Player is loading...
Dr. Marco Tognoli - Fuel-saving potential simulation and estimation when optimally predicting the supply set-point conditions of a District Heating Network with detailed system data-driven modelling
Dr. Marco Tognoli, Politecnico di Milano
Monday, Nov. 15, 2021 · 11:31 a.m. · 12m 55s
The reduction of primary energy consumption, as an energy efficiency measure, is required through the application of system interventions following the principles issued by the European Commission and the goal of minimizing the greenhouse emissions as stated by the Paris Agreement (2015). The aim of the present work is to first provide a reliable estimation tool, developed through a set of Machine Learning models, the dynamic thermal behaviour of a District Heating Network located at a university campus in Italy (Politecnico di Milano). In the next step, the advantages in term of fuel saving potential and heat supply quality are highlighted through the implementation of a thermal predictive demand control logic. The developed solution is readily appliable to District Heating Systems allowing to reduce the undesired thermal losses.
Embed
Transcriptions
Note: this content has been automatically generated.
00:00:00
the shuns just before lunch break break let's look ugly she's very pollutants workshops
00:00:07
to like oh actually should use work to to leave our next presenter
00:00:13
it's from a critical looks real this time from you'll do
00:00:19
and he's going to two girls some of juicy digits you should just should well at
00:00:25
least the floor i like i like long titles i mean i think it's a defect
00:00:31
so low riding my marketing only i'm a p. h. d. students and
00:00:35
currently a terminating my p. h. d. studies and today i'm going to
00:00:40
present yet again a work based on anticipated network and particularly on an integrated framework
00:00:46
modelling either the the man side and the supply side
00:00:51
so and this is motivated by also the recent news that we're not by um the
00:00:58
yeah i'm the goals all the first reason for the prize
00:01:01
agreement and also the recent ones all the climate uh change
00:01:05
conference value button at the station and and their and their
00:01:08
goals are we reducing the footprint and particularly also that the primary
00:01:13
uh any intrusion some primary energy savings in
00:01:16
those a critical system so while the sheep while
00:01:20
the the men's side is still fixed on front in some in some scenarios lots of potential
00:01:26
can be a a highlighted by the generation sites and particularly in the context of
00:01:31
the in the nasa and they will i will talk about the presidential in commercial applications
00:01:37
the um the uh far too boilers are generally
00:01:41
a simple devices which are integrated in those system
00:01:44
in order to provide the the man and generally in this case we're talking about hot water boilers
00:01:49
and um the the back on the uh on on the topic adam was
00:01:54
describing regarding family engine you the savings and their energy savings intervention these are generally
00:02:00
i have two main drawbacks of which are we deal those those the intermission are traditionally made
00:02:06
which are that they they need a huge capital
00:02:09
dimensions and also technical difficulties and doing those interventions
00:02:14
so and normal approach in this sense is represented by
00:02:18
implementing much optimised control strategies particularly in the supply side
00:02:23
uh which is bounded to the production of the
00:02:27
heap demands all these uh is the this rating system
00:02:31
and this is in contrast with the current control strategies which and
00:02:35
we see on the on the last part of all the screener representation
00:02:39
all they get a medical control strategy in which these are
00:02:44
not a add the those not do not
00:02:47
describe the um dynamic affects all of this system
00:02:51
still i am in in this in this study it deeply technically no
00:02:56
no we're not to compose as been selected as the the city can network
00:03:00
and this is controls is a is a medium size
00:03:03
is eating network and i also um and um this compose
00:03:07
all twenty two separate buildings which are connected by two distribution
00:03:12
looks and with as we can see on the left part
00:03:15
uh on on right part of the screen and um and within a a a single power plant
00:03:21
and um the whole system is to monitor a
00:03:25
particularly with respect to the term all ah property the
00:03:29
the the firm of variables but also we have additional
00:03:32
information regarding the environmental a sore radiance and electrical roles
00:03:37
and uh the key data on this uh on all these case studies that's
00:03:42
a d. a. given the medium to large size of this system which covers
00:03:48
more than one hundred thousand square meters of uh it it surfaces
00:03:52
and we have enough information to draw the thermal energy balance all beach subsystem separately
00:03:58
and this isn't how fuel cost of the uh of the current says this is not because given the
00:04:04
average price we're we're we're we're not mentioning the the the actual uh prices
00:04:09
uh or natural gas is not around all one point six
00:04:13
million euro per year um so that's um within this information
00:04:20
we can then defined the two main objectives in the system so basically on this preliminary get analysis
00:04:26
so we have a potential in with this in the thermal says of the distribution network
00:04:31
what also is the second objectives as being highlighted that there
00:04:34
are excessive thermal generation events which are caused by a mismatch
00:04:39
within the the current uh uh control strategy between the demand side and the supply
00:04:45
so um this two main activities which are then first
00:04:49
one related to the modelling of the each subsistence separately
00:04:53
then can be combined together or this all this models in
00:04:57
order to provides a a simulation environment for the optimisation phase
00:05:02
so all i disagree modelling approaches used
00:05:05
since we have nor a uh information regarding
00:05:09
the geometry and technical data present we just have the information of the primary and
00:05:14
secondary loop all of uh of the system while uh this mythology allows to also
00:05:20
include the dynamic there might be a better so in contrast with the tradition control strategy
00:05:26
and we'll also investigates the effects of the lag in the system
00:05:31
uh which is called by the the injection the did did depend i'm
00:05:37
between the injection point and the extraction point in the in in
00:05:40
in the system and given the limited control rise on all the optimal
00:05:45
controller a short term uh models ranging from uh uh
00:05:50
fifty minutes after one hour ahead will be will be considered
00:05:54
so uh in this scheme though we're all a first part of the only
00:05:59
investigation can we can can be seen particularly in the orange box we see
00:06:03
the uh development all the data driven pretty models all the thermal performances on each single
00:06:09
a subsystem uh so ranging from the uh from from the secondary side of the
00:06:16
uh all the buildings down to the generation side so we model basically all the elements present
00:06:22
the from the uh buildings distribution side and degeneration side
00:06:27
and the second dimension of the all of investigation is that different machine learning models are included
00:06:34
in order to then estimates the resulting uh uh accuracy of this of this uh models
00:06:40
and finally the third in a dimension of investigation is
00:06:44
represented by the production rises so we have different models within
00:06:47
a viable pretty tries in range from fifteen minutes up to one or at and
00:06:52
and um another important step is represented by the feature selection so after the training phase
00:06:57
uh all the separate models which are then also
00:07:01
a four c. d. implementation all they the uh genetic
00:07:05
the programming base the pipeline optimisation algorithm
00:07:10
a it to the feature selection serves as as two main objective the first one
00:07:14
is the reduction of the future so based on the estimation of the future importance
00:07:20
with respect to the dependent variable and also gaining
00:07:24
some respectability so today we we we talk about
00:07:27
lie in making the models general as a ball and this is an
00:07:31
important step in order to achieve partial it at least partially this cool
00:07:35
um so that after that when the when the set of feature to just importance of the features selected
00:07:41
in order to train faces a is then is then included
00:07:47
these models uh it is worth i think that they can be combined together in order to provide overall estimation of the system
00:07:53
and these key results which are based on the uh training phase that would be implementing
00:07:58
a chronological approach or uh either from the training and testing and also the validation phase
00:08:04
for c. d. ah it freezes for the distribution side uh that's a given the
00:08:09
limited time frame there is no relevant no much dependence over the external features so that's
00:08:15
to says those phenomena as an much more wider a time rise and
00:08:20
while there is a in inaudible importance by the supply temperature conditions and
00:08:25
the achieved the school or are always know where within for windows methodology
00:08:30
the word then six point five percent is the is the mean absolute relative
00:08:34
maybe shun and i'm a fall when the the the um different the feature selections
00:08:39
that so after selecting set of features we see that also there is a remarkable reduction
00:08:44
of uh of the models are are then an improvement of the production capabilities of the models
00:08:50
uh would compare to that to to the previous slide even in the worst case scenario and as predicted the
00:08:56
a higher d. of production horizon higher would be the ever
00:09:01
on the wall so ranging from fifteen minutes up until sixty minutes
00:09:05
and the moon so uh uh regarding the estimation all the time like introduced by the by
00:09:12
the dimension of the distribution network in the system so between the power plant and the furthest
00:09:19
a building in the system there is a fifteen minute
00:09:22
time languages measure by the difference in step between two
00:09:26
uh injection moments all of the all of the of the temperature measured and that can be
00:09:31
highlighted in this in this scheme by the uh it's a super position of the two peaks
00:09:36
and finally all these models are then combined together in order to first within the
00:09:42
granularity but also to provide the overall estimation of the three more performances of the system
00:09:46
uh such that they can be um to serve as the
00:09:51
control basis all the the constrain not optimisation uh we can see here
00:09:56
willow um which is subject to the bit minimisation all the return temperature
00:10:02
two words the power plants so this allows it does a reduction of
00:10:05
the thermal losses on distribution network while it is constrained by e. d.
00:10:10
thermal demanded predicted term of the mind of the system soul
00:10:14
uh in order not to the in record any any discomfort
00:10:19
and the the printer is also that has been also extended to a wider set of buildings
00:10:24
have a highlight that there is a potential of a reduction of these
00:10:30
um all the fuel costs by four percent which
00:10:34
in this scenario even the fuel costs of the system
00:10:38
equates to nearly sixty five thousand euros so it
00:10:41
is a it is a notable um um energy saving
00:10:45
that's uh it is also worth fighting again that it would be it requires a really a marginal investment cost
00:10:52
and also a d. v. d. information times are are really
00:10:56
uh uh a low were compared to traditional energy savings intervention
00:11:01
and in order does i'm going forward to the to the conclusion part since my time's almost up
00:11:07
and that's they saw it here in this in this investigation
00:11:11
the they probably did it it has been propose a framework which underlies framework for
00:11:18
modelling the overall basically the network would while
00:11:22
eh um while still retaining wall didn't know the tales which can then be used also
00:11:27
for for the start this which of course are not are not the subject of this
00:11:32
well all this presentation and um and it's also worth it i think that's not only
00:11:37
the thermo uh uh uh uh on a data argues but also data sourcing from the fields
00:11:43
uh ranging had again from the electrical and external olds i'm still and um
00:11:50
comparing it different type of and machine learning architecture allows to select the most promising technology
00:11:56
uh and the eight implementation on the supply uh on the
00:12:01
top of my supply a control strategy allowed a reduction all over
00:12:06
nearly sixty five thousand euro per year which is a um like for wanting us in this
00:12:11
in this sense a given the the the committed on a investment costs of this all this activity
00:12:18
and the start the as to be concluded and all somebody
00:12:22
they sit on a white versus the us a set of uh
00:12:26
'kay start this so this is faced with collaboration we did we did your research institute
00:12:30
ah so that in order to provide a more
00:12:33
generalised will approach of this uh all this mythological framework
00:12:37
while uh uh and it's it's the analysis is going through
00:12:41
a is going to be investigated with respect to the control rise
00:12:44
on on the uh optimised controller so this is some not all
Conference Program
Workshop introduction by Jérôme Kämpf & Henning Müller + Keynote speech Pr. Dr. Guglielmina Mutani - Place-based energy models for sustainable cities
Pr. Dr. Guglielmina Mutani, Politecnico di Torino
Nov. 15, 2021 · 9 a.m.
333 views
Q&A (Keynote speech: Pr. Dr. Guglielmina Mutani)
Pr. Dr. Guglielmina Mutani, Politecnico di Torino
Nov. 15, 2021 · 9:49 a.m.
Roberto Boghetti - Simulating District Heating Networks’ hydraulics with residual physics neural networks
Roberto Boghetti, Idiap Research Institute
Nov. 15, 2021 · 10 a.m.
Q&A (Roberto Boghetti)
Roberto Boghetti, Idiap Research Institute
Nov. 15, 2021 · 10:13 a.m.
Amine Weibel - Machine learning forecast for photovoltaic power stations (PV) with spatial interpolation on the COSMO-7 model. A real ‘use-case’ on the Czech Republic
Amine Weibel, HES-SO
Nov. 15, 2021 · 10:16 a.m.
Q&A (Amine Weibel, HES-SO Valais-Wallis)
Amine Weibel, HES-SO
Nov. 15, 2021 · 10:27 a.m.
101 views
Dr. Giuseppe Peronato - HeatLearn: a convolutional neural network to assess the building energy consumption at regional scale from open land-use maps
Dr. Giuseppe Peronato, Idiap Research Institute
Nov. 15, 2021 · 10:30 a.m.
Q&A (Dr. Giuseppe Peronato)
Dr. Giuseppe Peronato, Idiap Research Institute
Nov. 15, 2021 · 10:43 a.m.
Pr. Dominique Genoud & Jérôme Treboux - Integrating machine learning and AI in practical energy applications : a review of these last 10 years activities at the Institute of Information Systems
Pr. Dominique Genoud & Jérôme Treboux, HES-SO Valais-Wallis
Nov. 15, 2021 · 10:46 a.m.
Q&A (Pr. Dominique Genoud & Jérôme Treboux)
Pr. Dominique Genoud & Jérôme Treboux, HES-SO Valais-Wallis
Nov. 15, 2021 · 10:59 a.m.
Dr. Marco Tognoli - Fuel-saving potential simulation and estimation when optimally predicting the supply set-point conditions of a District Heating Network with detailed system data-driven modelling
Dr. Marco Tognoli, Politecnico di Milano
Nov. 15, 2021 · 11:31 a.m.
Q&A (Dr. Marco Tognoli)
Dr. Marco Tognoli, Politecnico di Milano
Nov. 15, 2021 · 11:44 a.m.
Pr. David Wannier & Jean-Marie Allder - Research advance in Energy prediction for electrical vehicles and green mobility hotels, the results of the eVIP project
Pr. David Wannier & Jean-Marie Allder, HES-SO Valais-Wallis
Nov. 15, 2021 · 11:46 a.m.
Q&A (Pr. David Wannier & Jean-Marie Allder)
Pr. David Wannier & Jean-Marie Allder, HES-SO Valais-Wallis
Nov. 15, 2021 · 11:58 a.m.
Dr. Kavan Javanroodi - Extending the concept of energy hub to facilitate sector and spatial coupling
Dr. Kavan Javanroodi, Solar Energy and Building Physics Laboratory (LESO-PB) at EPFL
Nov. 15, 2021 · 11:59 a.m.
143 views
Q&A (Dr. Kavan Javanroodi)
Dr. Kavan Javanroodi, Solar Energy and Building Physics Laboratory (LESO-PB) at EPFL
Nov. 15, 2021 · 12:16 p.m.
Pr. Pierre Roduit - Prediction of the electricity flexibility of households using Machine Learning
Pr. Pierre Roduit, HES-SO Valais-Wallis
Nov. 15, 2021 · 12:18 p.m.
Q&A (Pr. Pierre Roduit)
Pr. Pierre Roduit, HES-SO Valais-Wallis
Nov. 15, 2021 · 12:32 p.m.
Loïc Puthod - An open-data acquisition toolchain for AI applications
Loïc Puthod, Centre de recherche Crem
Nov. 15, 2021 · 1:59 p.m.
Q&A (Loïc Puthod)
Loïc Puthod, Centre de recherche Crem
Nov. 15, 2021 · 2:10 p.m.
Cédric Mugabo Serugendo - EnerMaps: The open data tool empowering your energy transition
Cédric Mugabo Serugendo, Centre de recherche Crem
Nov. 15, 2021 · 2:16 p.m.
&Q (Cédric Mugabo Serugendo)
Cédric Mugabo Serugendo, Centre de recherche Crem
Nov. 15, 2021 · 2:27 p.m.
Recommended talks
Energy conversion - Goals and contents
Prof. François Maréchal, EPFL
Feb. 20, 2012 · 4:49 p.m.
204 views