs ac :
P e-p ocessing and analysis o senso da a p esen se e al challenges due o hei inc eas-
ingly complex s uc u e and lack o consis ency. In his pape , we p esen TSx end, a so wa e ool
ha allows non-p og amme s o ans o m, clean, and analyze empo al senso da a by de ining and
execu ing p ocess wo k lows in a decla a i e language. TSx end in eg a es se e al exis ing echniques
o empo al da a pa i ioning, cleaning, and impu a ion, along wi h s a e-o - he-a machine lea ning
algo i hms o p edic ion and ools o expe imen de ini ion and acking. Mo eo e , he modula
a chi ec u e o he ool acili a es he inco po a ion o addi ional me hods. The examples p esen ed
in his pape using he ASHRAE G ea Ene gy P edic o da ase show ha TSx end is pa icula ly
e ec i e o analyze ene gy da a.
Keywo ds: ime se ies; p e-p ocessing; p edic ion; machine lea ning; deep lea ning
1. In oduc ion
The de elopmen and g ow h o in o ma ion and communica ion echnologies ha e
p opi ia ed he daily gene a ion o massi e da a. Today, much o he gene a ed da a come
om senso da a. Tempo al senso da a ha e become o g ea in e es o he academic and
p i a e sec o s, as s udying his so o da a allows o s udies o he e olu ion o da a o e
ime, p o iding end-use s wi h obus algo i hms and ools o decision-making.
The e a e se e al applica ions ha use senso da a o a ious pu poses [
1
,
2
], such as
handling and managing measu es such as empe a u e, humidi y, p essu e, gas, op ical, and
many o he s. Many companies in di e en indus ies a e becoming inc easingly awa e o
he g ea po en ial ha he esea ch and exploi a ion o empo al senso da a can o e [
3
,
4
].
The e a e many challenges in dealing wi h empo al senso da a. In he ollowing, we
men ion he main challenges conce ning his so o da a:
1.
One o he main challenges is dealing wi h he la ge olumes o da a ha a e gene a ed
by senso s each day. This can make s o ing, managing, and p ocessing he da a
di icul , equi ing specialized ools and echniques.
2.
Ano he challenge is he a iabili y and noise in he da a, which makes i di icul o
iden i y ends and pa e ns. This may equi e ad anced da a il e ing and cleaning
me hods o emo e i ele an o inaccu a e in o ma ion.
3.
Addi ionally, he lack o consis ency and complex s uc u e inhe en in empo al sen-
so da a make i s p ocessing and analysis e en mo e di icul . Sophis ica ed algo i hms
and echniques may be equi ed o analyze and in e p e he da a app op ia ely.
4.
Fu he mo e, empo al senso da a a e o en ha es ed om he e ogeneous sou ces.
This can be challenging, equi ing e icien in eg a ion app oaches o handle he da a
in a imely manne .
All he a o emen ioned challenges hinde he de inini ion o a wo k low ha allows
o p omising esul s o be ob ained. Fu he mo e, selec ing he mos sui able algo i hm o
sol e he p oblem cons i u es ano he bu densome ask. Sol ing his p oblem is ha d wo k
due o he na u e and a iabili y o each ela ed p oblem.
Ene gies 2023,16, 1581. h ps://doi.o g/10.3390/en16041581 h ps://www.mdpi.com/jou nal/ene gies
Ene gies 2023,16, 1581 2 o 29
One o he mos signi ican challenges in p edic ion p oblems o ime se ies and
any o he da a ype is he ime needed o design he co ec s a egy o he expe imen .
Di e en s udies ocus on he applica ion o speci ic algo i hms o a da a se ies using a s a ic
con igu a ion [
5
–
7
]; in o he wo ds, wi hou o e ing he possibili y o pa ame e ising he
di e en expe imen s. This consumes a g ea deal o ime o each expe imen de ini ion,
hus slowing down he esea che ’s main objec i e.
As s a ed be o e, ime se ies da a ha e gained he a en ion o he en i e esea ch
communi y, as i s analysis allows o an unde s anding o changes in he da a o e ime.
By analyzing he ends and pa e ns in he da a, insigh s in o how he da a a e e ol ing
can be acqui ed, and p edic ions abou u u e esul s can be made. This is pa icula ly
use ul in ields whe e changes o e ime can signi ican ly impac decision-making and
s a egy. Addi ionally, empo al senso da a analysis can help o iden i y po en ial issues o
anomalies, allowing o imely in e en ions and co ec i e ac ions.
This pape p oposes a ool called TSx end o ime se ies analysis. TSx end p esen s a
modula a chi ec u e and s anda dises he di e en s ages o expe imen a ion, c ea ing a
wo k low h ough a simple con igu a ion ile. The p oposed ool allows o he end-use s o
wo k on ime se ies da a wi hou p og amming knowledge using he a ailable echniques
and ocusing on de eloping he esea ch. The ool enables da a il e ing and cleaning o
emo e inco ec o nonessen ial in o ma ion. Finally, TSx end has he abili y o execu e
deep p edic ion and machine lea ning algo i hms ha a e commonly u ilized in a ious
domains, such as ene gy [
8
], medicine [
9
], o geoscience [
10
,
11
]. This allows o a wide
ange o applica ions and lexibili y in i s usage. The esul s a e p esen ed h ough he
isualiza ion module, enabling end use s o analyze he esul s a di e en wo k low s ages.
A summa y o he main con ibu ions o his pape is as llows:
•
The pape ’s main con ibu ion is he de elopmen o a ool called TSx end o ime
se ies analysis, which has a modula a chi ec u e and s anda dizes di e en s ages
o expe imen a ion.
•
The ool allows o end-use s o wo k on ime se ies da a wi hou p og amming
knowledge, simpli ying he p ocess o da a il e ing and cleaning o emo e inco ec
o nonessen ial in o ma ion.
•
TSx end o e s he possibili y o execu ing p edic ion algo i hms and isualizing he
esul s h ough a isualisa ion module, enabling end-use s o analyse he esul s a
di e en wo k low s ages.
I is impo an o no e ha , o he bes o he au ho s’ knowledge, he e is no exis ing
ool in he li e a u e ha has he same abili ies as TSx end. This ool is unique in i s abili y o
s anda dize he expe imen a ion p ocess, simpli y ime se ies analysis o end-use s wi hou
p og amming knowledge, and p o ide a isualisa ion module o a be e analysis o esul s.
In his pape , we apply he p oposed ool in an ene gy consump ion p oblem. We ob ained
he da a om he p edic ion ea u es con es on he Kaggle pla o m called ASHRAE—G ea
Ene gy P edic o III [
12
]. The da abase comp ised h ee yea s o hou ly me e eadings om
mo e han a housand buildings in di e en loca ions wo ldwide. I should be no ed ha he
da abase belongs o ASHRAE, a la ge building echnology associa ion [13].
The es o he pape is s uc u ed as ollows: Sec ion 2 e iews p e ious wo k on
his opic. Sec ion 3p esen s he design and unc ionali ies de ails o he p esen ed ool
(Tsx end). Sec ion 4ex ends he desc ip ion o he di e en modules ha comp ise he ool.
In o de o showcase he easibili y o he p oposed ool, Sec ion 5p esen s a eal-wo ld use-
case using TSx end and discusses he ob ained esul s. Finally, in Sec ion 6, he conclusions
and u u e esea ch a e p esen ed.
2. Rela ed Wo ks
Many applica ions aim o make he wo k o esea che s wo king wi h ime se ies da a
easie [14–18]. This is d i en by he need o abs ac p og amming knowledge o a highe
le el, allowing o esea che s who a e no expe s in da a science o ocus on s udying
he da a a he han he echnical de ails o he algo i hm. Some ools ha can be used
Ene gies 2023,16, 1581 3 o 29
o achie e his abs ac ion, isola ing he esea che om he echnical complexi ies, a e
desc ibed in he ollowing.
To make he li e a u e e iew easie o unde s and and be e highligh he no el y
o his esea ch, a compa a i e s udy o lib a ies was conduc ed. Va ious aspec s o he
lib a ies we e analysed, such as hei abili y o collec and use he e ogeneous da a sou ces,
he capaci y o he p ocessing ools a ailable wi hin he lib a y, and he REST API abili ies o
he lib a ies. Fu he mo e, we also e alua ed whe he he lib a ies can implemen A i icial
In elligence (AI) algo i hms, i hey ha e been used in eal-wo ld use-cases and i hey
a e use - iendly. The ease o use is pa icula ly impo an when he lib a ies a e used by
non-expe da a-mining use s. Th ough his analysis, we aim o p o ide a comp ehensi e
compa ison o he di e en lib a ies and hei ea u es, o aid in he unde s anding o he
signi icance o his esea ch.
One example o such a ool is Enlopy [
19
], which is an open-sou ce ool de eloped
in Py hon ha o e s a a ie y o me hods o p ocessing, analyzing, and plo ing ime
se ies da a. This ool has modules ha a e p ima ily ocused on s udying ime se ies da a,
wi h capabili ies such as analysis echniques, g aphing, da a augmen a ion, and ea u e
ex ac ion, mainly in he ene gy domain. Howe e , i should be no ed ha his ool equi es
a signi ican amoun o p og amming knowledge, making i inaccessible o esea che s
who a e no expe s in da a science.
The TSSA ool [
14
] p ima ily ocuses on he p ep ocessing s age, o ob ain he co ela-
ion be ween he esis ance a iables o memo ies in di e en de ices. This is no sui able
o wo k wi h he e ogeneous da a and equi es ex ensi e knowledge in da a science. Addi-
ionally, i does no ha e a REST API o da a e ie al, which limi s i s unc ionali y and
accessibili y o use s.
Ano he ool ha add esses he wo k wi h ime se ies is s esh [
20
]. This ool enables he
s udy o ime se ies da a by ex ac ing ea u es and aining simple classi ica ion o eg ession
models. Howe e , i should be no ed ha his ool equi es a signi ican amoun o p og am-
ming knowledge o be used o i s ull ex en . Addi ionally, i does no o e a wo k low ea u e
o guide use s h ough hei wo k p ocess. In [
18
], a Visual Wa ning Tool o Financial Time
Se ies (VWSTFTS) is p esen ed based on scaling analysis. The p oposed me hod uses he
ime-dependen Gene alised Hu s Exponen (GHE) me hod o analyse inancial ime se ies
and iden i y empo al pa e ns in GHE p o iles. By applying his me hodology and using a
isual ool, he esea che s can analyse signi ican and pe iphe al s ock ma ke indices. The
p oposed me hod o e s a new way o iden i ying pa e ns in inancial ime se ies da a and
can be used o p o ide ea ly wa ning signals o ma ke luc ua ions. Howe e , he ool is
limi ed in i s capabili ies and does no suppo AI echniques. Addi ionally, i is p ima ily
used in he inancial domain and migh no be sui able o o he ields.
Da s [
21
] is one o he mos comp ehensi e ools a ailable, p o iding a wide ange o
algo i hms o wo king wi h ime se ies. Da s suppo s bo h uni a ia e and mul i a ia e
ime se ies and models. Acycle [
17
] is a specialized so wa e o paleoclima e esea ch
and educa ion ha ocuses on signal p ocessing, pa icula ly o cyclos a ig aphy and
as och onology. I includes a ious models, such as sedimen a ion noise and sedimen a ion
a e, which a e speci ic o sedimen a y esea ch. Acycle’s ully implemen ed g aphical
use in e ace makes i easy o use s o ope a e and na iga e he so wa e, making i use -
iendly o bo h esea che s and educa o s. I should be no ed ha nei he Da s no Acycle
includes a p ep ocessing phase, meaning ha he da a need o be cleaned, ans o med and
p epa ed be o e being used wi h hese ools. This could mean ha addi ional s eps and
esou ces a e equi ed o he e ec i e use o hese ools.
Ano he s a e-o - he-a ool o ime se ies analysis is Ka s [
22
]. This ool aims o
make ime se ies analysis mo e accessible o esea che s wi h a solid backg ound in da a
science. Ka s o e s a a ie y o o ecas ing algo i hms, such as indi idual o ecas ing
models, ensemble models, a sel -supe ised lea ning model (me a-lea ning), back es ing,
hype pa ame e i ing and empi ical p edic ion in e als. This ool is a comp ehensi e
and powe ul op ion o esea che s looking o pe o m ad anced ime se ies analysis.
Ene gies 2023,16, 1581 4 o 29
TSFEL [
23
] is a ool ha p ima ily ocuses on he p ep ocessing s age o ime se ies
analysis, p o iding a ange o op ions o explo a o y analysis and ea u e ex ac ion. One
o i s no able ea u es is he inclusion o a se o uni es s o e i y he uns. Howe e ,
like many o he ools, TSFEL can be challenging o use o esea che s wi hou s ong
p og amming skills.
SSTS [
16
] is a ool o sea ching pa e ns in ime se ies da a. I u ilises a syn ac ic
app oach, analysing he s uc u e and ela ionships wi hin he da a. SSTS allows o use s
o speci y complex pa e ns using o mal g amma and sea ch o ins ances o a pa e n.
The ool can iden i y pa e ns ha a e di icul o de ec using adi ional me hods such
as s a is ical analysis o machine lea ning. Addi ionally, SSTS can ex ac ea u es om
ime se ies da a o u he analysis o o ain p edic i e models. I is a powe ul ool o
sea ching pa e ns in ime se ies da a ha can be used in a ious ields, such as inance,
heal hca e, and anspo a ion.
cleanTS [
15
] is an au oma ed ool o cleaning uni a ia e ime se ies da a. I u ilises
machine lea ning echniques o emo e noise, ou lie s and missing alues, helping o
imp o e he accu acy and eliabili y o ime se ies analysis. I also makes i easie o iden i y
pa e ns and ends in he da a. cleanTS can also be used o in e pola e missing da a and
esample he da a a di e en ime scales. This ool can be applied in a ious domains, such
as inancial ime se ies, senso da a and o he ime se ies da a. I is a use ul ool o he
p ep ocessing o ime se ies da a, making i mo e eliable and accu a e o u he analysis.
Ano he lib a y ha allows o wo king wi h ime se ies da a is G eyKi e [
24
]. This
ool p o ides p ep ocessing capabili ies as well as p edic ion models o he e ogeneous
da a. One o i s unique ea u es is i s own algo i hm, which can be used o c ea ing
p edic ion models, called Sil e ki e. G eyKi e is also aimed a da a scien is s. Ano he
comple e ool is Au oTS [
25
], which u ilizes o he lib a ies o he gene a ion o p edic ion
models wi h he help o a amewo k. As wi h o he ools, i equi es a good unde s anding
o p og amming. Table 1p o ides a compa ison o he lib a ies men ioned abo e wi h he
ool p oposed in his pape .
Table 1.
Compa ison o he p oposed ool wi h he ools men ioned ea lie o analysing empo-
al senso da a. Legend:
3
—Fea u e suppo ed,
7
—Fea u e no suppo ed,
∼
—Fea u e no ully
suppo ed o wi h explici limi a ions, ?—Unknown; in o ma ion no a ailable abou he ea u e.
Name He e ogeneous
Da a Collec ion
P ep ocessing
Da a
REST
API
AI
Tools
Applica ions
Wide Range o A eas
Use
F iendly
Enlopy 3 3 7 7 7 7
TS esh 7 3 7 3 ?7
Ka s 7 7 7 3 3 ∼
Da s 3 7 7 3 3 7
SSTS ? 3 7 3 3 7
TSSA 7 3 7 7 ?7
cleanTS 7 3 7 3 3 7
G eyKi e 3 3 7 3 3 7
TS el 3 3 7 7 7 ∼
Au oTs 3 7 7 3 3 7
Acycle 3 7 7 3 7 3
VWSTFTS 3 3 7 7 7 ∼
TSx end 3 3 3 3 7 3
The p oposed ool is designed o acili a e da a p ocessing o non-expe use s wi h
a simple, use -o ien ed in e ace. The ool is p ima ily aimed a he p ocessing o ime
se ies da a such as ene gy consump ion, bu can also p ocess and analyse a ious o he
ypes o da a. Addi ionally, he lib a y includes commonly used da a p ocessing, cleaning,
and ans o ma ion ools. I s design allows o he easy in eg a ion o new me hods and
algo i hms. A wide ange o da a analysis algo i hms is also a ailable. One o he key
s eng hs o TSx end is i s use -o ien ed design and communica ion h ough i s REST API,
which makes he ool dynamic and easy o use.
Ene gies 2023,16, 1581 5 o 29
All he p esen ed ools a e comp ehensi e lib a ies o analysing empo al senso da a,
helping esea che s o s udy hese ypes o da a wi hou he need o ex ensi e knowledge
o da a science. Howe e , almos all o hese ools equi e p og amming knowledge, which
can be a hind ance o esea che s who wan o ocus solely on da a analysis. In con as ,
TSx end no only p o ides a ool o wo king wi h empo al senso da a, bu also includes
an a chi ec u e ha allows o esea che s o c ea e a wo k low o hei esea ch by edi ing
a con igu a ion ile. This enables esea che s o conduc expe imen s s ep by s ep, analysing
he da a a each s age, wi hou he need o p og amming knowledge.
3. Design and Func ionali ies o TSx end
TSx end comp ises a se ies o modules o speci ic asks. The modules ha comp ise
his ool a e g ouped in o di e en da a science pa adigms, each o which o e s he
possibili y o execu ing di e en algo i hms. In he ollowing sec ion, he so wa e design
o ou ool will be in oduced, and a b ie explana ion o i s unc ionali ies will be p o ided.
3.1. Design
The main idea behind designing his ool is o c ea e a se ies o dynamic and lexible
modules, allowing o a wide ange o ope a ions on he inpu da a and p o iding a inal
esul . One o he p ima y ad an ages o his design app oach is ha i p o ides a obus
ool, whe e each componen can be execu ed sepa a ely. This allows o each s ep o he
pipeline o be un sepa a ely, wi hou necessa ily ha ing o un he whole p ocess e e y
ime he expe imen is pe o med.
As shown in Figu e 1, TSx end is o ganised in o i e modules: one module o s o e he
logge s o he ool, a con igu a ion module, a module o p ocess da a, a module o execu e
machine lea ning algo i hms and a module o execu e deep lea ning algo i hms. In he
ollowing, we b ie ly in oduce each o hese modules.
Figu e 1. Design o he p oposed ool.
3.2. Func ionali ies
Fi s ly, h ough a plain ex ile, he con igu a ion module allows o he inse ion o
a se ies o pa ame e s o con igu e he expe imen a ion and au oma e he whole p ocess
wi hou p og amming knowledge.
The da a p ocessing module includes a se ies o da a p ep ocessing algo i hms, such as
elimina ing missing alues, elimina ing ou -o - ange da a in ou inpu da a, and a se ies o
algo i hms aiming o ob ain he di e en ela ionships be ween he di e en cha ac e is ics
o ou inpu da a. These algo i hms we e chosen because hey cause one o he mos
common p oblems when wo king wi h empo al se ies da a.
The machine lea ning module includes he s anda d algo i hms used in his pa adigm,
such as XGBoos [
26
], Decision [
27
] and Reg ession T ees [
28
]. These algo i hms we e

Ene gies 2023,16, 1581 6 o 29
chosen because hey a e among he mos widely used in he s a e-o - he-a o sol e
p edic ion p oblems based on s uc u ed ime se ies da a.
In he deep lea ning module, he same selec ion policy o he in eg a ed algo i hms
was ollowed as in he p e ious module. The mos widely used algo i hms o ime se ies o-
cused on he deep lea ning pa adigm we e selec ed, such as Con olu ional Neu al Ne wo k
(CNN) [29], Long-Sho Te m Memo y (LSTM) [30] and Mul ilaye Pe cep on [31].
Finally, he module eco de s ep esen he esul s and can ack hem. Fo his, we
in eg a ed Ml low [
32
] because i is a ool ha is used o keep ack o he expe imen s
pe o med by any use . This ool helps o show he s a us o he expe imen s wi h an acces-
sible, use - iendly in e ace. In he ollowing sec ions, we will explain he unc ionali ies
o each module in mo e de ail.
4. Module Desc ip ions
The p oposed ool p esen s a modula a chi ec u e ha acili a es he e o less in-
co po a ion o u u e unc ionali ies. In he subsequen sec ions, he cha ac e is ics and
unc ionali ies o each module will be desc ibed in de ail.
4.1. Coo dina ion Module
Ou ool is mainly ocused on ca ying ou explo a o y da a analyses in a as and agile
way. In his way, he end-use s gain a be e unde s anding o he da a, and can iden i y
po en ial issues o p oblems wi h he da a and quickly de elop mo e e ined analyses. The
ool o e s a se o p edic ion algo i hms ha a e mainly used o sol e ime se ies p oblems,
such as neu al ne wo ks. Al hough he algo i hms p o ided by he ool can be applied o
any da a ype, he p oposed lib a y specialises in empo al senso da a ega ding ene gy.
Se e al wo ks add ess ene gy- ela ed p oblems using da a science echniques o
iden i y oppo uni ies o ene gy-e iciency imp o emen s [
33
–
35
]. Howe e , only a ew can
au oma e he en i e expe imen a ion p ocess. Fu he mo e, end-use s equi e p og amming
skills o use he lib a y.
The con igu a ion module is one o he mos impo an . I allows o di e en execu-
ions o he de ined expe imen o be plabbed. The objec i e o his module is o con igu e
a oad map so ha use s can schedule he whole expe imen a ion s a egy. The main
ad an age o his app oach is ha i can endow end-use s wi h a obus ool, which can be
handled wi hou p og amming knowledge.
The con olle ile, called MLp ojec , con ains a de ini ion o he algo i hms ha a e o
be used and he con igu a ion pa ame e s. The ile eads he da a in he YAML iles and
execu es he use -de ined wo k low au oma ically.
The en i e oadmap o he expe imen a ion is con igu ed h ough a se ies o YAML
iles [
36
], whose p ima y unc ion is o de ine and se ialize he di e en p ocesses o all
kinds o p og amming languages. Each ile has a se ies o pa ame e s ha he esea che
can edi . The pa ame e s can be consul ed in [
37
], whe e all he di e en op ions accep ed
by he ool ha e been de ailed. The nomencla u e used o his ile is based on dic iona ies
(key: alue pai ).
Each module o he ool has one o hese con igu a ion iles, and he con olle ile
con ols i s execu ion. This allows o he con igua ion o some o he unc ions ha a e
explici ly o e ed by he ool’s sub-modules. I is possible o choose he algo i hms ha will
un du ing he expe imen a ion o each module, as well as he inpu and ou pu da a o he
execu ion o hese algo i hms, and he numbe o ows in he da ase , which is o g ea help
when simpli ying he analysis o la ge da ase s and educe expe imen a ion imes. Once his
module is con igu ed, TSx en can au oma ically execu e each s ep con igu ed in he ile.
4.2. Da a P e-P ocessing Module
In his module, he esea che begins o p epa e he da a ha will be used in he
expe imen a ion. The module is pa amoun o empo al senso da a (especially ene gy da a)
since he da a a e collec ed om he e ogeneous da a sou ces (senso s, da abases, e c.) [
38
,
39
].
Ene gies 2023,16, 1581 7 o 29
Gene ally, he aw da a need o be p e-p ocessed o eed he di e en lea ning algo-
i hms, so he da a equi e a cleaning and p epa a ion p ocess o imp o e he algo i hms’
pe o mance. This module comp ises di e en submodules, which a e desc ibed below.
4.2.1. Pa i ioning
Da a pa i ion echniques ans o m he cu en da ase s o gene a e new ones. These
echniques a e essen ial in ime se ies on ene gy da a because hey gene a e new, hidden
knowledge ha can imp o e he esul s o e ed by lea ning algo i hms.
Nume ous wo ks include his so o echnique [
40
–
42
]. The da a-pa i ioning module
akes ca e o pa i ioning, g ouping, and ex ac ing he necessa y in o ma ion so he use
can expe imen , using a gi en ime in e al. I helps o s udy he da ase de ined in a
speci ic ime in e al, which can be a smalle han he o iginal da ase .
Ano he impo an ea u e is he possibili y o g ouping he da ase ields a di e en
hie a chical le els, gene a ing a new da ase o analysis in he expe imen . Fo example, i
is possible o g oup acco ding o speci ic da ase ield and ob ain se e al subse s o da a o
analyse, gene a ing new, hidden knowledge.
Fo his pu pose, dynamic ee gene a ion was implemen ed o c ea e a new da ase
ollowing he pa en –child–g andchild hie a chy. I should be no ed ha , deepe in o he ee,
he execu ion becomes less e icien . This pe mi s he gene a ion o new, speci ic da ase s
and ocuses he expe imen a ion on da a wi h a mo e well-de ined ea u e se . Finally, he
module o e s he possibili y o de ining whe e he gene a ed da ase is s o ed and whe e he
execu ions o he expe imen s a e pe o med. The s uc u e o his ee can be seen in Figu e 2.
Figu e 2. Example o ee gene a ion.
4.2.2. Missing Values
Dealing wi h incomple e da a is a common challenge when wo king wi h da ase s,
especially ega ding ene gy e iciency [
43
,
44
]. Incomple e da a can a ise o a ious easons,
including aul y senso s, los da a du ing ansmission, o missing da a due o human e o .
This ype o da a can cause mal unc ions in machine lea ning p edic ion algo i hms, so i is
essen ial o apply p e-p ocessing echniques o deal wi h hese e o s and achie e be e
esul s. To add ess his issue, a submodule was c ea ed wi hin he da a p ocessing module
and he ocus was placed on emo ing missing alues om he da ase .
This submodule o e s he op ion o ea ing he missing alues by con igu ing he ile
and he possibili y o indica ing he pa h in which he da ase is loca ed in o de o apply
he missing alue echniques. The cu en ly implemen ed algo i hms a e he in e pola ion
algo i hm, which pe o ms an in e pola ion be ween he di e en ows o he da ase , and
ano he , mo e agg essi e one, which elimina es he ows wi h missing alues.
The in e pola ion algo i hm [
45
] allows o he gene a ion o new da a and enhances he
quali y o he o iginal da ase . I in ol es es ima ing he alue o a ield a in e media e
Ene gies 2023,16, 1581 8 o 29
poin s be ween known da apoin s. The ow elimina ion algo i hm is mo e agg essi e and,
unlike he p e ious one, elimina es he ow. This algo i hm is necessa y when he numbe
o missing alues wi hin a ow o he da ase is oo high.
Once ei he algo i hm is execu ed, he sys em gene a es a g aphical esou ce o e i y
ha he missing alues ha e been co ec ly emo ed, showing he numbe o eco ds in
each ield. The esou ces a e s o ed in he log module so ha he esul s a e sa ed and can
be consul ed a any ime. I should be highligh ed ha his p ocess ans o ms he o iginal
da ase , hus imp o ing he quali y o he da ase and, he e o e, he pe o mance o he
lea ning algo i hms.
4.2.3. Ou lie s
Ou lie s a e ano he common p oblem o be aced when sol ing da a science p ob-
lems [
46
–
48
]. These ypes o alues a e all oo equen in ene gy p oblems because he
de ices ha a e esponsible o collec ing he in o ma ion and sending i o ou sou ce o
knowledge a e p one o manipula ion by ex e nal sou ces om he en i onmen in which
hey a e ins alled.
The e a e cases in which he senso s a e exposed o empe a u es ha do no co e-
spond o he ac ual empe a u e o he en i onmen ha exis s a ha momen , ei he due
o exposu e o a o eign hea sou ce o o a cold sou ce. These ac ions lead o a se ies o
ou lie s ha a e ou side o he ypical pa e n o measu emen s.
To sol e his ype o p oblem, his p ep ocessing submodule was c ea ed, which
ocuses on he elimina ion o hese ou lie s om he da ase wi h which he esea che is
going o wo k. This includes he possibili y o selec ing he ields o be analyzed in he
con igu a ion ile, as well as indica ing he pa h on which he da ase o which we a e going
o apply he ou lie p ep ocessing echniques is loca ed.
In his i s laye o ou a chi ec u e, w included he “z sco e mean” algo i hm. This
algo i hm de ec s he alues ha a e wi hin a ange de ined by us as missing alues,
and pe o ms an a e age o e i s nea es neighbou s o modi y ha alue and ob ain a
p edic ion o wha alue should exis a ha momen . The sys em gene a es a g aphical
esou ce o check ha he ou lie s ha e been co ec ly elimina ed by means o a simple box
plo . The esou ces a e s o ed in he logge s module, so he esul s a e s o ed and can be
que ied la e in a mo e e ec i e way.
This p ocess also ans o ms ou da ase , i.e., he applica ion o his algo i hm modi ies he
da ase wi h which he esea ch con inues o mo e e ec i ely apply ou
lea ning algo i hms.
4.2.4. Fea u e Selec ion
Fea u e selec ion is a p ep ocessing echnique ha allows o us o ob ain aluable
in o ma ion o lea n wha ype o ea u es a e mos ele an and how he di e en a iables
ha make up ou da ase ela e o each o he and he p oblem we a e analysing. Like he
p e ious ones, i is one o he mos widely used echniques in da a p ocessing.
In p oblems such as hose posed by ene gy e iciency in buildings [
49
,
50
], his is
ad an ageous because, in mos cases, he e a e a la ge numbe o a iables, o which,
depending on he p oblem we a e ying o sol e, we will need a se ies o a iables o o he s.
The c ea ion o his module manages o show he ela ionship be ween he di e en
a iables o he da ase on which we a e wo king. Fo his o wo k co ec ly, i is ad isable
he e a e no missing alues.
In ou ool, i is possible o choose he ields ha a e o be analysed, as well as he
ou e om which he da a o be analysed a e ob ained. One o he algo i hms ha is
implemen ed is a co ela ion algo i hm, which gene a es an image- ype esou ce, wi h a
hea map displaying he co ela ion be ween he di e en selec ed ields. This algo i hm
shows he deg ee, be ween 0 and 1, o co espondence be ween one a iable and ano he ,
wi h 0 being he lowes deg ee and 1 he highes deg ee o co espondence. Ano he o
he implemen ed algo i hms is a p op ie a y algo i hm called FSMeasu e ha manages o
Ene gies 2023,16, 1581 9 o 29
ob ain he measu es o mean, s anda d de ia ion, en opy, chi-squa e and dispe sion o he
da a o he di e en a iables o he da ase wi h which we a e wo king.
This ype o analysis helps us in he selec ion o he inpu ields o ou ime se ies,
which we a e going o inse in o ou lea ning algo i hms. The esou ces gene a ed by his
ype o algo i hm a e s o ed in he logge s module so ha hey can be consul ed la e in a
con enien way.
4.3. Machine Lea ning Module
The machine lea ning pa adigm has been widely applied o sol e ene gy e iciency
p oblems [
51
–
54
]. This ype o pa adigm allows o he inpu o a se ies o da a and he
execu ion o p ocesses ha esul in an ou pu . Depending on he p oblem being sol ed, his
ou pu can p edic a esul wi h a ange o success o classi y da a in o a speci ic ca ego y.
In he case o ene gy p oblems, mos o hese p oblems a e eg ession p edic ion
p oblems. In his lib a y, which is based on ime se ies analysis, he inpu da ase ob ained
om p ep ocessing and explo a o y analysis is used as h e inpu . A se ies o algo i hms
a e un, which p oduce p edic ions abou he building’s consump ion as ou pu . This helps
o op imize consump ion and plan a se ies o ene gy-sa ing s a egies o he end use .
This machine lea ning module con ains se e al sub-modules, which include algo-
i hms based on sol ing ime se ies p oblems o make p edic ions using ou da ase s. As a
ime-se ies- ocused ool, hese algo i hms can all sol e eg ession p oblems.
These algo i hms gene a e a se ies o esou ces ha help he esea che o d aw
conclusions abou he model gene a ed in ou expe imen a ion. The models gene a ed in
ou expe imen a ion a e s o ed in he logge module. Each o he algo i hms in eg a ed in
he ool is explained below.
4.3.1. Random Fo es Reg ession
Random Fo es [
55
] a e ages many models wi h noise and impa ial, educing he
a iance. T ees a e ideal candida es o bagging, as hey can eco d complex in e ac ion
s uc u es in he da a.
Fo he p edic ion o a new elemen , a ee lea is del ed h ough he ee lea es. Then,
i is assigned he label a he inal node. This p ocess is i e a i ely ailed h ough all he
ees o be assembled in he un and he node ha ob ains he highes coe icien is epo ed
as he p edic ion.
The ad an ages o andom o es s is ha hey a e highly adap able o la ge amoun s
o da a, un e icien ly and handle a la ge numbe o ea u es, and many ene gy-e iciency-
ela ed jobs a e sol ed using his ype o algo i hm [56–58].
Fo hese easons, we implemen ed his submodule, which uns he andom o es
eg ession algo i hm included in he sklea n lib a y [
59
]. The main con igu a ion allowed
in his submodule includes he numbe o K olds o be pe o med wi hin he algo i hm, he
measu emen c i e ia o each selec ed K old, and he inpu s and ou pu s o he model.
In addi ion, o he pa ame e s a e equied o u he ine- une he execu ion, such as he
dep h o he ees, he es ima ion and he minimum numbe o child en. The submodule,
as in he p e ious one, displays he sco es o he di e en uns, as well as he mean and
s anda d de ia ion o hese esul s.
The esou ces gene a e a g aph showing he ee gene a ed wi h he a ious calcula-
ions pe o med by he algo i hm, and a epo wi h hei espec i e sco es,as well as he
mean and s anda d de ia ion o hese esul s. This includes he model gene a ed by he
algo i hm. These esou ces can also be isualized in he logge module, and he models
gene a ed by he execu ion o hese algo i hms can be eused.
4.3.2. Decision T ee Reg ession
A decision ee [
27
] is a p edic ion model ha has been used in coun less ields,
especially in he ene gy ield [
34
,
60
,
61
], and is one o he mos widely used algo i hms in
his ield.
Ene gies 2023,16, 1581 16 o 29
In he use-case de eloped in his pape , he in e pola ion algo i hm based on i s nea es
neighbou s was used. Missing alues o ields such as wind speed,p ecip dep h,dew
empe a u e and ai empe a u e we e emo ed. I is impo an o no e ha his ype o
algo i hm ans o ms he o iginal da ase . The inal esul s a e shown in Figu e 8, e i ying
ha all he missing alues we e emo ed. Appendix A.3 illus a es how o emo e missing
alues h ough he alg_missing pa ame e .
Figu e 8. Remo ing missing alues.
5.6. Noise Reduc ion
The nex s ep comp ises noise educ ion. F om he p e ious explo a o y da a analysis,
we know he e a e ou lie s in ou da ase , which is easonable when dealing wi h empo al
senso da a abou ene gy.To achie e his, he z-sco e-mean algo i hm will be applied (see
Appendix A.4). This me hod iden i ies and emo es ou lie s om a da ase by calcula ing
he z-sco e o each da apoin and compa ing i o a h eshold. I he z-sco e o a da apoin is
g ea e han he h eshold, i is conside ed an ou lie and emo ed om he da ase .
To use he z-sco e-mean algo i hm, i is necessa y o speci y he h eshold alue, which
speci ies how many s anda d de ia ions a da apoin needs o be om he mean o be
conside ed an ou lie . A common h eshold alue is 3, meaning ha a da apoin needs
mo e han 3 s anda d de ia ions om he mean o be conside ed an ou lie . Howe e , he
app op ia e h eshold alue will depend on he speci ic cha ac e is ics o he da a and he
goals o he analysis. Figu e 9shows all he ou lie s o he use-case. I is impo an o no e
ha he a iables wi h he mos a iabili y in hei alues a e dew empe a u e,ai empe a u e,
and wind di ec ion, which a e also he mos in luen ial a iables in he ene gy da ase .

Ene gies 2023,16, 1581 17 o 29
Figu e 9. Ou lie s de ec ion.
5.7. Fea u e Ex ac ion
This p ocess, one o he mos c i ical in p ep ocessing, helps us o ob ain he mos
ele an a iables om ou da ase . Ou algo i hms a e shown in [37].
In his p ocess, we an a a iable co ela ion algo i hm ha shows he ela ionship
be ween ou a iables. Figu es 10 and 11 showcases he co ela ion be ween a iables dew
empe a u e and ai empe a u e. In his case, he a iables a e highly co ela ed, wi h a alue
o 0.9. Wind di ec ion and wind speed ha e a co ela ion, bu his is no e y s ong; he
alue we ob ained was 0.5.
By execu ing he algo i hm ha measu es he cha ac e is ics o he di e en a iables
in ou da ase , he eliabili y o he di e en a iables can be de e mined. In ou use-case, i
can be seen ha he mos eliable a iables a e dew empe a u e,ai empe a u e and wind
speed. The e o e, i can be e i ied ha hese a e he mos c ucial a iables o p edic ing
he ene gy consump ion o ou buildings.
In his manne , he e ec i eness o he algo i hms was e i ied wi h he ene gy da ase
and assis ed us, oge he wi h he explo a o y analysis, in conduc ing a comp ehensi e
s udy o he di e en a iables ha comp ise ou expe imen a ion.
Ene gies 2023,16, 1581 18 o 29
Figu e 10. Co ela ion be ween selec ed a iables.
Figu e 11. S a is ical measu es o selec ed a iables.
5.8. Employing Machine Lea ning Techniques o Gene a e P edic ions
One o he key poin s owa ds which ene gy-based in o ma ion p ocessing is di ec ed
is he execu ion o algo i hms o p edic ing ene gy consump ion. Fo his eason, he
decision was made o implemen hese ypes o algo i hms in TSx end. The ool can
make a p edic ion by unning classical machine lea ning algo i hms o see how p e iously
Ene gies 2023,16, 1581 19 o 29
p ep ocessed da a beha e. In his use-case, he execu ion o a single algo i hm, XGBoos , is
demons a ed. The es o he implemen a ions can be iewed in [37].
This algo i hm uses a K- old numbe o models wi h XGBOOST Reg esso algo i hms.
I makes p edic ions and s o es each esul in he a ay called sco es o ob ain he mean
squa ed e o o he ene gy consump ion p edic ion. An impo an a iable in he con igu-
a ion is n_spli s (see Appendix A.6), because i indica es he numbe o imes he ecei ed
da ase will be di ided. Finally, me ics such as mean squa ed e o ,s anda d de ia ion and
numbe o pa i ions o ou da a a e compu ed.
As discussed in Sec ion 4.5, he ou pu s will be s o ed in he logge module in eg a ed
by ML low. These esul s a e s o ed as g aphs, plain ex o iles ha can be consul ed by
use s a any ime, helping he use o in e p e he expe imen .
Tables 3and 4show he sco es and s a is ical measu es, espec i ely, o he Xgboos
algo i hm, while Figu e 12 depic s he pe o mance o he Xgboos algo i hm. In ou use-
case, he h ee mos c ucial a iables men ioned in p e ious sec ions, dew empe a u e,ai
empe a u e, and wind speed, we e selec ed. Based on he selec ed a iables ha we e used
as inpu , he Xgboos algo i hm was execu ed, and en spli s we e made o gene a e he
mean e o when making he p edic ion.
Figu e 12. Pe o mance o XGBoos algo i hm.
Table 3. Sco es o Xgboos algo i hm.
Display Sco es Sco es
0 0.0132
1 0.0130
2 0.0129
3 0.0131
4 0.0132
5 0.0126
6 0.0136
7 0.0132
8 0.0126
9 0.0134
Ene gies 2023,16, 1581 20 o 29
Table 4. Mean, s anda d de ia ion, and numbe o he Xgboos algo i hm.
MAE STD Display Sco es
0.0131 0.001 10
The esul shows us a squa ed e o o 0.0131, which is qui e low, wi h a s anda d
de ia ion o 0.001, indica ing ha he execu ion wi h he passed inpu da a is co ec .
This sugges s ha he inpu a iables a e well-co ela ed wi h he a ge a iable (ene gy
consump ion), and ha he algo i hm is e ec i ely lea ning he ela ionship be ween hem.
A es could be un o e i y he esul s, bu he goal o his wo k was o un he ool and
see ha i wo ks co ec ly.
5.9. Using Deep Lea ning Algo i hms o Make P edic ions
As we men ioned in he p e ious module, he end in he s udy o ene gy da a is
p edic ing consump ion using machine lea ning algo i hms.
Wi h his, he ool can pe o m a p edic ion execu ed on classical neu al ne wo k
algo i hms o see how he p e iously p ep ocessed da a beha e. In his use-case, he
execu ion o a single algo i hm, he long- e m memo y (LSTM) algo i hm, is demons a ed.
The es o he implemen a ions can be seen in [37].
Nex , he da a we e sequenced, and he in e als we e di ided acco ding o he n_s eps
(see Appendix A.7). Once he da a we e ob ained, we c ea ed a model using, in his case, an
LSTM ne wo k. The inpu s we e he same as hose used in he XGboos algo i hm execu ed
in he p e ious sec ion. The a iables dew empe a u e,ai empe a u e, and wind speed we e
used o p edic he building’s ene gy consump ion.
Finally, we ob ained he me ics we measu ed in ou model ( mse, mae, mse). The
LSTM model is s o ed in he sys em, and he e olu ion o he model o e he execu ion
o di e en pe iods is shown a g aph. The esul s o he expe imen a ion a e p esen ed in
Table 5and Figu e 13.
Figu e 13. G aphical ep esen a ion o he esul s o LSTM model execu ion.
Ene gies 2023,16, 1581 21 o 29
Table 5. E alua ing he pe o mance o an LSTM model in ou case s udy.
Measu e E o Display Sco es
mse 0.1292
mse 0.1055
mae 0.0167
These esul s can be used by use s, as men ioned in Sec ion 4.5, he ou pu s gene a ed
by he di e en execu ions wi hin ou sys em will be s o ed in he logge s module, whe e,
wi h he help o ML low, hey can be consul ed a any ime by he use . This will allow
o hem o pe o m an analysis o hei expe imen a ion and c ea e epo s, and also help
when making a decision abou he expe imen a ion.
In his case, he RMSE, MSE, and MAE alues we e all ela i ely low, which sugges s
ha he model is making ela i ely accu a e p edic ions. The RMSE alue o 0.1292 indica es
ha , on a e age, he model’s p edic ions a e o by abou 0.1292 uni s. The MSE alue o
0.1055 and he MAE alue o 0.0167 a e bo h lowe han he RMSE alue, which u he
suppo s he conclusion ha he model is making accu a e p edic ions.
6. Conclusions and Fu u e Wo k
This pape in oduces TSx end, a new so wa e ool designed o assis non-p og amming
use s in analysing empo al senso da a. TSx end simpli ies he p ocess o ans o ming,
cleaning, and analysing empo al senso da a h ough he use o a decla a i e language
o de ining and execu ing wo k lows. This ool aims o empowe use s o ake con ol o
hei da a, enabling hem o make imely and well-in o med decisions du ing he esea ch
p ocess. Wi h TSx end, non-p og amming use s can quickly analyse hei da a, allowing
o hem o ocus on de eloping hei esea ch and unde s anding hei esul s a he han
s uggling wi h p og amming.
The pape ’s main con ibu ion is he de elopmen o he ool i sel and he ea u es
ha make i unique. The ool was implemen ed using a modula a chi ec u e, which allows
o he s anda disa ion o di e en s ages o expe imen a ion. TSx end allows o da a
ans o ma ion, cleaning, and impu a ion, as well as he execu ion o p edic ion algo i hms
and he isualisa ion o esul s a di e en wo k low s ages. The ool also allows o he use
o di e en echniques o ime se ies analysis and makes expe imen a ion mo e accessible
o non-p og amme s. Finally, can ob ain he i s se o esul s o he analysed da ase
in a as and agile way. In his sense, his helps end-use s o es ablish imely and p ope
s a egies du ing he decision-making p ocess. The main di e ence be ween he p oposed
ool and he o he s ha we e implemen ed is he possibili y o de ining a wo k low quickly
and eadily, wi hou p io p og amming knowledge.
TSx end was implemen ed using an easy- o-use app oach. This led o a conside able
imp o emen in he use expe ience when execu ing algo i hms and ob aining ini ial
esul s, which could be used o analyse he da a. Fu he mo e, he modula a chi ec u e
o he ool enables he inco po a ion o addi ional echniques. The esul s ob ained using
he ASHRAE G ea Ene gy P edic o da ase demons a e he e ec i eness o TSx end in
analysing ene gy da a. This ool is a aluable addi ion o he ield o ime se ies analysis
and can acili a e he wo k o esea che s and p ac i ione s in a ious domains.
This ool cons i u es a s a ing poin , opening up he possibili y o u u e wo ks.
Nex , we will implemen and in eg a e mo e e icien algo i hms o p o ide end-use s wi h
mo e powe ul echniques. To make he use expe ience iendlie , we will de elop a mo e
in ui i e use in e ace. Finally, i would be in e es ing o conside he p oblems wi h a
mul i- a iable, as well as de eloping a comple e ool in a dis ibu ed en i onmen .

Ene gies 2023,16, 1581 22 o 29
Au ho Con ibu ions:
R.M.-J.: Concep ualiza ion, So wa e, Da a cu a ion, Valida ion, W i ing
O iginal D a , W i ing, W i ing Re iew and Edi ing. K.G.-B.: Me hodology, W i ing O iginal D a ,
W i ing Re iew and Edi ing, Fo mal Analysis. J.G.-R.: Concep ualiza ion, W i ing Re iew and
Edi ing, Supe ision. All au ho s ha e ead and ag eed o he published e sion o he manusc ip .
Funding:
This wo k was pa ially unded by FEDER/Jun a de Andalucía (DEEPSIM p ojec , A-
TIC-244-UGR20), Spanish Minis y o Science (SINERGY p ojec , PID2021-125537NA-I00) and he
Nex Gene a ionEU unds (IA4TES p ojec , MIA.2021.M04.0008).
Da a A ailabili y S a emen :
Da a is publicly a ailable a h ps://www.kaggle.com/compe i ions/
ash ae-ene gy-p edic ion/da a (accssed on 23 Decembe 2022). See mo e de ails in [12].
Con lic s o In e es : The au ho s decla e no con lic o in e es .
Nomencla u e
Nomencla u e and Fo mula Desc ip ion
REST API Rep esen a ional S a e T ans e (REST) Applica ion P og amming In e ace (API).
use - iendly Desc ibe applica ions, websi es, and o he digi al p oduc s ha a e easy o use and unde s and.
pipeline Se ies o s eps o s ages ha a se o da a go h ough in o de o be p ocessed, analyzed, o o he wise ans o med
end-use The people who use he so wa e ha was de eloped by he p og amme s.
oadmap Plan o s a egy ha ou lines he s eps o be aken o achie e a speci ic esea ch goal o se o goals.
se ializacion P ocess o con e ing an objec o da a s uc u e in o a o ma ha can be s o ed o ansmi ed o e a ne wo k.
dic iona y Da a s uc u e ha is used o s o e a collec ion o key- alue pai s.
measu es o mean Desc ibe he a e age alue o a da ase .
s anda d de ia ion Measu e o he sp ead o dispe sion o a da ase .
en opy Quan i y ha ep esen s he amoun o diso de o andomness in a sys em.
chi-squa e
S a is ical es used o de e mine whe he he e is a signi ican di e ence be ween an obse ed equency dis ibu ion and a
heo e ical dis ibu ion.
hea map G aphical ep esen a ion o da a, whe e indi idual alues a e ep esen ed as colo s.
bagging Technique used in ensemble lea ning o imp o e he s abili y and accu acy o p edic i e models.
ine- une Technique used in deep lea ning o adjus he pa ame e s o a p e- ained model on a new da ase .
RMSE The squa e oo o he mean o he squa ed di e ences be ween he p edic ed and ac ual alues.
MSE The mean o he absolu e di e ences be ween he p edic ed and ac ual alues.
MAE The mean o he absolu e di e ences be ween he p edic ed and ac ual alues.
Appendix A
Appendix A.1. Pa i ion Da a
Lis ing A1: Con ig/main.yaml.
1e l: pa i ion - da a
2deepl : ""
3mlea n: ""
4n_ ows : 0.0
5elemen s : ""
6ou pu _di : Da a / ain_ash ae
Lis ing A2: Con ig/pa i ion-da a.yaml.
1da e_ini : 2016 -01 -01 00:00:00
2da e_end : 2016 -12 -31 00:00:00
3 ields_include: None
4pa h_da a : ain . cs
5g oup_by_pa en : si e_id , me e
6ou pu _di : Da a / ain_ash ae
Ene gies 2023,16, 1581 23 o 29
Lis ing A3: heade unc ion.
1de Pa i ionDF (da e_ini , da e_end , pa h_da a , n_ ows , ields_include ,
g oup_by_pa en , ou pu _di )
2
3## Pa ame e s ##
4da e_ini :[ ( s ing ) Co ec Da e ] . T ans o m in da a ime da e ini , in
o de o ge da a om da ase .
5da e_end : [ (s ing ) Co ec Da e ]. T ans o m in da a imeda e end , in
o de o ge da a om da ase .
6pa h_da a : [ (s ing ) pa h ] Pa h o igin da ase .
7n_ ows : [ ( in ) ] Numbe ows da ase .
8 ields_include : [ ( lis s ing ) name ield da ase ]. Fil e by da ase
ields.
9g oup_by_pa en : [ (lis s ing ) name ield da ase ]. G oup by da ase
le els.
10 ou pu _di : [ ( s ing ) name di ec o y ]. Oupu di ec o y , o sa e da a.
Appendix A.2. Explo a o y Analysis
Lis ing A4: Con ig/main.yaml.
1e l : explo a o y - analysis
2deepl : ""
3mlea n: ""
4n_ ows : 0.0
5elemen s : ""
6ou pu _di : Da a / ain_ash ae
Lis ing A5: Con ig/explo a o y.yaml.
1 ield_x : imes amp
2 ield_y: si e_id
3g aph : line
4_ esample : W
5measu es : me e _ eading , ai _ empe a u e , dew_ empe a u e , p ecip_dep h_1_h
, sea_le el_p essu e , wind_speed
6inpu _di : Da a / es _icpe_ 2
Lis ing A6: heade unc ion.
1de Visualiza ion ( n_ ows , ield_x , ield_y , g aph , measu es , _ esample ,
inpu _di , elemen s )
2
3
4## Pa ame e s ##
5n_ ows : [ ( in ) ] Numbe s ows Da aSe . This pa ams ge om Con ig /main.
yaml
6elemen s :[ ( s ing ) name elemen s ] Fil e by elemen s . This pa ams ge
om Con ig / main . yaml
7 ield_x : [ ( s ing ) name ield ] Field X g aphs . Usually his imes amp .
8 ield_y : [ ( s ing ) name ield ] Field Y g aphs .
9g aph : [ (line | missing ) ] Line G aphs ime se ies o show missing alues
g aph .
10 measu es : [ ( lis s ing ) measu es ] De e mina e ields . Example : si e_id ,
me e
11 esample : [ ( s ing ) W, M, Y ] Resamples Week (W), Mon h (M), Y( Yea ). Only
show da a g aph .
12 inpu _di : [ ( s ing ) name di ec o y ] Inpu di ec o y o ge da a .
Ene gies 2023,16, 1581 24 o 29
Appendix A.3. Elimina ion o Missing Values
Lis ing A7: Con ig/main.yaml.
1e l : missing - alues
2deepl : ""
3mlea n: ""
4n_ ows : 0.0
5elemen s : ""
6ou pu _di : Da a / ain_ash ae
Lis ing A8: Con ig/missing- alues.yaml.
1 ields_include: None
2inpu _di : Da a / ain_ash ae
3alg_missing: in e pola e
Lis ing A9: heade unc ion.
1de missing_ alues ( n_ ows , ields_include , inpu _di , elemen s , alg_missing )
2
3
4## Pa ame e s ##
5n_ ows : [ ( in ) ] Numbe s ows Da aSe . This pa ams ge om Con ig /main.
yaml
6elemen s :[ ( s ing ) name elemen s ] Fil e by elemen s . This pa ams ge
om main.yaml
7 ield_include : [ ( lis s ing ) name ield Da aSe ] Fil e by Da aSe
ields.
8inpu _di : [ ( s ing ) name di ec o y ] Inpu di ec o y o ge da a .
9alg_missing : [ ( s ing ) name algo i hms ] Name Algo i hms missing alues .
[ in e pola e , d op ]
Appendix A.4. Elimina ion o Noise
Lis ing A10: Con ig/main.yaml.
1e l: ou lie s
2deepl : ""
3mlea n: ""
4n_ ows : 0.0
5elemen s : ""
6ou pu _di : Da a / ain_ash ae
Lis ing A11: Con ig/ou lie s.
1 ields_include: None
2inpu _di : Da a / ain_ash ae
3alg_ou lie s : z -sco e - mean
Lis ing A12: heade unc ion.
1de ou lie s ( inpu _di , n_ ows ,q1 ,q3 , ields_include , alg_ou lie s ):
2
3## Pa ame e s ##
4n_ ows : [ in ] numbe o ows o be ex ac ed , 0 ex ac s all.This pa ams
ge om main.yaml
5 ield_include : [ ( lis s ing ) name ield Da aSe ] Fil e by Da aSe
ields.
6inpu _di : [ ( s ing ) name di ec o y ] Inpu di ec o y o ge da a .
7alg_ou lie s : [ ( s ing ) name algo i hms ] Name Algo i hms ou lie s . [
z_sco e_me hod_mean]
8q1 : [ ( in ) ] % ou lie s emo e .
9q3 : [ ( in ) ] % ou lie s emo e .
Ene gies 2023,16, 1581 25 o 29
Appendix A.5. Fea u e Ex ac ion
Lis ing A13: Con ig/main.yaml.
1e l : ea u e - selec ion
2deepl : ""
3mlea n: ""
4n_ ows : 0.0
5elemen s : ""
6ou pu _di : Da a / ain_ash ae
Lis ing A14: Con ig/ ea u e-selec ion.yaml.
1 ields_include : me e _ eading , ai _ empe a u e , dew_ empe a u e ,
p ecip_dep h_1_h , sea_le el_p essu e , wind_speed
2inpu _di : Da a / ain_ash ae
3alg_ s : FSMeasu es
Lis ing A15: heade unc ion.
1de ea u e_selec ion ( n_ ows , ields_include , inpu _di , elemen s , alg_ s )
2
3## Pa ame e s ##
4n_ ows : [ ( in ) ] Numbe s ows Da aSe . This pa ams ge om Con ig /main.
yaml
5elemen s :[ ( s ing ) name elemen s ] Fil e by elemen s . This pa ams ge
om main.yaml
6 ield_include : [ ( lis s ing ) name ield Da aSe ] Fil e by Da aSe
ields.
7inpu _di : [ ( s ing ) name di ec o y ] Inpu di ec o y o ge da a .
8alg_ s : [ ( s ing ) name algo i hms ] Name Algo i hms ea u e selec ion . [
FSMeasu e , co ela ion ]
Appendix A.6. P edic ion by Running Machine Lea ning Algo i hms
Lis ing A16: Con ig/main.yaml.
1e l: ""
2deepl : ""
3mlea n : xgb
4n_ ows : 0.0
5elemen s : ""
6ou pu _di : Da a / ain_ash ae
Lis ing A17: Con ig/xgb.yaml.
1model_inpu : ai _ empe a u e , cloud_co e age , dew_ empe a u e ,
p ecip_dep h_1_h , sea_le el_p essu e , me e _ eading
2model_ou pu : me e _ eading
3n_spli s : 5
4objec i e : eg : squa ede o
5inpu _di : Da a / ain_ash ae
6