Player is loading...
Pr. Pierre Roduit - Prediction of the electricity flexibility of households using Machine Learning
Pr. Pierre Roduit, HES-SO Valais-Wallis
Monday, Nov. 15, 2021 · 12:18 p.m. · 14m 06s
The increase of new renewables asks for new sources of flexibility to scope with their unpredictability on long term. To control the flexibility of households, it is necessary to be able to predict it. This presentation will present the different methods used to predict positive and negative flexibility of heating appliances and the performances achieved on real data.
Embed
Transcriptions
Note: this content has been automatically generated.
00:00:01
okay i think we can now d.
00:00:04
start from from copying from big set up to the the operational world um on on flexibility
00:00:13
yeah and so on this forecasting it's not supplied to market it's kind of the first deviation of it
00:00:18
as you just have to write a few functions and and think equality now you have to stick for telling us
00:00:25
how to pick that these yeah so thank you for the uh i suppose really to do to give a talk yeah
00:00:31
i just want to prevent a bit what we're doing on the flexibility of all
00:00:34
sorts um that's about affirmative appeared person already also in uh in the ones or
00:00:40
well for some of the questions are more able than it one so let's let's go um i to the stuff
00:00:46
uh we are working a on the basis of to your pen projects so
00:00:50
that's complex and the most projects um where we deploy united infrastructure in uh
00:00:56
about one that also and other elements but mainly i will focus and is also today
00:01:01
and we implemented the months i manage mentioned is also to be
00:01:04
able to control um the eating about its appliances mainly cutting of
00:01:09
ah it bums because most of the also that where in the
00:01:12
us about the project but you go where quite wesson feelers seafood bums
00:01:18
but ah yeah id is also to prototype also
00:01:21
kind of um services such such as analysis of uh
00:01:26
uh eating performance or i don't watch analysis of the consumption to give
00:01:30
advice is to try and if you look to work was done uh
00:01:36
you see but i'm more less buttons installation any reasons market uh well what because
00:01:42
mark matter a gateway just my next to it connected to a an optical line
00:01:47
um and then you have what's on the whites um we but control uh
00:01:52
uh it faster to that we were acting to be able to uh to connect
00:01:57
um not to act on the point and you have multiple sense are different my lesson so that
00:02:02
were added to be able to have a independent measurement of the uh it comes or the appliances
00:02:09
as well of temperature of the winter but of the uh the domestic hot water and someone
00:02:14
all this data is collected that's something like five second waits a pass by second period
00:02:20
it's if yeah if you what of positions so it's quite a lot of data so we are laughing at data
00:02:25
uh the first thing that we are are when you want to go into flexibility
00:02:29
is to um to talk about two type of activity you can she so you can
00:02:33
try to increase the the power of the points with the remote control is not so easy because you can not just because just
00:02:40
shut down that when you can this could decrease the consumption you could try to use the rebound effect we'll talk about that afterwards
00:02:47
so what it wanted wanted to have is to have a place a a production of the lot curve
00:02:51
so uh we work in the data for this more less to a one but also because on that
00:02:57
and you have to be a a a huge map
00:03:00
i'm giving you the power uh as a wage powerball sold
00:03:04
you have on the left uh as a global lot curve and you can see
00:03:09
what is quite logical uses kind of light at midday and a huge line is evening
00:03:15
and you have the inference of eating in the morning and evening mainly intermittent season
00:03:20
that is appalling and you have the one the double curve balls a lot curvy forty bump uh oh
00:03:26
before to the eating more less where it's would use and if you want to go to the um
00:03:34
what you can control also predicting the flexibility that we haven't where we are able to to go to to to to use
00:03:40
you have these two each maps that will produce basins oppose the user
00:03:43
energy use for domestic hot water production and also for the space between
00:03:48
and what you get on the low side is not really a surprise so uh it is that we perform in the
00:03:53
summer uh because yes but mainly in the morning in the
00:03:56
evening and for the domestic out what's a little bit more
00:04:03
stable doing the yeah but still use it a bit more during the winter and each you are nothing
00:04:09
to use nights a bit more in the quite a lot in the morning a bit exhibiting and soul
00:04:16
so if you want to go to the negative activity cutting this
00:04:18
it wants um uh you want to just go first to the um
00:04:23
what does a parameter you should use to be able to products uh as an olds and such
00:04:29
the first one was quite logical is that uh uh
00:04:32
eating is infinitely uh correlated to uh uh to other temperature
00:04:37
it's not too surprised and what you see is domestic out what i'm doing so yeah is much more stable but that if you also
00:04:44
what is the energy you have at your disposition uh at your disposal to um to actions agreed
00:04:54
what does i think we wanna i wanted to go is uh is was all about what kind of model we want to have because uh
00:04:59
if you are where you're talking about the s. m. for big buildings most of the time you are making a physical model and trying to yeah
00:05:06
model you building bumble bee windows model everything to be able to know how much any energy you can store not
00:05:12
for small ourselves extremely complicated to get all the information about sites and we wanted
00:05:17
to have a look about what instruments of putting the others all sorts so mainly uh
00:05:24
what we just show he is the one that eighty wheeling and when we are working we for
00:05:30
we've pools of put of one also is not really a something but if you're putting it
00:05:35
in a group of two also falls all nine also eighteen hours on this one that will give you more nice
00:05:41
the histogram of the of weight our piles all that you get and it's not really surprise but if you increase the number of the
00:05:48
pool uh you ever have you will have more more a normal distribution
00:05:53
and what you can see is that if you go in that direction
00:05:57
starting from a pull to fall for t. two fifty also rock coming close to
00:06:02
the lower limits where to put the quite interesting ah quite fitting to a normal distribution
00:06:07
so we worked on uh after that's more on tools because working pretty thin and all sold
00:06:13
consumption is something conception of energy the predicting when's it been pretty will start to not start
00:06:18
it extremely complicated especially for small buildings but in the pool is starting to be more interesting
00:06:24
if you look to the data uh on
00:06:28
the negative fixing hazel cuttings loads um uh
00:06:32
we used a different parameters for protecting its
00:06:36
ah buys values of the of the lot curve
00:06:40
real tall using the mythological few good features like temperature no edition i'm not time features because
00:06:46
yes it's still quite an overly uh influence of all the people are uh da fins
00:06:52
all sorts and also the things like uh all the days screwed up a bigger bodies
00:06:58
uh if you have a look to the graph you on the mikes between radiation temperature and eighteen
00:07:03
you can see also that you have some discrepancy uh uh what i would call
00:07:08
discrepancies as an example in the evening windsor edition just that you have an apart
00:07:13
degrees of the eating quite logical uh it's it's come in the building you get the sense altogether eating sedition directly
00:07:21
when when the sun just go off each way it's quite a long time
00:07:27
before the things which is why zinc and you have this kind of uh speak
00:07:30
of synchronisation of the points that just go off after the id been part of the building and we go off
00:07:37
it's not really directly lower liberated us is kind of a delay of the war
00:07:42
damage distribution system from the concept of quite an influence also and um the system
00:07:49
so we does it quite a lot of um models custard models
00:07:53
but that is this one a little regulation on the forest a media
00:07:58
a best of one uh like g. b. m. um uh exposed to imax just to test a different kind of model and to see
00:08:04
how much you could achieve a production i would that's a that's a best modern could achieve could ogle better than that
00:08:12
but that was that was the first white and here you have to ask was the production of the loads
00:08:17
um that was quite cool if you go uh for a half an hour forty five minutes to still
00:08:23
okay but you see that if you are going to five um us yeah the production is before we
00:08:30
still edgy then was a best smaller new case and yeah i can give you a curve
00:08:35
of the predicted values uh arts five thousand fifteen minutes so fifteen minutes of production
00:08:40
is really good uh five uh was you still have a cruel but it's not perfect
00:08:48
but
00:08:51
let's go to another field notes negative flexibility negative fix it is quite easy says predicting the lot curve
00:08:56
or the point now or it's what could we achieve if we are making cuts produce a positive flexibility
00:09:03
an increase of consumption so we worked on uh so we wanna get faculty palms
00:09:10
as a positive flexibility a positive consumption um and um uh the
00:09:17
main causes of this week after as a as a those are the
00:09:21
end of the cat is make it was by synchronisation of the pens
00:09:25
and also composition of the energy that was not use for doing that
00:09:31
it's extremely tedious to get data on the real cuts because set me that you want to have a data
00:09:37
for a few hours a day if we uh uh if we mind for every type of freaks an early days
00:09:42
that mean that you have to make cuts and adding cuts that i've not inference on the next gotten sunday
00:09:46
is quite well full that mean a lot of data
00:09:49
collection uh that means that we twice fast to analyse its
00:09:55
v. simulation is trying to first simulates the sink wages the synchronisation
00:10:00
of impact so the the fact that if you are making a catch
00:10:04
you will achieve in synchronisation of some it been that will start all at the same time producing this big
00:10:13
uh what reading is a is just use a standard lot curve and
00:10:17
just sift the laissez bump as if they had been synchronised way cats
00:10:22
so you push the button that normally get started doing is it got to the end of the cat
00:10:27
that means that you are not a compensating for the the last energy but still you are
00:10:32
making the synchronisation having an analysis of the data um we did that for the whole yeah
00:10:39
and we did also a principal component analysis based on the curve yeah yeah uh i see
00:10:46
and i'm not having a log about the synchronisation
00:10:51
but you see we just to first um a component of the p. c. a. uh you are the first company
00:10:57
that is really being i don't know well all you can now also i've uh a component that is rising and falling
00:11:05
this kind of two direction is quite easy to understand the first one
00:11:09
is may need between summer and winter if you are in the winter
00:11:13
as an oldies i yeah normally after the cuts will be we mean ah yeah
00:11:18
in is uh some ivy synchronise of the bump you a lot more
00:11:21
less no conception of the what's inside the first dimension of the p. c.
00:11:26
if you go to the second one it's small base and time of the day in the
00:11:30
morning as most of the time to just shut down all we'd use the consumption that means that
00:11:37
that's logical that's even if you are making the synchronisation there will be already action of the conception of the walls
00:11:43
or if you are close to the end as often want if we're going to go the other ones and go to an increase
00:11:51
so what he did since yeah and it's twelve two parameters that where uh uh
00:11:55
uh what's consumption before the cats and also time of the day and he you have this kind of uh
00:12:02
maybe an an quartile of all the data that we collected we simulation
00:12:06
for the yeah it's becoming to fit quite well uh that means that
00:12:11
fifty percent of uh in this uh we if fifty percent of the categories
00:12:14
of the yeah are between these two uh yeah so that means that you are
00:12:18
yeah but to be quite good still we have to learn the model to be able to do something but for for some noises it was already something
00:12:24
and perhaps uh
00:12:28
what we did finally is also to compare that to we'll cuts
00:12:31
support i misread grammy you may it making cuts and tended buttons
00:12:35
and if you make a k. it's in often depends on to us that doesn't mean that twenty points are impacted
00:12:40
so we just took the ones that way back to it and back at it to make a comparison with that
00:12:46
and we see that's a form of decreasing is quite okay but there is already quite a big
00:12:53
influenced the uh um uh that is just that's when you synchronise after
00:12:58
cut most of the time the pond we go to domestic hot water production
00:13:02
that means that you have an increase of the power because you haven't i yeah a consumption for producing what what are
00:13:08
what you do with the flow of tampa of water that is going to building a free free forty forty five degrees
00:13:14
but still that was the first step we still have to work on sat and goes there
00:13:22
so to conclude production of the log of of it bums
00:13:26
it's quite good on my side um uh give you a good
00:13:31
evaluation of the fact is that you have been also if you pull them which group of forty fifty years old
00:13:37
we work on the positive flexibility so this kind of using the rebounding effect
00:13:41
and i'm using the of which both up our before the cats and
00:13:45
uh i was a day seems to be a efficient to get something
00:13:50
ah it would be nice problem all cats to compare the simulation and a ditty
00:13:55
and it's yeah seizes differences and achieve twelve about the moderns ian
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
Energie solaire : de l'état de l'art de la recherche, au développement d'une plateforme de prédiction pour les distributeurs
Dominique Genoud, HES-SO Valais Wallis - IIG
Jan. 29, 2016 · 2:21 p.m.
120 views