HYPERGRYD. This p ojec has ecei ed unding om he Eu opean Union’s Ho izon 2020 esea ch and
inno a ion p og amme unde g an ag eemen No 101036656
WP3 – ICT Modules and Simula ion Tools
Task 3.2 - De elopmen o demand esponse
managemen , p edic ion, and op imisa ion
me hods o 4 h and 5 h DHC ne wo ks
D3.3 - Me hodologies and amewo ks
o op imize demand esponse and peak
load sha ing o 4 h-5 h DHC in h ee
ypical Eu opean clima es.
Re . A es(2024)6909400 - 30/09/2024
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 2
DISCLAIMER
The opinion s a ed in his epo e lec s he opinion o he au ho s and no he opinion o he
Eu opean Commission.
All in ellec ual p ope y igh s a e owned by HYPERGRYD conso ium membe s and a e p o ec ed by
he applicable laws. Rep oduc ion is no au ho ised wi hou p io w i en ag eemen .
The comme cial use o any in o ma ion con ained in his documen may equi e a license om he
owne o ha in o ma ion.
ACKNOWLEDGEMENT
This p ojec has ecei ed unding om he Eu opean Union’s Ho izon 2020 esea ch and inno a ion
p og amme unde g an ag eemen Nº 101036656.
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 3
P ojec
P ojec Ac onym
HYPERGRYD
P ojec Ti le
Hyb id coupled ne wo ks o he mal-elec ic in eg a ed Sma Ene gy Dis ic s
G an Ag eemen
numbe
101036656
Call iden i ie
H2020-LC-GD-2020
Topic iden i ie
LC-GD-2-1-2020
Inno a i e land-based and o sho e enewable ene gy echnologies and hei
in eg a ion in o he ene gy sys em
Funding Scheme
Resea ch and Inno a ion Ac ion
P ojec du a ion
42 mon hs (F om 1 Oc obe 2021)
Coo dina o
ARCbcn
Websi e
h p://hype g yd.eu
Deli e able
Deli e able No.
3.3
Deli e able i le
Me hodologies and amewo ks o op imize demand esponse and peak load sha ing
o 4 h-5 h DHC in h ee ypical Eu opean clima es
Desc ip ion
D3.3 (Me hodologies and amewo ks o op imize demand esponse and peak load
sha ing o 4 h-5 h DHC in h ee ypical Eu opean clima es) summa ize he main
ac i i ies in Task 3.2 (De elopmen o demand esponse, p edic ion, and op imisa ion
me hods o 4 h 5 h DHC ne wo ks)
Mos powe - o-hea solu ions ope a e independen ly, so his ask aims o in eg a e
demand esponse a a highe le el, op imizing s o age (buildings, subs a ions, piping
ne wo ks) using su plus o low-exe gy hea /cooling sou ces. I will de elop
echniques o ene gy o ecas ing and dynamic con ol s a egies, using machine
lea ning me hods like A i icial Neu al Ne wo ks and Gene ic Algo i hms. KTH will
apply ad anced ML echniques o RES models (was e hea , P2H, CHP, e c.), wi h DHC
models using wea he da a om pa ne s (SONNE, ENVI, EURAC). ML me hods will be
e alua ed o compu a ional cos , p i acy, and accu acy. GSY and ENCO will es and
in eg a e he de eloped algo i hms.
WP No.
WP3
Rela ed ask
Task 3.2 - De elopmen o demand esponse, p edic ion, and op imisa ion me hods
o 4 h 5 h DHC ne wo ks
Lead Bene icia y
3 - KTH
Au ho (s)
Mus apha Habib (KTH), Qian Wang (KTH)
Con ibu o (s)
Jo ge Leao Pi occhi (IDP)
Type
R
Dissemina ion
PU
Language
English – GB
Due
30/09/2024
Submission da e
30/09/2024
Ve sion
Da e
Au ho s
Desc ip ion
V.0.1
13/09/2024
Mus apha Habib (KTH)
Fi s e sion o in e nal e iewing
V.0.2
27/09/2024
Da id Ve ez (ARC)
Re iew
V.1.0
28/09/2024
Jo ge Leao Pi occhi (IDP)
Re iew
V.2.0
28/09/2024
Mus apha Habib (KTH)
Final e sion a e e iew
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 4
Table o Con en s
1. Execu i e Summa y .................................................................................................... 7
2. In oduc ion ................................................................................................................ 9
2.1. Scope ......................................................................................................................... 9
2.2. Audience ................................................................................................................... 9
2.3. Abb e ia ions ............................................................................................................ 9
2.4. Con ibu ions o pa ne s ....................................................................................... 10
2.5. Rela ion o o he ac i i ies ..................................................................................... 11
2.6. S uc u e ................................................................................................................. 11
3. Me hodology ............................................................................................................ 12
3.1. Machine and deep lea ning me hods applied in his esea ch. ............................. 12
3.1.1. Da a clus e ing ........................................................................................................ 12
3.1.1.1. Euclidean dis ance measu e ................................................................................... 14
3.1.1.2. DTW dis ance measu e ........................................................................................... 14
3.1.2. DHN hea powe p edic ion using neu al ne wo ks ............................................... 15
3.1.2.1. Mul i-laye pe cep on ........................................................................................... 16
3.1.2.2. T aining ................................................................................................................... 18
3.2. Me aheu is ic op imisa ion applied on DHN-HP .................................................... 19
3.2.1. Sys em modelling and alida ion ............................................................................ 20
3.2.2. Op imisa ion p oblem o mula ion ........................................................................ 21
3.2.3. Pa icle Swa m Op imisa ion .................................................................................. 22
3.3. Op imal con ol o HP-RES-TES wi hin DHN. .......................................................... 23
3.3.1. Sys em modelling .................................................................................................... 24
3.3.1.1. Hea pump and ene gy s o age sys em. ................................................................. 24
3.3.1.2. s a e o cha ge modelling o he ba e y ene gy s o age sys em .......................... 24
3.3.1.3. Nonlinea op imisa ion ........................................................................................... 25
4. Me hod alida ion wi hin use-cases ........................................................................ 26
4.1. No dic clima e – da a clus e ing o handling low-quali y da a om DHN me e s .. 26
4.1.1. Time-se ies clus e ing .............................................................................................. 27
4.1.1.1. Nu sing homes da a clus e ing ................................................................................ 27
4.1.1.2. Schools da a clus e ing ............................................................................................. 29
4.2. Load p edic ion ........................................................................................................ 31
4.2.1. Long- e m assessmen ............................................................................................. 31
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 5
4.2.2. Seasonal assessmen ................................................................................................ 34
4.3. Con inen al/Alpine Clima e – op imal demand esponse managemen in DHN a a
communi y le el ..................................................................................................................... 37
4.3.1. Use case desc ip ion. ................................................................................................ 37
4.3.2. Simula ion esul s ..................................................................................................... 39
4.4. Medi e anean Clima e – op imal con ol o hea pump wi h a pho o ol aic/s o age
sys em in cen alized DHN subs a ion ................................................................................... 41
4.4.1. Use-case desc ip ion ................................................................................................ 41
4.4.2. Simula ion esul s ..................................................................................................... 43
4.4.2.1. Sys em pa ame e s and me hod alida ion en i onmen ....................................... 43
Conclusions ............................................................................................................................. 48
Summa y o achie emen s .................................................................................................... 48
Rela ion o con inued de elopmen s. ................................................................................... 49
Re e ences .............................................................................................................................. 50
Lis o Figu es
Figu e 1 Da a collec ion and analysis lowcha . ................................................................................. 13
Figu e 2 Di e en aligning ules o (a) Euclidean dis ance and (b) DTW dis ance ma ching. ........... 15
Figu e 3 Example o an MLP model s uc u e. .................................................................................... 17
Figu e 4 DH/HP and TES connec ion opology in he p oposed subs a ion scheme. ......................... 19
Figu e 5 Simple o e iew o he connec ion opology o HESS in he cen alized hea ing subs a ion.
............................................................................................................................................................. 24
Figu e 6 Clus e ed da ase s o nu sing homes o he yea s 2017, 2018, and 2019 using Euclidean k-
means. ................................................................................................................................................. 28
Figu e 7 Clus e ed da ase s o nu sing homes o he yea s 2017, 2018, and 2019 using DTW k-means.
............................................................................................................................................................. 29
Figu e 8 Clus e ed da ase s o schools o he yea s 2017, 2018, and 2019 using Euclidean k-means.
............................................................................................................................................................. 30
Figu e 9 Clus e ed da ase s o schools o he yea s 2017, 2018, and 2019 using DTW k-means. ... 31
Figu e 10 Annual p edic ion assessmen using a school da ase ex ac ed om Clus e 1 (a) using
Euclidean k-means (b) using DTW k-means. ....................................................................................... 32
Figu e 11 Annual p edic ion assessmen using a school da ase ex ac ed om Clus e 2 (a) using
Euclidean k-means (b) using DTW k-means. ....................................................................................... 32
Figu e 12 Annual p edic ion assessmen using a nu sing home da ase ex ac ed om Clus e 1 (a)
using Euclidean k-means (b) using DTW k-means. .............................................................................. 33
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 6
Figu e 13 Annual p edic ion assessmen using a nu sing home da ase ex ac ed om Clus e 2 (a)
using Euclidean k-means (b) using DTW k-means. .............................................................................. 34
Figu e 14 P edic ion pe o mance o a school da ase (a) in win e (b) in summe . ........................ 35
Figu e 15 P edic ion pe o mance o a nu sing home da ase (a) in win e (b) in summe .............. 35
Figu e 16 P edic ion pe o mance in mild-season (a) o a nu sing home (b) o a school ................ 36
Figu e 17 Sa elli e iew o G oßschönau communi y wi h buildings conside ed o op imal ene gy
coo dina ion. ....................................................................................................................................... 38
Figu e 18 Ene gy op imisa ion lowcha . ........................................................................................... 38
Figu e 19 Hea demand wi h PV powe p oduc ion o he conce ned buildings. .............................. 40
Figu e 20 HP op imal condense empe a u e se poin wi h associa ed TES SOC a ia ion in a 2-day
simula ion (use -cen ic scena io) ....................................................................................................... 40
Figu e 21 HP op imal condense empe a u e se poin wi h associa ed TES SOC a ia ion in a 2-day
simula ion (communi y-cen ic scena io) ........................................................................................... 41
Figu e 22 Elec ic powe dis ibu ion in he ENVIPARK esea ch acili y. ........................................... 42
Figu e 23 Hea ing dis ibu ion ne wo k in ENVIPARK. ........................................................................ 42
Figu e 24 Simula ion condi ions (a) hea load (b) PV powe (c) elec ici y p ice. ............................... 44
Figu e 25 HP condense empe a u e se poin : (a) wi h lexibili y Scena io 1 (b) wi h lexibili y
Scena io 2. ........................................................................................................................................... 45
Figu e 26 BESS powe se poin : (a) wi h lexibili y Scena io 1 (b) wi h lexibili y Scena io 2. ............ 45
Figu e 27 HP supplied hea : (a) wi h lexibili y Scena io 1 (b) wi h lexibili y Scena io 2. .................. 46
Figu e 28 BESS SOC a ia ion: (a) wi h lexibili y scena io 1 (b) wi h lexibili y scena io 2. ............... 47
Figu e 29 TES maximum wa e empe a u e: (a) wi h lexibili y Scena io 1 (b) wi h lexibili y Scena io
2. .......................................................................................................................................................... 47
Lis o Tables
Table 1 MLP aining pa ame e s ........................................................................................................ 19
Table 2 S a is ic measu es o he gene a ed no malized da ase s in clus e -wise. ............................ 31
Table 3 MSE and MAE e alua ion using nu sing homes no malized da a o he yea 2019. ............ 37
Table 4 Simula ion pa ame e s ........................................................................................................... 39
Table 5 PSO pa ame e s ...................................................................................................................... 39
Table 6 Sys em pa ame e s ................................................................................................................. 43
Table 7 Flexibili y impac assessmen o di e en op imisa ion me hods. ........................................ 48
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 7
1. Execu i e Summa y
The goal o he HYPERGRYD p ojec is he de elopmen o a se o eplicable and scalable cos -
e ec i e echnical solu ions o allow he in eg a ion o Renewable Ene gy Sou ces (RES) wi h
di e en dispa chabili y and in insic a iabili y inside The mal G ids as well as hei link wi h he
Elec ical G ids, including he de elopmen o inno a i e key componen s, in pa allel wi h inno a i e
and in eg a ed ICT se ices o med by a scalable sui e o ools o he p ope handling o he inc eased
complexi y o he sys ems om building o Local Ene gy Communi y (LEC) le els and beyond, and
accele a e he sus ainable ans o ma ion, planning, and mode niza ion o Dis ic Hea ing and
Cooling (DHC) owa ds 4 h and 5 h gene a ion.
HYPERGRYD also aims o de elop eal- ime managemen o elec ical and he mal ene gy lows in he
coupled ene gy ne wo k complex, including he syne gies be ween hem. The e o e, HYPERGRYD
aims a h ee o e -a ching Gene al Objec i es:
• To p o e Sma Ene gy Ne wo ks as he u u e o E icien Ene gy Managemen in DHC in
syne gy wi h he Elec ical G ids in LEC/Sma Ci ies o he u u e;
• To de ine he oadmap o design and plan u u e DHC as well as he mode niza ion o he
exis ing ones in di e en clima es and RES pene a ion le els owa d 4 h-5 h gene a ion,
• To demons a e HYPERGRYD RES-based Enabling Technologies, Sma Ene gy G id Solu ions
empowe ed by new ICT ools and se ices as he key o his e olu ion.
Du ing he p ojec , HYPERGRYD’s solu ions will be implemen ed ac oss ou Li e-In-Labs cases in
h ee ep esen a i e clima es, wi h special conside a ion o hei cos e ec i eness and po en ial
eplicabili y o inally achie e hese h ee main objec i es. One o he main solu ions p oposed by
HYPERGRYD is he exploi a ion o DHN da a, ypically gene a ed by sma senso s and ene gy me e s,
o come up wi h applicable ools ha enable he implemen a ion o op imal ene gy managemen o
mode n DH ne wo ks ha inco po a e enewables and low- empe a u e supply. This will open he
pa h owa ds a smoo h ansi ion o ac ual DH ne wo ks o 4 h and 5 h gene a ion DH.
The pu pose o his deli e able in sho is o:
• De elop di e en machine lea ning echniques designed o da a clus e ing o deal wi h low-
quali y da a issued om senso s and ene gy me e s ins alled in DHN subs a ions, and o
iden i y abno mal pa e ns and p o iles.
• De elop a deep lea ning echnique o long- e m p edic ion o DH load in di e en building
ca ego ies.
• Design and simula e an op imal coo dina ed demand side managemen based on
disagg ega ed da a o a DH-powe ed communi y.
• Design and simula e a cen alized demand-side managemen based on agg ega ed da a
applicable o he main DHN subs a ion in he p esence o di e en ene gy s o age
echnologies, enewables, and hea pumps.
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 8
• Valida e he abo emen ioned echniques on eal da ase s gene a ed in h ee di e en
Eu opean clima es.
On behal o Au ho s
Mus apha Habib, KTH
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 9
2. In oduc ion
2.1. Scope
This deli e able ou lines he de elopmen and alida ion o da a-d i en echniques o op imizing
ene gy usage in DHN-powe ed buildings and communi ies, le e aging da a om sma me e s,
senso s, and building managemen sys ems. I explo es how such da a can e eal end-use beha iou
o in o m ad anced con ol and managemen policies. The scope includes clus e ing and p edic ion
me hods, combined wi h me aheu is ic and de e minis ic op imisa ion algo i hms, o achie e
op imal con ol and demand managemen . The solu ions p esen ed capi alize on ene gy lexibili y
wi hin DHN communi ies, o e ing bo h dis ibu ed con ol ia ad anced IoT communica ion and
cen alized con ol based on agg ega ed da a a he main DHN subs a ion. In he scope o hese
echniques, his deli e able p oposes a se o ools ha can se e in upg ading he cu en DHN
owa d he 4 h and 5 h gene a ions.
2.2. Audience
This deli e able p o ides ou comes ha can be applicable inpu s o Task 3.3 (De elopmen o no el
machine lea ning echniques d i en by IoT o LEC-in eg a ed DHC ne wo ks), in addi ion o he
ac i i ies in WP5 conce ning demons a ion and alida ion s a egies in demo si es.
2.3. Abb e ia ions
RES: Renewable Ene gy Sou ces
LEC: Local Ene gy Communi ies
DHN: dis ic hea ing ne wo k
DC: Di ec cu en
AC: Al e na i e cu en
ML: Machine lea ning
DL: Deep lea ning
ESS: ene gy s o age sys em
TES: The mal ene gy s o age
BESS: Ba e y ene gy s o age sys em
HP: Hea pump
IoT: In e ne o hings
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 16
One model is o be c ea ed o each clus e (de eloped in 3.1.1). In addi ion o ou doo wea he
condi ions, hea ene gy demand in buildings can be a ec ed by many o he ac o s. In his wo k, he
chosen da a inpu s, o each hou , a e lis ed below:
• The o ecas ed ou doo empe a u e (in °C) is ypically accessible ia an online se ice.
• Hou o he day: om 1 ep esen ing "1:00-2:00" o 24 ep esen ing "23:00-00:00",
• Day o he week: an in ege , om 1 o Monday o 7 o Sunday,
• Mon h o he yea , an in ege , om 1 o Janua y o 12 o Decembe ,
• Holiday indica o , a Boolean, 1 o a holiday and 0 o a wo king day.
Those pa ame e s a e se basically o gi e an es ima ion o he hea load end, which changes
essen ially acco ding o di e en imeslo s. They can also gi e an app oxima e assessmen o he
occupancy a e and human ac i i ies ela ed o he building ype (M.M. Alam e al., 2017).
As he analysed buildings a e no esiden ial, i can be possible o ind some indica o s ha ing
co ela ions wi h he occupancy a e wi h an accep ed app oxima ion, which is he eason behind he
choice o he las ou ANN inpu s lis ed abo e. In his con ex , "hou o he day" can gi e an
app oxima e guess o he daily scheduled wo k ime o schools and nu sing homes. In he same way,
"day o he week" has been in oduced o di e en ia e be ween wo king days and weekends, which
is mo e signi ican in he case o schools as nu sing homes s ill ha e pa ial wo k ac i i ies on
weekends. Wi h he same impac , he inpu s "mon h o he yea " and "holidays indica o " ha e been
in oduced o highligh aca ion pe iods.
3.1.2.1. Mul i-laye pe cep on
A Mul i-Laye Pe cep on (MLP) is a ype o eed o wa d a i icial neu al ne wo k ha consis s o
mul iple laye s o in e connec ed nodes, called neu ons o uni s. I is a ounda ional and widely used
a chi ec u e in he ield o deep lea ning [4-5]. The e m "pe cep on" e e s o a basic building block
ha models a single neu on, and "mul i-laye " indica es ha mul iple laye s o hese pe cep ons a e
s acked oge he .
In an MLP, he neu ons a e o ganized in o laye s, ypically consis ing o an inpu laye , one o mo e
hidden laye s, and an ou pu laye . The inpu laye ecei es he inpu da a, and he ou pu laye
p oduces he inal ou pu o he model. The hidden laye s a e in e media e laye s be ween he inpu
and ou pu laye s, esponsible o p ocessing and ans o ming he inpu in o ma ion h ough a se ies
o non-linea ope a ions. Figu e 3 shows an illus a i e a chi ec u e o an MLP wi h an inpu laye o
i e neu ons; h ee hidden laye s wi h 12 neu ons each and one ou pu laye wi h one single neu on.
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 17
Figu e 3 Example o an MLP model s uc u e.
Each neu on in he MLP is associa ed wi h a se o lea nable pa ame e s, including weigh s and biases.
The weigh s de e mine he s eng h o he connec ions be ween neu ons, while he biases in oduce
an o se o bias e m o he neu on's ou pu . These pa ame e s a e adjus ed du ing he aining
p ocess o op imize he pe o mance o he MLP on a gi en ask.
The unc ioning o an MLP in ol es wo main s eps: o wa d p opaga ion and backwa d p opaga ion.
In o wa d p opaga ion, he inpu da a is ed h ough he ne wo k, and he ac i a ions o each neu on
a e compu ed laye by laye , ul ima ely p oducing he ou pu . Backp opaga ion in ol es compu ing
he g adien s o he ne wo k's pa ame e s conce ning a chosen loss unc ion, allowing o he
adjus men o he weigh s and biases in a di ec ion ha minimizes he loss. This i e a i e p ocess o
o wa d and backwa d p opaga ion is epea ed un il he model con e ges o an op imal s a e. Eq (4)
o mula es he o wa d p ocess h ough MLP a chi ec u e.
1
( ) ( ) ( ) ( 1)
01
( ) ( ( ))
L
U
L L L L
ii
i
X a w w X
−−
=
=+
(4)
Whe e:
L
is he laye numbe .
()L
is he
L
unc ion laye .
()
0L
w
and
()L
i
w
a e espec i ely bias and weigh ec o s co esponding o laye
L
.
X
is he inpu ec o .
a
is he ac i a ion unc ion (i can ep esen di e en ac i a ion unc ions be ween di e en laye s).
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 18
1L
U−
is he numbe o laye s be o e L.
The chosen ac i a ion unc ion is he Sigmoid unc ion, i is o mula ed as:
1
() 1x
ax e−
=+
(5)
3.1.2.2. T aining
Du ing he aining phase, an op imisa ion algo i hm is used o upda e he weigh s and biases o he
MLP o minimize a loss unc ion, which is o mula ed in (6). As an op imisa ion me hod, his s udy
add esses he use o Limi ed-memo y B oyden-Fle che -Gold a b-Shanno (LBFGS). I is an
op imisa ion algo i hm used o sol ing uncons ained op imisa ion p oblems. The algo i hm
combines he BFGS me hod wi h limi ed memo y s o age, allowing i o app oxima e he in e se
Hessian ma ix wi hou explici ly compu ing o s o ing i . The limi ed-memo y aspec o LBFGS makes
i memo y-e icien and sui able o la ge-scale op imisa ion p oblems. Howe e , he de ailed
ma hema ical desc ip ion o LBFGS is ou side he scope o his a icle.
(6)
Whe e and a e he ac ual and p edic ed DH load espec i ely,
P
N
is he numbe o da a samples.
The me hod "neu al_ne wo k.MLPReg esso ", which is a pa o he Py hon package "sklea n", was
chosen as an implemen a ion ool. The aining pa ame e s a e se in Table 1, and hose pa ame e s
a e eached a e some ial-and-e o es s. As aining da a, wo da ase s o hou ly DH hea
consump ion o 2017 and 2018 o nu sing homes and educa ional buildings loca ed in T ondheim,
No way a e used (mo e in o ma ion abou he used da ase s will be gi en in 4.1.1).
The pseudocode o a ained MLP is summa ized below:
Ini ializing
(1)
0
w
and
Acqui ing aining da a X
While ( aining s opping c i e ia is no me ){
Calcula ing he loss unc ion
Upda ing weigh s and bias o all laye s
MLP eed o wa d}
(1)
i
w
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 19
Table 1 MLP aining pa ame e s
MLP pa ame e
alue
Numbe o neu ons in he inpu laye
5
Numbe o hidden laye s
5
Numbe o neu ons in one hidden laye
50
numbe o neu ons in he ou pu laye
1
Ac i a ion unc ion
Sigmoid
Op imize
LBFGS
Lea ning a e
Cons an (0.1)
Loss unc ion
MSE
3.2. Me aheu is ic op imisa ion applied on DHN-HP
In his s udy, we assume a i ual scena io, in which, we ha e a esiden ial communi y
connec ed o DHN, in which, HPs wi h TES a e ins alled a each building. The pu pose behind his
assump ion is o assess he impac o he maximum ene gy lexibili y o e ed by HP-TES in he op imal
exploi a ion o he DHN ene gy. Howe e , in eali y, his assump ion may no always be alid.
he p oposed me hod consis s o adjus ing a each ime s ep (one hou ), in coo dina ion, he s a e o
cha ge (SOC) o TES uni s in all buildings in he DHN communi y by a ying he co esponding HP
condense empe a u es. Fo me hod alida ion, his sec ion add esses he de elopmen o
ma hema ical models necessa y o simula ion, as well as he op imisa ion p oblem o mula ion. As
s a ed ea lie , a i ual scena io is es ed, in which, all buildings a e connec ed o he same DHN
ne wo k ia HP/TES uni s. The connec ion opology assumes ha HPs a e used as DH empe a u e
boos e s as desc ibed in Figu e 4. The e o e, he de eloped me hods can be applied as an assessmen
amewo k owa d he in eg a ion o 4 h and 5 h gene a ion DH (4GDH and 5GDH) which is
cha ac e ized by low- empe a u e supply wi h high pene a ions o local RES.
Figu e 4 DH/HP and TES connec ion opology in he p oposed subs a ion scheme.
Fo a success ul me hod implemen a ion, a ew simpli ying assump ions a e p oposed as below:
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 20
The DHN empe a u e supply is educed by 20 % (compa ed o he cu en empe a u e le el) o
simula e 5GDH.
1. All HP uni s a e equipped wi h a iable-speed comp esso s.
2. The possibili y o he buildings connec ed o he same low- ol age (LV) g id o sha e hei
ex a PV powe s (in e -communi y).
3. The only conside ed elec ici y consump ion ( o op imisa ion) is he one ela ed o he HP
comp esso 's ope a ion.
3.2.1. Sys em modelling and alida ion
Fo HPs, he coe icien o pe o mance (COP) is de ined based on he sou ce and sink
empe a u es as desc ibed in (7), which,
sou ce
T
and
sink
T
a e he sou ce (DHN supply) and sink
(condense ) empe a u es in HP espec i ely [Kel in]. 𝜂𝐻𝑃 is an empi ical coe icien ha calib a es
he ideal COP o he eal one, in ou case, i is ixed a 0.15.
sin
sin s
k
HP
k ou ce
T
COP TT
=−
(7)
The consumed elec ici y
_el HP
P
[kW], ela ed o a supplied HP hea
_HP ou
Q
[kW] is o mula ed in
(8). I is mainly ela ed o comp esso ope a ion.
_
_
HP ou
el HP
Q
PCOP
=
(8)
Ideally, he hea consumed by HPs in he e apo a ion p ocess is o mula ed in (9) (neglec ing hea
was es in he HP uni s):
_ _ _HP in HP ou el HP
Q Q P=−
(9)
As i will be explained la e ,
_HP in
Q
is no hing bu he hea ex ac ed om DH ne wo k [kW].
Rega ding TES, a s a i ied mul i-node model o wa e anks is adop ed. The di e en ial equa ion o
he e alua ion o wa e empe a u e in one node
i
in he wa e ank is o mula ed in (10):
, , , , ,
i
i p IN i OUT i COND i CONV i LOSS i
dT
mC Q Q Q Q Q
d = − + + −
(10)
Whe e
,IN i
Q
and
,OUT i
Q
a e espec i ely he supplied and ex ac ed hea a he node
i
[kW]. In his
s udy, his applies only o he i s node (
1i=
);
i
m
is he wa e mass o laye
i
[kg];
Cp
is he wa e
speci ic hea (4182 J/kg.°C).
,CONV i
Q
is he hea powe gained o los ia con ec ion mode a he node
i
. i is o mula ed as below [kW]:
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 21
,.( ). .( ( 1) ( )) .( ). .( ( ) ( 1))
CONV i UP IN OUT DOWN OUT IN
Q m m Cp T i T i m m Cp T i T i
= − − − − − − +
(11)
Whe e
IN
m
and
OUT
m
a e espec i ely he inle and ou le wa e lows [kg/s];
UP
and
DOWN
a e
espec i ely he low di ec ion indica o s:
1
0
IN OUT
UP
i m m
else
=
,
1
0
OUT IN
DOWN
i m m
else
=
(12)
,COND i
Q
is he hea ans e ed o o om laye
i
ia conduc ion mode [kW], i is o mula ed as below:
,
..
( ( 1) ( )) ( ( ) ( 1))
COND i
k A k A
Q T i T i T i T i
dX dX
= − − − − +
(13)
Whe e
k
is he he mal conduc i i y o wa e (0.598 W/m.K),
A
is he c oss-sec ion a ea o he
wa e [m2];
dX
is he model node hickness [m].
,LOSS i
Q
is he hea loss h ough he ank me al wall in laye
i
[kW], i is o mula ed as below:
,. .( ( ) )
LOSS i AMB
Q kwU T i T=−
(14)
Whe e
kw
is he mal conduc i i y o he wall (2.5 J/m2.°C); U is he wa e node la e al su ace [m2];
and
AMB
T
is he ambien empe a u e, chosen 23 °C.
To upda e he wa e empe a u e a each node, he di e en ial equa ion is sol ed nume ically using
he equa ion below:
, , ,
2
( ) ( )
. . . . COND i CONV i LOSS i
T i Q Q Q
Cp R dX
= + −
(15)
Whe e
()Ti
is he empe a u e inc emen a ion a laye
i
[° C];
is he wa e densi y [kg/m3];
R
is he diame e o he wa e c oss-sec ion a ea in he ank [m].
3.2.2. Op imisa ion p oblem o mula ion
As a close app oxima ion o he TES SOC [%], in his s udy, he a e age wa e empe a u e ac oss he
ank heigh is used as desc ibed in (16), in which
MAX
T
is he maximum empe a u e ha can he
s o age ank be ope a ed wi h ( ixed a 65 °C); N is he numbe o nodes in he ank model.
1
100. .
N
i
i
MAX
T
SOC NT
=
=
(16)
The hea supplied by HPs o TES uni s is o mula ed as below:
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 22
. .( (0)) (0)
0
p HP COND HP COND
HP OUT
mC T T i T T
Qelse
−−
−
−
=
(17)
Whe e
HP COND
T−
is he HP condense se poin , o be de ined by he op imize a each ime s ep [° C];
(0)T
is wa e empe a u e a he TES uppe laye (
0i=
) [° C]. When he op imize wan s o u n o
he HP, i sends condense empe a u es se poin s lowe han he sou ce empe a u e (55 °C) o
p e en hea low e e se in simula ion.
Op imisa ion has wo main weigh ed e ms in he objec i e unc ion o mula ed in (18): one ela ed
o he ene gy cos educ ion a each building, and one ela ed o he whole communi y as one en i y.
2
_ , , , , _ , , , , , ,
1 1 1 1 1 1
( ). ( )
PV
BN
N
H N H M
el HP i PV i el HP i PV i loss j
i i i j
J P P EP P P P
= = = = = =
= − + − −
(18)
Whe e:
H
is ime op imisa ion ime ho izon in hou s [hou ];
B
N
is he numbe o conce ned
buildings;
,,PV i
P
is he supplied ene gy om he PV sys em ins alled on building
i
.
i
a he ime
[kWh] ( o buildings ha ing no PV sys ems, his a iable is se o ze o);
EP
is he
elec ici y p ice a he ime
[€/kWh], i has wo possible alues: one ela ed o buying p ice and one
o selling one;
PV
N
is he numbe o ins alled PV sys ems;
,,loss j
P
is he elec ic loss o he LV
connec ion cables
j
a he ime
[kW];
and
a e weigh ing coe icien s.
A se o cons ain s a e o be espec ed du ing he op imisa ion, which a e he minimum and
maximum SOC limi s o each TES, and he HP condense empe a u e se poin as o mula ed below:
MIN MAX
SOC SOC SOC
(19a)
HP COND MIN HP COND HP COND MAX
T T T
− − − − −
(19b)
Acco ding o he alues o he weigh ing coe icien s, he op imisa ion c i e ia can be de ined,
ei he by going owa d educing he elec ici y cos o each consume indi idually (by inc easing
) o owa d maximizing he ene ge ic independence o he en i e communi y by sha ing a ailable PV
ene gies be ween di e en buildings (by inc easing
). Fo he las objec i e unc ion e m, he
elec ic powe losses be ween di e en buildings a e o be aken in o conside a ion, he e o e,
in o ma ion abou he in e nal low- ol age cabling inside he communi y is needed.
3.2.3. Pa icle Swa m Op imisa ion
Pa icle Swa m Op imisa ion (PSO) is selec ed as he op imisa ion echnique. I was de eloped
by Ebe ha and Kennedy (Kennedy and Ebe ha 1995), inspi ed by he social beha iou o bi d
locking o ish schooling. PSO sha es many simila i ies wi h e olu iona y compu a ion echniques
such as Gene ic Algo i hms. I is sui able o op imisa ion p oblems wi h a high numbe o decision
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 23
a iables and lexible objec i e unc ions, which is he use-case o his s udy. The sys em is ini ialized
wi h a popula ion o andom alues o solu ions and sea ches o op ima by upda ing gene a ions.
Howe e , unlike GA, PSO has no e olu ion ope a o s such as c osso e and mu a ion. In PSO, he
po en ial solu ions, called pa icles, ly h ough he p oblem space by ollowing he cu en op imum
pa icles. De ailed in o ma ion will be gi en in he ollowing sec ions.
PSO is ini ialized wi h a g oup o andom pa icles (solu ions) and hen sea ches o op ima by
upda ing gene a ions. In e e y i e a ion, each pa icle is upda ed by he ollowing wo "bes " alues:
he i s one is he bes solu ion ( i ness) i has achie ed so a , and he alue is called
pbes
. Ano he
"bes " alue ha is acked by he pa icle swa m op imize is he bes alue ob ained so a by any
pa icle in he popula ion. This bes alue is a global bes and called
gbes
.
A e inding he wo bes alues, he pa icle upda es i s eloci y and posi ions wi h he ollowing
equa ion (20a) and (20b):
( ) ( )
1 1 2 2
( 1) ( ) . ( ) ( ) . ( ) ( ) i i c pbes i p i c gbes i p i+ = + − + −
(20.a)
( 1) ( ) ( 1)p i p i i+ = + +
(20.b)
Whe e i he numbe o i e a ions;
() i
is he pa icle eloci y;
()pi
is he cu en pa icle (solu ion);
()pbes i
and
()gbes i
a e he bes posi ions o he pa icle and he g oup espec i ely.
1
and
2
is a andom numbe be ween 0 and 1.
1
c
and
2
c
a e lea ning ac o s.
3.3. Op imal con ol o HP-RES-TES wi hin DHN.
This subsec ion p esen s and explains he op imisa ion-based con ol o he hyb id ene gy s o age
sys em (HESS) o local ene gy communi y (LEC) managemen . Fi s , he p oposed hyb idiza ion
a chi ec u e is p esen ed and explained. Then, he ma hema ical models o he in ol ed sys ems in
HESS a e de ined, which will be he basis o he simula ion s udy la e . Using he de eloped models,
he nonlinea op imisa ion algo i hm will be o mula ed. Finally, he use-case speci ica ions used o
alida ion will be desc ibed, highligh ing he i ual scena io ha is going o be implemen ed in he
simula ion.
P oposed HESS Hyb idiza ion A chi ec u e
Based on he use-case speci ica ions, which will be desc ibed a he end o his sec ion, his s udy
p oposes he hyb idiza ion a chi ec u e shown in Figu e 5. I consis s o TES ed by an ai -sou ce HP.
The HP comp esso is powe ed by a combina ion o a ba e y ene gy s o age sys em (BESS) and
pho o ol aic (PV) sys em in addi ion o he low- ol age (LV) g id. Two main con ollable asse s a e
adop ed o implemen he op imisa ion-based con ol o HESS, which a e HP and BESS. Fo he i s
one, a a iable speed comp esso is needed o adjus con inuously he e ige an low, and
consequen ly, he HP sink empe a u e le el. Fo he second one, a DC- o-AC in e e /cha ge is
needed o in e ace BESS wi h he local AC bus and con ol he cha ging/discha ging a es. Howe e ,
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 24
his s udy does no include any op imal sizing o hese sub-sys ems, and he ela ed in es men cos
is excluded om he inal e alua ion.
Figu e 5 Simple o e iew o he connec ion opology o HESS in he cen alized hea ing subs a ion.
3.3.1. Sys em modelling
3.3.1.1. Hea pump and ene gy s o age sys em.
Fo his ask, he same ma hema ical app oaches explained in subsec ions 3.1.3.1 o HP and
TES, a e adop ed.
3.3.1.2. s a e o cha ge modelling o he ba e y ene gy s o age sys em
BESS SOC le el o he nex ime in e al is es ima ed ia he ollowing equa ion:
.
( ) ( ) 100. B
IT
SOC T SOC C
+ = −
(21)
()SOC
is he measu ed alue [%] while
()SOC T+
is he p edic ed one a he ime ho izon
T
;
C
is he BESS a ed capaci y [Ah] and
B
I
is he unning cu en [A] which is calcula ed hanks o
he nex simple equa ion:
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 25
B
B
B
P
IV
=
(22)
B
V
is he BESS ol age [72 V], o simplici y i is conside ed cons an ( h ee 24V uni s in se ies);
B
P
is he exchanged BESS powe [W], i akes posi i e alue in case o cha ging and nega i e alues in
case o discha ging. Fo simplici y, BESS sel -discha ge p ocess is no modelled.
3.3.1.3. Nonlinea op imisa ion
In his s udy, Sequen ial Quad a ic P og amming (SQP) is chosen as an op imize o he op imisa ion
p oblem, i is an app oach ha s a s wi h an ini ial guess o he decision a iables
x
. A each ime
s ep, i o ms a quad a ic app oxima ion o he Lag angian, which is he objec i e unc ion
() x
combined wi h a weigh ed sum o he cons ain unc ions
()hx
and
()gx
, and ind he op imal
solu ion in an i e a i e way (Boggs, P. and Tolle, J, 1995).
1
min ( ) 2
. . ( ) 0
( ) 0
TT
x x Qx F x
s g x
hx
=+
=
(23)
( , ) ( ) ( ) ( )
TT
L x x h x g x
= − −
(24)
Q
and
F
a e espec i ely he quad a ic cos ma ix and he linea cos ec o ;
is a ec o o
Lag angian mul iplie s o he equali y cons ain s;
is a ec o o Lag angian mul iplie s o he
inequali y cons ain s.
The p oblem is hen shi ed om minimizing
() x
o minimizing
( , , )L x
which holds he
objec i e unc ion along wi h he cons ain s. The op imisa ion algo i hm i e a i ely upda es he
decision a iables
x
, Lag ange mul iplie s
and
un il con e gence. The goal is o ind an x ha
minimizes he Lag angian subjec o he cons ain s.
The objec i e unc ion adop ed in his s udy is o mula ed as ollow:
,,,
22
,2
,,
.( )
( ) ( )
()
HP i
i B i PV i i TH
i
MAX i MAX MIN i MIN
HP i B i PV i i TH
i
Q
EP P P i EP EP
COP
J T T T T
QP P i EP EP
COP
− −
= + − + −
− −
(25)
Subjec o he ha d cons ain o mula ed below:
,.
100. Bi
MIN i MAX
IT
SOC SOC SOC
C
−
(26)
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 32
The da ase s ex ac ed om Clus e 2 o schools, gene a ed by bo h clus e ing me hods, show
ela i ely less modelling capabili y, which is ansla ed o he a e age p edic ion pe o mance
summa ized in Figu e 11.
Figu e 10 Annual p edic ion assessmen using a school da ase ex ac ed om Clus e 1 (a) using Euclidean k-means (b)
using DTW k-means.
Figu e 11 Annual p edic ion assessmen using a school da ase ex ac ed om Clus e 2 (a) using Euclidean k-means (b)
using DTW k-means.
Mainly due o he di e en occupan s' ac i i y schedules, he hea load annual p o ile o nu sing
homes is less luc ua ing, as Figu e 12 shows, compa ed o schools. This is because DH load p o iles
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 33
o schools a e much sha pe , wi h dis inc nigh se backs and weekday/weekend shi s han hose
o nu sing homes. Mo eo e , hea loads in summe o schools a e mode a ely lowe han hose o
nu sing homes, which is highligh ed by he black dashed ec angles in Figu e 10 and Figu e 11. The
eason is ha no o icial scheduled ac i i ies a e ca ied ou du ing he summe in schools, which is
no he case o nu sing homes.
Clus e 2, issued by he wo s udied clus e ing echniques o nu sing home da a, shows di e en
pa e ns compa ed o Clus e 1 (see Figu e 13). In his case, he annual hea load o he
co esponding buildings is conside ably lowe . Low SH demands o hese buildings can ep esen one
po en ial jus i ica ion. Mo eo e , Clus e 2 also includes some da a anomalies, such as he ou lie
alue in Figu e 13 (a) and he measu emen gap in Figu e 13 (b), which a e bo h highligh ed by black
dashed ec angles.
Figu e 12 Annual p edic ion assessmen using a nu sing home da ase ex ac ed om Clus e 1 (a) using Euclidean k-
means (b) using DTW k-means.
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 34
Figu e 13 Annual p edic ion assessmen using a nu sing home da ase ex ac ed om Clus e 2 (a) using Euclidean k-
means (b) using DTW k-means.
4.2.2. Seasonal assessmen
To gain a deep insigh in o he p edic ion p ocess in di e en seasonal pe iods, h ee-week scena ios
ha e been selec ed: cold, wa m, and mild. Below, mo e explana ions, along wi h some g aphs, a e
p esen ed o his con ex . Da ase s, chosen a bi a ily om Clus e 1, gene a ed by DTW k-means
o each building ype, ha e been selec ed.
In Figu e 14, he ANN model shows a high p edic ion capabili y o schools in he win e compa ed o
he summe scena io. The eason is ha in win e , hea load peaks appea clea ly du ing wo king
days daily, while in summe , his load pa e n is negligible. The hea demand is subs an ially lowe
du ing he weekend in win e , while in summe his di e ence is no signi ican . This ma e is
highligh ed by he black dashed ec angles in Figu e 14 (a) and Figu e 14 (b). Peak ene gy demands
in win e a e linked mainly wi h he need o SH and he no iceable occupancy a e. In he summe ,
SH demands a e insigni ican , and occupancy is almos null due o he annual aca ion.
Conce ning nu sing homes, he ANN model shows clea ly low p edic ion capabili ies, especially in
summe (see Figu e 15 (b)), as in his case, he ou doo empe a u e ac o is less dominan o DH
load (SH demand is signi ican ly low). In his pa icula case, he DHW load ep esen s he main ene gy
demand po ion, which is ela ed o ac o s o he han wea he da a. This e ec is less dominan in
he case o schools due o he impac o o he impo an aspec s, mainly ela ed o he occupan 's
ac i i ies.
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 35
Figu e 14 P edic ion pe o mance o a school da ase (a) in win e (b) in summe .
Figu e 15 P edic ion pe o mance o a nu sing home da ase (a) in win e (b) in summe .
Fo load p edic ion assessmen in mild seasons, he ac ha ou doo empe a u e is no longe a
dominan pa ame e o de e mining he hea load leads o inaccu a e p edic ions o nu sing homes
(see Figu e 16 (a)), howe e , ANN is s ill showing ela i ely be e esul s when i comes o school hea
load p edic ion.
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 36
Figu e 16 P edic ion pe o mance in mild-season (a) o a nu sing home (b) o a school
P edic ion pe o mance e alua ion
Table 3 lis s mean squa ed e o (MSE) and mean absolu e e o (MAE) alues o he p edic ion
pe o mance o he p oposed ANN eg ession me hod o all clus e s, compa ed o a e e ence me hod
de eloped by (T. O Timoudas e al., 2022). The au ho s he e de eloped an ANN app oach ha includes,
along wi h he his o ical ou doo empe a u es, and his o ical hea loads wi h di e en ime leng hs.
I has been ound ha be e esul s we e eached using his o ical alues o ou doo empe a u e and
load up o 24 hou s. This app oach has been alida ed on he same da ase s used in his a icle, ei he
o aining o es ing. Be e esul s ha e gene ally been ob ained using Clus e 2 da ase s using bo h
clus e ing echniques (Euclidean and DTW k-means). One jus i ica ion o ha is ha Clus e 2
gene ally ga he s less luc ua ing load p o iles, as shown in Figu e 13. This makes i easie o MLP o
ack smoo he load cu es appea ing in he win e as hey e lec SH demands. Such load cu es a e
highly dependen on wea he da a, mainly, ou doo empe a u es, which explains he high p edic ion
pe o mance in his case. Figu e 12 shows mo e luc ua ing cu es, which can be a ec ed by o he
addi ional ac o s, such as occupancy change a es. Fo all clus e s, he able shows clea ly ha he
compa ison is in a ou o he MLP model, p esen ed in his s udy, compa ed o he ANN baseline
me hod.
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 37
Table 3 MSE and MAE e alua ion using nu sing homes no malized da a o he yea 2019.
E alua ion
c i e ia
P oposed MLP
ANN p oposed in [10]
MSE
Clus e 1 (Euc k-means)
0.0153
0.0219
Clus e 1 (DTW k-means)
0.0154
Clus e 2 (Euc k-means)
0.0073
Clus e 2 (DTW k-means)
0.0072
MAE
Clus e 1 (Euc k-means)
0.0975
0.1106
Clus e 1 (DTW k-means)
0.0977
Clus e 2 (Euc k-means)
0.0639
Clus e 2 (DTW k-means)
0.0638
4.3. Con inen al/Alpine Clima e – op imal demand esponse managemen in
DHN a a communi y le el
4.3.1. Use case desc ip ion.
The op imisa ion-based demand side managemen (DSM) policy de eloped in his s udy is alida ed
on da a gene a ed by sma me e s which a e ins alled in he Sonnenpla z li e-in lab in G oßschönau,
Aus ia. I consis s o a local ene gy communi y powe ed by LV and local DH ne wo ks (see Figu e 17).
The use s connec ed o he ne wo k consis o h ee public buildings, wi h an o e all hea ed su ace
o 2000 m2 and an es ima ed a e age powe consump ion o 150 kW: wo comme cial buildings wi h
an es ima ed powe consump ion o 170 kW, and ou esiden ial buildings wi h an es ima ed powe
consump ion o 24 kW. The gene a ion sou ces include one o he exis ing biomass boile s wi h a
capaci y o 130 kW, connec ed o a 10 m3 wa e ank, and he p oposed sola sys em, wi h an o e all
capaci y o 250 kW.
This s udy assumes ha he DH ne wo k is connec ed o each building h ough he subs a ion
opology shown in Figu e 4. HP plays he ole o a empe a u e boos e ha upg ades he DH supply
empe a u e om ~55°C o ~70°C. I is also conside ed a connec ion poin o he elec ic g id ( ia he
comp esso ope a ion) wi h he he mal ne wo ks/asse s. In his s udy, he op imisa ion is ca ied ou
based on he lowcha p esen ed in Figu e 18. Because compu ing ime inc eases exponen ially wi h
he numbe o in ol ed buildings, in his s udy, he simula ion o he DSM s a egy is limi ed o ou
buildings, labelled con en ionally as ollows: Building 4.5, Building 6.0, Building 7.0 wi h 10.5 kWp PV,
and Building 8.0 wi h 11 kWp PV.
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 38
Figu e 17 Sa elli e iew o G oßschönau communi y wi h buildings conside ed o op imal ene gy coo dina ion.
Figu e 18 Ene gy op imisa ion lowcha .
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 39
Table 4 Simula ion pa ame e s
Pa ame e
Value
Elec ici y buying p ice
0.45 €/kWh
Elec ici y selling p ice
0.15 €/kWh
TES olume
500 li e s
TES inle wa e low
0.018 kg/s
TES ou le wa e low
0.013 kg/s
TES max SOC
90 %
TES min SOC
10 %
DH supply emp
55 °C
HP condense min emp
50 °C
HP condense max emp
70 °C
PV powe cabling loss
3 %
A e age CO2 emissions
127 g/kWh
Table 5 PSO pa ame e s
Pa ame e
Value
Numbe o pa icles
80
Decision a iables
4
Maximum i e a ions
50
Ine ia weigh
w
0.5
Cogni i e coe icien
1
c
0.6
Social coe icien
2
c
1.6
4.3.2. Simula ion esul s
In his sec ion, a simula ion s udy is ca ied ou o he assessmen o he de eloped op imisa ion-
based DSM. The simula ion spans a wo-day scena io using IoT da a gene a ed on May 30-31, 2023.
Da a o he hea load o he ou buildings, conce ned o simula ion, wi h he PV powe p oduc ion
o wo o hem, a e displayed in Figu e 19. The main pa ame e s used in he simula ion a e summa ized
in Table 4 Fo be e con ol pe o mance assessmen , he s udy assumes he p esence o unde -sized
TES uni s a each ele an building's subs a ion, speci ically u ilizing 200-li e TES. The op imisa ion-
based DSM changes, a each ime s ep (one hou ), all HP condense empe a u e se poin s. The main
PSO op imisa ion pa ame e s a e lis ed in Table 5. The simula ion asses he op imisa ion ou come o
wo cases wi h wo objec i e unc ions weighing modes:
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 40
Figu e 19 Hea demand wi h PV powe p oduc ion o he conce ned buildings.
Case 1 (use -o ien ed op imisa ion): in his case, only he i s e m o he objec i e unc ion is
conside ed. The e o e, ene gy op imisa ion is in a o o pee - o-pee o pee - o-u ili y ading. The
goal is o educe he ene gy cos o each building indi idually by d i ing di e en HP uni s op imally.
The HP empe a u e se poin , as well as he SOC e olu ion o each TES, a e displayed in Figu e 20. The
simula ion showed an ene gy cos o 0.00 €, 0.00 €, -3.95 €, and -3.80 € o Building 4.5, Building 6.0,
Building 7.0 and Building 8.0 espec i ely. Posi i e alues e e o ene gy buying p ices while nega i e
ones e e o selling p ices. The o al exchanged ene gy wi h he g id u ili y was –51.76 kWh, simila ly,
he nega i e sign e e s o an ene gy expo ed o he g id u ili y.
Figu e 20 HP op imal condense empe a u e se poin wi h associa ed TES SOC a ia ion in a 2-day simula ion (use -cen ic
scena io)
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 41
Case 2 (communi y-o ien ed op imisa ion): in his case, only he second e m o he objec i e
unc ion is conside ed. The goal is o maximize he ene ge ic independence o he conce ned buildings.
The me hod is based on sha ing op imally he PV ene gy be ween buildings o gi e highe lexibili y o
he HP uni s o wo k in op imal coo dina ion and independen ly om he g id u ili y. The HP
empe a u e se poin , as well as he SOC e olu ion o each TES, a e displayed in Figu e 21. The
simula ion showed an ene gy cos o 6.24 €, 3.57 €, -2.89 € and –3.22 € o Building 4.5, Building 6.0,
Building 7.0 and Building 8.0 espec i ely. The o al exchanged ene gy be ween he communi y and
he g id u ili y was –19.63 kWh.
Figu e 21 HP op imal condense empe a u e se poin wi h associa ed TES SOC a ia ion in a 2-day simula ion
(communi y-cen ic scena io)
The simula ion esul s e eal ha in Case 1, he cos -e ec i e HP ope a ion is e iden o each
end-use . In ac , in Building 4.5 and Building 6.0, HP uni s did no expe ience any signi ican wo k.
Though, a la ge unused su plus o PV ene gy was expo ed o he g id u ili y. In con as , Case 2
add esses his su plus h ough he op imal coo dina ion o all in ol ed HP uni s. Howe e , Ca bon
emission e alua ion shows ha , in Case 1, a o al o 0.39 kg CO2 emissions ela ed o he HPs ope a ion
is calcula ed, while in Case 2, 3.58 kg was calcula ed. S iking a balance be ween use -cen ic and
communi y-o ien ed op imisa ions is essen ial and ailo ed o each use-case speci ica ion, which is
d i en by economic and ecological a ge s.
4.4. Medi e anean Clima e – op imal con ol o hea pump wi h a
pho o ol aic/s o age sys em in cen alized DHN subs a ion
4.4.1. Use-case desc ip ion
The Li e-in Lab ENVIPARK consis s o en buildings, ou o which a e mainly used as labo a o ies, i e
as o ices, and one as a can een and es au an . The DHN is supplied by wo hea exchange s wi h 1.2
MW each wi h a supply empe a u e a ound 75°C du ing he win e season, and o domes ic hea
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 48
in Figu e 24 (c) and he elec ic powe impo ed om he g id o each scena io. The able shows how
signi ican is he impac o ex ending he ope a ion lexibili y bounda ies gi en o each ESS echnology
(TES and BESS) in he pe o mance o he hea ing cos op imisa ion. Adjus ing he a ia ion in e al o
he TES maximum wa e empe a u e om 10°C o 15°C and BESS SOC a ia ion in e al om 40% o
80% has led o a cos sa ing o up o 12.4% using SQP chosen in his a icle, and 13.9% using pa e n
sea ch algo i hm. Howe e , ex ended lexibili y limi s ha e led o mo e compu ing ime, which is
explained by he p olonged sea ch space o each algo i hm. I he ene gy cos op imisa ion p ocess is
a pa o eal- ime con ol, his issue is c ucial and should be conside ed.
Table 7 Flexibili y impac assessmen o di e en op imisa ion me hods.
Op imisa ion algo i hm
Hea ing cos
Execu ion ime
Cos sa ing
(Baseline
Scena io 1)
Scena io 1
Scena io 2
Scena io 1
Scena io 2
SQP
1466 €
1284 €
8 min
12 min
12.4%
In e io -poin
(F. Capi anescu e al.,
2007)
1456 €
1262 €
22 min
28 min
13.3%
Ac i e-se
(Zhong e al., 2023)
1468 €
1290 €
6 min
7 min
12.1%
Pa e n sea ch
(Tao e al., 2021)
1434 €
1234 €
1.7 hou
1.6 hou
13.9%
Conclusions
Due o he wide a ailabili y o da a om sma me e s, senso s, and building managemen sys ems
in DHN-powe ed buildings, da a-d i en app oaches, o op imizing ene gy usage a building and
communi y le els, ep esen powe ul ools. Such da a can help in unco e ing end-use beha iou
enabling by ha he de elopmen o ad anced con ol and managemen policies. In his con ex , his
deli e able shows he de elopmen and alida ion s eps o se e al da a-d i en echniques o e ing
clus e ing and p edic ion unc ions. Combining p edic ion models wi h me aheu is ic and
de e minis ic op imisa ion algo i hms, his deli e able shows also op imal con ol and demand
managemen ha ake ad an age o a ailable ene gy lexibili y in DHN ene gy communi ies, ei he
as a dis ibu ed con ol solu ion hanks o ad anced IoT communica ion, o as a cen alized one ha
elies on agg ega ed da a in he main DHN subs a ion.
Summa y o achie emen s
Based on he conclusions desc ibed abo e, he main achie emen s o he wo ks p esen ed in his
deli e able a e lis ed below:
• Acco ding o each use case speci ica ions, di e en da a collec ion solu ions ha e been
adop ed o be e da a analy ics. This includes wo king wi h classical iles-based da a (.xls o
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 49
.cs ), ela ional and ime se ies da abases ( h ough SQL and API que ying), o h ough IoT
communica ion.
• Ad anced ML da a mining echniques may ep esen powe ul ools o deal wi h low-quali y
and noisy da a, ob ained usually om empo al mal unc ioning o senso s o in e ne
disconnec ion. This can also iden i y uncommon DH consump ion pa e ns and abno mal use
beha iou , which imp o es signi ican ly he pe o mance p edic ion models ha a e buil in
clus e -wise app oach.
• Le e aging DL o long- e m p edic ion o DH load showed adequa e o good pe o mances
in di e en scena ios.
• By aking ad an age o p edic ion models wi h me aheu is ic and de e minis ic op imisa ions,
i is possible o build ad anced model-based con ol and managemen policies ha op imize
ene gy usage in DHN communi ies.
• When disagg ega ed da a in DHN communi ies a e no a ailable, agg ega ed one can be
u ilized o enable cen alized managemen based on he op imal exploi a ion o he ene gy
lexibili y in he main subs a ion.
Rela ion o con inued de elopmen s.
Following he las poin lis ed abo e, he e is s ill an ongoing ac i i y ha consis s o disagg ega ing
powe consump ion da a a a building le el using ML-based non-ins uc i e load moni o ing (ML-
NILM). This echnique can disco e he ope a ion s a e o di e en HVAC ins alla ions by analyzing
hei elec ic powe consump ion indica o s (ac i e and eac i e ene gies, powe ac o , phase
cu en ).
This echnique has he po en ial o assess powe and hea lows in he building and up o he
communi y le el, which enables he implemen a ion o ene gy managemen s a egies applicable o
sec o coupling.
D3.3 Me hodologies and amewo ks o op imize demand esponse and peak load sha ing o 4 h-
5 h DHC in h ee ypical Eu opean clima es. 50
Re e ences
A. Ja ed, B.S. Lee and D.M. Rizzo, “A benchma k s udy on ime se ies clus e ing”, Machine Lea ning
wi h Applica ions, 1, 2020, 100001.
Boggs, P. and Tolle, J, 1995, Sequen ial Quad a ic P og amming. Ac a Nume ica, 4, 1-51.
J. Kennedy and R. Ebe ha (1995). Pa icle Swa m Op imisa ion. P oceedings o he 1995 IEEE
In e na ional Con e ence on Neu al Ne wo ks (Pe h, Aus alia): IEEE Se ice Cen e,
Pisca away, NJ, IV: pp 1942-1948.
M. Jespe , F. Pag, K. Vajen and U. Jo dan, “Annual Indus ial and Comme cial Hea Load P o iles:
Modelling Based on k-Means Clus e ing and Reg ession Analysis”, Ene gy Con e sion and
Managemen : X, Volume 10, 2021, 100085.
M.M. Alam, H. Bao, P.X.W. Zou and J. Sanjaya, “Beha iou Change o Building Use s and Ene gy
Consump ion”, Encyclopedia o Sus ainable Technologies 2017, Pages 189-196.
O.I. Abiodun, A. Jan an, A.E. Omola a, K.V. Dada, N.A.E. Mohamed and H. A shad. S a e-o - he-a in
a i icial neu al ne wo k applica ions: A su ey. Heliyon, Volume 4, 2018, 00938.
S. Aghabozo gi, AS Shi kho shidi and TY Wah, Time-se ies clus e ing–a decade e iew. In Sys 2015;
53:16–38.
Thomas OhlsonTimoudas, Yiyu Ding, Qian Wang. A no el machine lea ning app oach o p edic sho -
e m ene gy load o u u e low- empe a u e dis ic hea ing. REHVA 14 h HVAC Wo ld
Cong ess 22 – 25 May, Ro e dam, The Ne he lands
Zheng Zhang, Ping Tang, Lianzhi Huo, and Zengguang Zhou. MODIS NDVI ime se ies clus e ing unde
dynamic ime wa ping. In e na ional Jou nal o Wa ele s, Mul i esolu ion and In o ma ion
P ocessing HYPERLINK "h ps://www.wo ldscien i ic.com/ oc/ijwmip/12/05". (2014) .
Flo in Capi anescu, Me ludin Gla ic, Damien E ns , Louis Wehenkel, 2007, In e io -poin based
algo i hms o he solu ion o op imal powe low p oblems. Elec ic Powe Sys ems Resea ch 77,
508–517
H. Zhong, S. Meng, X. Zhang, Z. Wei, C. Zhang and L. Du, 2023, "Mul iphysics-Cons ained Fas Cha ging
o Li hium-Ion Ba e y Wi h Ac i e Se P edic i e Con ol," in IEEE T ansac ions on In elligen
T anspo a ion Sys ems
Hai Tao, Fa aedoon Waly Ahmed, Halkaw Abdalqadi kh ahmed, Mohsen La i i, Hi oki Nakamu a,
Ya eng Li, 2021, Hyb id whale op imiza ion and pa e n sea ch algo i hm o day-ahead ope a ion
o a mic og id in he p esence o elec ic ehicles and enewable ene gies, Jou nal o Cleane
P oduc ion 308, 127215
