Speech Transcript - DIRECT LOAD MANAGEMENT OF ELECTRIC VEHICLES

0:00:13	to men from is that there and also uh rubber the bob thomas of course not
0:00:18	um
0:00:19	so i'm going to talk about the let to got be he goes so these um actually is a work
0:00:23	that we are the general rising to out those types of remotes so
0:00:27	state you ones for future development of these work
0:00:30	um
0:00:31	so a goal
0:00:32	and there could be a little bit about the the this patch um and and uh
0:00:38	a traditional additional load forecasting
0:00:40	and uh also the approach is are put forward for row controlling loads
0:00:46	and and then i we discuss what what proposed as an architecture two
0:00:49	manage the scheduling a uh i think a lot of the speakers before or a the fabled what of into
0:00:54	using the problem or
0:00:56	closing the loop to the consumer are making uh the consumer are
0:01:00	uh a smart to um
0:01:02	i would have to make a a a brief analogy suppose you have a store where you only sell one
0:01:07	product and the
0:01:08	any customer the calm thing the store has a different request but no matter
0:01:13	um no matter what price you fix the customer we by whatever of was the preset the amount
0:01:19	uh and we get out of the store so
0:01:22	or the C D's is this would be a fantastic market to operate any store
0:01:27	um if you could just slide the price be what you wanted
0:01:31	uh because the customer always by is the same amount of wanted uh we do any is smart decision about
0:01:37	the price
0:01:38	so changing um
0:01:40	the cease them uh i think these right now means that you actually will allow this some or to change
0:01:46	the demand once this store poles
0:01:48	it decoder surprise
0:01:50	a in this will make these market all but it normally otherwise we will be stock with the enormous amount
0:01:56	of regulations
0:01:57	uh and the continue to to just not all but it these efficiently in
0:02:02	as a mark as the a market
0:02:05	so uh um now a that you be a goes uh uh uh of friends them for was um they
0:02:11	friends because have no emissions they don't defined a noise so we gonna let to L at T C you
0:02:15	can be produced by a we're from any many sources
0:02:18	a how and and but they are a high voltage levels there the bow out of a are um plasma
0:02:24	muddy bees
0:02:25	uh a you much isn't turning go putting your to battery to charge would would be a but apps equivalent
0:02:31	to to to to that amount
0:02:33	i energy requirements so
0:02:35	um
0:02:37	and just need a big problem in in people are afraid or a or for what can happen in fact
0:02:42	it was even suggested then in california you to get a spectral permission mission
0:02:45	an at three got car
0:02:47	um a because the you D D was really afraid of having to to eventually you deal with black kind
0:02:53	of what's and other the problem
0:02:55	um
0:02:56	so i in jan and that all um
0:02:58	we why these the problem because that has nothing but control along the load right so the or or or
0:03:03	the the this patch
0:03:05	has to match demand and supply in have time
0:03:09	one of the big problems in uh uh in uh the like this is that the is all storage right
0:03:14	so that has to happen you a time
0:03:16	um
0:03:17	the
0:03:19	just to make an example you you could store energy if you had to be good a or or a
0:03:24	a a and that take upon to right you could pump what the up instead of down
0:03:28	and in general in terms of scale it's it's really complicated to store energy efficient to D
0:03:33	so it's it's a it's a as it's that well was done a either a loss
0:03:38	uh so the the prediction on the the the the pretty job see something that can excite the fantasy you
0:03:43	of signal was people body
0:03:45	it turns out that a lot of the techniques that we developed in the eighties uh either
0:03:49	broadly applied to this problem a lot of statisticians stations also work on this problem so use a very much
0:03:54	you field and
0:03:55	uh a as long as the load continues to to to to go to a in in a way that
0:04:00	i
0:04:01	where handle by these models
0:04:03	um yeah things
0:04:04	don on you need uh a new significant to research
0:04:08	um and so i and that it time can but we have used ten percent accuracy the prediction of below
0:04:13	the D is fed to the
0:04:15	um the to to to these uh up to make optimal dispatch of energy and and think scroll uh from
0:04:20	there so that is the load
0:04:23	these audience not
0:04:25	uh this is sadly um
0:04:27	so i mean yeah with these at the mean allergies or the learnt
0:04:30	yeah is the power with the D's there are an essential essentially by any um um any um apply that
0:04:37	you plan can by the like that to there are known and by your head condition and so on
0:04:41	um and and the E these mathematically effectively a signal is a complex signal
0:04:46	is that i'm bill up uh or of uh a um
0:04:49	if it is or the represents the product between the touch
0:04:53	and the current conjugate to gate
0:04:55	uh a at a particular or a line
0:04:58	and the again these are complex because that a is them but up the complex envelope up of a signal
0:05:03	which
0:05:03	ease a around the the the operating frequency fifty hz for be you rope sixty hz for being nine states
0:05:10	um is not just
0:05:11	uh the generated obviously uh
0:05:14	by you single system in terms one we we look at bus C sorting are look that the
0:05:19	uh uh a a a bath systems so were shown before the that these select a to read is where
0:05:24	uh a sample of the transmission systems
0:05:26	and the every line in that system obviously bundles up a a a lot of requests the come from the
0:05:32	distribution that a clear that live
0:05:34	to your um
0:05:35	um
0:05:37	so that that that to die
0:05:39	or or of a a request
0:05:41	uh that for this low
0:05:43	uh one have generated by you think a would describe as of dom at as like electrical vehicles of these
0:05:48	right so they on with the search and
0:05:50	uh the state for they back to T and they want to feel it up
0:05:53	uh a up to a certain amount
0:05:55	and in a lot of the discussion the is done right now the the the question is whether you would
0:05:59	actually to
0:06:00	a a new provide energy uh in in in chunks but actually that's very bad for the bout in fact
0:06:06	you should actually continues it
0:06:09	um and so
0:06:10	uh in the model that we adopt we assume that these uh electrical be close have not be have to
0:06:15	ball once they are on on they we say you on in charge
0:06:18	for the entire duration
0:06:20	and i say that the same applies to that appliance is like your dishwasher your order
0:06:24	washing machine and someone
0:06:26	and that not going but is that the data are more storage systems
0:06:30	oh as team said to if you did date or see there's that conditioners they we have in a measure
0:06:34	state the temperature sure they want to read it to a them point
0:06:38	and they will uh uh C adjust just their request
0:06:42	based on what they money to use the state of the uh of the system
0:06:46	so i'm not going to
0:06:48	this anything that is a element
0:06:50	uh to though
0:06:51	um but i think that for example the the methods that's from was a mentioning of pricing that could be
0:06:57	easily applied to those
0:06:59	um
0:07:00	so a a is mad at me dating has made the low but but is not steal clear to how
0:07:06	to make it can above
0:07:08	uh a mean initially was saying that he's going to have by january
0:07:12	some uh where you
0:07:14	uh uh of uh closing that have but that are a lot of a you should we send the price
0:07:18	uh to each individual device should we send the price to where her
0:07:22	energy management system a lot of proposals this is one of them
0:07:27	and that that he's one is based on the fact that if i want to have them in sections
0:07:31	i space if i
0:07:33	well i what i want
0:07:34	uh and actual specified to because these has to be automated
0:07:38	to a computer
0:07:39	so how do i specify what mike our once
0:07:43	to a compute system that can automatically
0:07:46	uh schedule it
0:07:47	um and schedule it's to optimize the shape of these low
0:07:52	okay so i'm assuming but these can be directly controlled because
0:07:56	uh the load we
0:07:58	the many quests and the digitising these request is what these the talk he's about
0:08:03	and that i can uh delay uh they the time at which these request is mac
0:08:08	um
0:08:10	so um that a be two approaches i'm i'm gonna be very brief on these
0:08:15	uh one is deemed be man side matt
0:08:17	but the much money money will this small
0:08:20	when is price based and one is lack control
0:08:23	uh and again i i i T meg and he said that you may have pricing
0:08:28	uh a five minutes before and then after a uh um after a why you wouldn't want to just to
0:08:34	pricing
0:08:35	uh you need to some form of control
0:08:38	um and the i i very much uh it agree that uh a right i think may cause problem so
0:08:44	the time but i think that a cat in you are can see that is one it's that the additional
0:08:49	one is time of use
0:08:50	depending on the time of day to to be like set phone systems
0:08:53	you are charge a different eight
0:08:55	um and the the didn't have time i think he's the new um
0:09:00	is the new wave for a a uh where are the despite is that actually a uh are computed
0:09:05	based on the spot market of energy which already exists and the consumer is made aware of that
0:09:11	uh and some how some automated system takes advantage of that
0:09:15	the lower controlling is another strategy uses skyline
0:09:18	to S be and you walk so
0:09:20	for some incentive steve out to have your load into adapted all my conditioning condition we being in not that
0:09:26	on an optimal
0:09:27	i'm
0:09:28	by the uh uh T D um and i get some benefit out of these might cost is be used
0:09:33	or negative even directly money based on how many times the use my sir
0:09:37	so these are it would these call than see that said be say could be part
0:09:41	or of O over these the cease them that uh can but lies to optimize the the neck
0:09:48	um so that problem so obviously we'd plan seeing
0:09:52	they rebound P
0:09:53	problem that was mentioned by solely needs set be a problem of a a a have to imply i think
0:09:57	in need use a dynamic that system so there was a question before and stuff but
0:10:01	so there is a question whether these would cause oscillations which writing the worst
0:10:05	um
0:10:06	and uh you can obviously understand that the the time of use uh maybe be actually a a a a
0:10:11	problem a D uh approach because he's site up much had of time and i think shown you uh explained
0:10:17	that but was on not gonna spend more time and these
0:10:21	um the problem a lot "'cause" statement to with this sort of having a a a a a control
0:10:25	uh made by they they the control station is it is very hard to understand how much you inconvenience this
0:10:30	to the the the consumer so
0:10:32	um
0:10:34	it's it's hard to predict with a few what with sky beans were arms
0:10:38	uh or how you are going to make the control so that you don't in
0:10:42	i actually people this a skype select we have enough people it's a so
0:10:46	that is have the questions i thing is that takes that into account because you can set up your system
0:10:51	and respond to be
0:10:52	uh uh while a lot of time at is on top down and so you may be very unhappy happy
0:10:56	what the ut
0:10:58	so i think it is a good
0:11:00	so motion because by specifying X C you what you want that this time
0:11:05	uh then you can you both pricing and
0:11:09	safety uh you me
0:11:11	decide how to shape
0:11:13	uh the load
0:11:14	uh uh and and they you didn't know that that "'em" that is trying to a marriage here as so
0:11:19	how how do you
0:11:20	do you make energy to fact it so that every time the to one sending thing there is a section
0:11:25	that goes on
0:11:26	um so the idea is that you have actually to specify what you want so
0:11:31	a obviously not a words are are uh a like a tickle be can so we assume that is that
0:11:37	the additional load that you can pretty used
0:11:39	we ten percent of accuracy and that
0:11:41	so an imposed to these this is a lead to be got a he call wrote
0:11:45	uh and the key idea is S one bond but it so to say that each or a that this
0:11:49	process the cost it use the there's is matt
0:11:52	you be load
0:11:54	is not something that the axe price but he thing that are informed about
0:11:58	uh so that is an that i'll a process uh which should we assume it's a possible process
0:12:03	uh T I a a are uh they are rivals
0:12:06	or these that that they got be equals
0:12:08	associated to to each other i well
0:12:10	uh uh is a a a a a charge model
0:12:12	you want to be charged for for the system on C yeah up i
0:12:16	and so uh the the good of we that we have is that we can actually change these time mean
0:12:21	to a a a a a a time
0:12:23	yeah i D E which is greater than T I i
0:12:26	uh so that the shape
0:12:28	oh these is mapped to load just
0:12:30	is that's close as possible to what helps a a a um paying less for the consumer inconvenience sing i'm
0:12:37	the least scroll C board
0:12:38	and be sensitive to price
0:12:41	so
0:12:42	uh obviously that i'll a process uh uh um is uh it is uh is the simple step staircase profile
0:12:48	i
0:12:49	and V is the uh the team on the process so
0:12:52	we are discussing how to perform these mapping
0:12:55	okay so so that we were want to communicate
0:12:58	is the time about i arrival
0:13:00	and the charge
0:13:01	and
0:13:02	and then and then that uh it's very easy to understand that you would have to one ties
0:13:07	but also i them
0:13:08	uh but i in to think part is it one time i is the amount of charge for example you
0:13:14	use for beats to quantise and you have sixteen classes
0:13:18	what you basically do least turn this P the control problem is actually
0:13:22	unit net for a problem because this classification so these saying that each charge should be lost with this screwed
0:13:28	side
0:13:29	essentially once every one of these requests
0:13:32	in a Q
0:13:33	so what you are doing is effectively storing D on tell of sorting
0:13:37	energy you're using that
0:13:39	uh to work around the problem that you can not store energy
0:13:44	and at this point you can sort of the lack of hmmm but was that are like that the certain
0:13:48	time
0:13:49	you just as of this that and number of a drive so but a certain
0:13:53	uh a time you need
0:13:54	and the schedule or should makes a certain amount of these two D pot
0:13:59	i assistant time
0:14:01	and the i if you uh the like the the the difference between the top of departure at time K
0:14:07	T and in time came as T for is so the amount that should be part
0:14:11	in these in about
0:14:12	then you can be shape the load
0:14:14	uh essentially one bad being the shape we
0:14:18	uh with
0:14:18	electron you parts that has exactly the same duration which
0:14:23	ease the load profile i or of the radical be he called
0:14:27	um
0:14:28	this is actually to think point you can actually to stack up
0:14:32	uh and i mean this Q and is is
0:14:34	that's used for for an only meeting
0:14:37	because i i have to a a uh are essentially a a back
0:14:41	the information all should start
0:14:43	yeah is to find the that he's uh that they are the process match the optimum
0:14:49	uh departure sure of process
0:14:52	uh uh test and time
0:14:53	and you just send back this time stamp that says if you want ride
0:14:58	at this time
0:14:59	please go ahead and switch
0:15:01	K
0:15:02	so these is the feedback mac
0:15:04	um
0:15:05	i so at this point these um it is just essentially where we are thinking
0:15:10	is that the we got the eagles we'll interact some soft after it could be in your i for or
0:15:14	or or are there's not P D A or we could be in your house to could be just some
0:15:18	soft in computer
0:15:19	which we are but it does and home energy management system
0:15:22	and we've commuting at is like at the seizure and
0:15:25	to two uh uh uh a place that could be situated weighted a at the level of each that's form
0:15:30	or or or perhaps even better that at the substation level
0:15:33	so the substation level
0:15:35	would be in he's the beep what is the expected profile five i
0:15:39	or or brought consumption for that they
0:15:42	and and you also some network management to that could be a question is stick play scene
0:15:46	renewable energy especially specially if these is available in your uh uh uh at area
0:15:51	and uh essentially what they don't the decision on would consist on is
0:15:56	deciding what is the optimal
0:15:59	a departure rate
0:16:00	so that the shape
0:16:02	all of these
0:16:03	process
0:16:03	matches that's as much as possible
0:16:06	the energy that you have available
0:16:09	so i'm just gonna clue very quickly with sample mean and is at some no gonna show that the that
0:16:14	the the actual of these optimize but
0:16:17	um
0:16:18	the these um but is going to be more soon i'm i just a to the paper on uh we
0:16:23	money motion to the battery column
0:16:25	with these uh uh uh formulation of these optimization so i imagine that the the the profile or or power
0:16:32	available is these beam a profile D is the combination of all
0:16:36	uh you're a a a a that i had
0:16:38	bead for the consumption
0:16:40	or potentially leading you know where
0:16:42	and nine as the the the the the additional load
0:16:46	that has to be mad
0:16:47	that is present in the network
0:16:50	um and a uh suppose in is you are controlled electrical be equal loads
0:16:56	you can actually just you deal i is a a the the the the the scheme that they just proposed
0:17:02	in select optimally these the patch time over a finite or as a windows
0:17:07	to bad to a in the getting line
0:17:09	so these would be the schedule old
0:17:10	so as you can see you see what you expect to you see that these provide a do not just
0:17:15	as much as possible
0:17:17	um the blue line
0:17:19	how are and and the parents it is not a great you you have to imagine a these were work
0:17:23	better in C state
0:17:25	um
0:17:26	however at the important
0:17:27	a point is that it is also
0:17:29	paying a
0:17:31	for the delaying the consumers
0:17:33	uh uh and then further it is not necessarily matching these exactly
0:17:37	uh it doesn't have also the degrees of freedom necessary to to these exactly
0:17:41	but a a a a a a mouse so it tries to also meet them uh uh either a certain
0:17:45	time that is not excessively high
0:17:48	so here we have a
0:17:50	it peaks of eight plus a have a process
0:17:52	three you a part time you need
0:17:54	uh and we have a new look ahead or eyes on all five time you neat uh it could be
0:17:59	mean it's or five me periods
0:18:02	um
0:18:03	and and uh uh we we find that you the the um
0:18:08	the delay is no
0:18:09	more than fifty you in this particular example so you would be delayed
0:18:13	about the a and now way
0:18:15	and the the is the main that uh these is that you can
0:18:18	essentially being to these to a fifty percent reduction of deviation
0:18:22	uh from the leading up profile uh and the the modulation is that you are not trying to sort of
0:18:28	uh let the dynamic that system find a query be "'em" are actually engineering the optimal shape
0:18:33	so this is the mixture of control pricing
0:18:36	you're a word of pricing but you are not giving a single price to every age
0:18:41	a you what actually doing very different actions for the different an agent to to read these profile
0:18:46	and these
0:18:47	includes
0:18:48	my doc was keep these
0:18:50	so uh in conclusion an efficient market
0:18:52	need to have some way of sailing on good opportunities need to be aware of pricing
0:18:58	um
0:18:59	but these these uh a a a a market that has a lot of safety constraints
0:19:04	and therefore
0:19:05	uh some optimization framework is necessary
0:19:08	a i think that more than a simple feedback control
0:19:11	uh there is a simple feedback control system that quote control the load
0:19:16	i think in the low needs to be a bundled them put all see billy uh that we
0:19:21	as the service
0:19:22	so network took rose and Q thing
0:19:25	can place
0:19:26	uh you know
0:19:27	a a through we in uh in uh in the storage technology
0:19:31	um
0:19:32	and so the basic principle of to do so at used one bundle indeed you dies the loads so that
0:19:37	to one can exactly express what everybody
0:19:40	um what i do you once
0:19:42	um and i think that is a lot of space and signal processing to eight to that process
0:19:49	okay thank you
0:20:07	yeah
0:20:10	and i want to uh in my job might but at P that these much so the use a much
0:20:14	i want to charge
0:20:24	so
0:20:25	so in his model but we model the delay a uh we've ease uh um
0:20:30	any inconvenience cost
0:20:32	right
0:20:33	so we are of trying to minimize that to in our objective function
0:20:37	um um
0:20:38	and you can imagine that you could put it as a hard constraint
0:20:42	um
0:20:43	uh we have and that it with i think constraint because we wanted to see
0:20:47	uh just by playing with a different weights they we gave to every cost
0:20:52	could be predictably have
0:20:54	it is a novel amount of course
0:20:57	now dealing with D V D a constraint on the in of course all might also ease complicated to these
0:21:02	model because
0:21:03	it would be much better in terms of also source but
0:21:06	to lose the a C fig
0:21:09	uh a on the consumer to just scale you the number better that we go at this time
0:21:14	okay
0:21:15	so i think more practical in these environment also force got be D D would be
0:21:21	just trying to play with the parameters and the cost so that you in fact
0:21:25	give something which is this T
0:21:27	which is a little bit the net model i mean they they don't all when i want to have the
0:21:31	five
0:21:32	like they just the i to put enough D
0:21:35	that i one be to uh this i i
0:21:38	and i think that's a more reasonable
0:21:40	more than
0:21:47	okay thank you think i mean

DIRECT LOAD MANAGEMENT OF ELECTRIC VEHICLES

Tomorrow's Smart Grids

Presented by: Anna Scaglione, Author(s): Mahnoosh Alizadeh, Anna Scaglione, University of California Davis, United States; Robert J. Thomas, Cornell University, United States