August 29, 2024 • 38 mins
In this episode of The Energy Beat Podcast, we dive deep into the intersection of artificial intelligence and energy management with special guest Dr. Bill Burke, CEO of Virtual Peaker. Get ready to explore how AI is not just a buzzword but a powerful tool transforming the energy industry from the ground up.
We kick off by tracing the origins of AI in popular culture and quickly shift to the present day, where Bill shares his journey of founding Virtual Peaker in 2015. Hear about his ambitious vision to make energy demand as controllable as a gas turbine generator and how his company is turning that vision into reality. From the early days of programmable thermostats to the latest AI-driven innovations, Bill walks us through the evolution of demand management technology.
Discover how AI is helping utilities manage the unpredictability of renewable energy sources like wind and solar. Bill explains the role of machine learning in predicting energy consumption, balancing grid demand, and even optimizing when electric vehicles should charge to avoid overloading the grid. We also delve into the challenges of AI adoption, including the trust issues utilities face and the importance of seamless customer engagement.
This episode is packed with insights on how AI can enhance customer support, predict grid constraints years in advance, and ultimately smooth the path towards a fully sustainable and efficient energy grid. Bill and host Jen Szaro discuss real-world applications of AI, like Virtual Peaker's top-line demand control technology, which offers precise control over energy demand and promises to delay costly infrastructure upgrades.
Finally, we look ahead to the future—what does AI adoption in the energy sector look like over the next five years? How can utilities and consumers prepare for this technological shift? And what are the limitations of AI in solving the industry's biggest challenges?
Whether you're an energy professional, a tech enthusiast, or simply curious about the future of AI, this episode provides a comprehensive and engaging overview of how artificial intelligence is set to revolutionize the way we produce, consume, and manage energy.
🎧 Tune in to learn how AI is leading the charge towards a greener, more efficient energy future—no Terminator scenarios included!
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Jen Szaro, AESP (00:08):
Welcome back to the Energy Beat podcast.
Picture it.
New Zealand 1863, author SamuelButler released an article
Darwin Among the Machine whichis credited by many to be the
beginning of the pop culturetrope of artificial intelligence
we know today to be thebeginning of the pop culture
trope of artificial intelligencewe know today.
Fortunately for us, butler'svision of us creating our own
(00:28):
successors has not yet come topass.
In fact, today AI shows thepromise of helping us solve some
of our greatest generationalchallenges, and without the
threat of global annihilation.
And today we are joined by DrBill Burke, ceo of Virtual
Peeker, to talk about how AIwill help us tackle some of our
(00:50):
greatest contemporary energychallenges.
Welcome, bill.
Bill Burke, Virtual Peake (00:54):
Hello Jen.
Thanks so much for having meback on the Energy Beat podcast.
I'm excited to be here.
Jen Szaro, AESP (00:59):
We're excited to have you.
You're doing such great stuffwith your organization.
I can't wait to dive in alittle deeper.
So before we do that, tell us alittle bit about yourself and
tell the audience all aboutVirtual Peeker.
Bill Burke, Virtual Peaker (01:10):
Yeah , thanks a lot, Jen.
I'm Bill Burke.
I'm the founder and CEO ofVirtual Peeker.
I started the company back in2015,.
Really with the key vision ofhow do we make demand as
controllable as a gas turbinegenerator, and today we're
working to really revolutionizehow we manage demand on the
electricity grid, making it moreefficient, more sustainable and
(01:33):
ensuring a prosperous andsustainable legacy for future
generations.
Jen Szaro, AESP (01:39):
So tell us.
How did you come up with thisidea for the company?
What drove you to launch suchan amazing and innovative
approach?
Bill Burke, Virtual Peaker (01:47):
Yeah , thanks for that, jen.
So I've been in this space fora very long time.
I started back in 2006 workingon a project, doing my graduate
work at UC Berkeley, working ona project to create the first
programmable communicatingthermostat project to create the
first programmablecommunicating thermostat and
ever since then I've beenfocused on this area of how do
(02:08):
we help utilities get behind themeter and control devices.
I focused my dissertationresearch on it.
I worked on that at GEAppliances, really applying this
idea of getting behind themeter to help utilities control
load across the entire home.
And then this was very earlydays of Internet of Things, you
(02:29):
know, I think, 2010.
So Nest Thermostat hadn't evenentered the market yet, and so
Internet-connected products werevery new.
Spent some time there at GE andthen left to start Virtual
Peeker, really with this bigidea of making demand as
controllable as a gas turbinegenerator by helping utilities
(02:49):
control those internet-connectedproducts.
Things like thermostats andwater heaters and batteries and
EVs and that big vision ofdemand as controllable as a gas
turbine generator has taken sometime to really come to fruition
, but with our latest technologydeployments that include, you
know, some innovative AIapproaches, some innovative
optimization approaches, wethink we're getting very close
(03:13):
to really bringing this toreality.
I'm pretty excited about it.
Actually, it's something I'vebeen working on literally since
2006.
Jen Szaro, AESP (03:21):
So I'm not going to lie, you're kind of
fulfilling my college dreams bywhat you're doing.
I remember starting out as aphysical chemist in this
industry and working in thephotovoltaic side of things and
watching the changes happen intechnology and it was never fast
enough for me and I alwaysenvisioned this whole idea of
distributed generation, kind ofreplacing power plants.
(03:44):
And now you're making thathappen.
Did you have any doubts whenyou're going through that
process of building the companyand how did you push past that?
Bill Burke, Virtual Peaker (03:54):
Yeah , yeah, starting a company is
not for the faint of heart, forsure.
Of course, doubts along the way.
So why do people care aboutthis concept of demand is
controllable as a gasregenerator.
It really comes back tophotovoltaics, which you were
working on as a physical chemist, right?
The wind doesn't blow all thetime, the sun doesn't shine all
(04:16):
the time, and so renewableenergy is intermittent.
It's intermittent in nature,and keeping the grid in balance
is something that has to be donein nature, and keeping the grid
in balance is something thathas to be done, and today that's
primarily being done by usinggas peaker plants or some sort
of fossil fuel energy source.
And so if we can turn thedemand side of it into a
(04:37):
controllable resource thatutilities can utilize to help
balance the grid, which is whatwe're working on, we think that
we smooth the path toward atotally green grid.
That's the reason we're reallydoing all this, and it's taken
some time.
Everything has taken some timein this energy transition, but
we're pretty excited to be, Ithink, at the forefront of
(04:59):
making this a reality.
Jen Szaro, AESP (05:01):
Yeah, some of your projects are just really
outstanding in this reality.
Yeah, some of your projects arejust really outstanding.
So, as we move forward with thelatest and greatest in
innovation in this area ofmanaging demand and using it to
kind of fill in those peaks andvalleys, we've got the advent of
AI.
It's relatively new, fairlywell misunderstood, I think, by
(05:22):
any non-technical crowd and eventhose who are in some technical
spaces, and the terminology is,you know, very confusing at
times.
I think so for our listeners,can you quickly spell out what
you see as the relationshipbetween AI and machine learning,
just for those who might notunderstand that terminology and
(05:42):
what the differences are betweenthe two or how they're
technically, how they're relatedand kind of what that is going
to mean for our industry?
Bill Burke, Virtual Peaker (05:51):
Yeah , yeah, yeah, absolutely.
I'd be happy to talk about that.
So I would first maybe pushback a little bit on the concept
of AI being new.
I think it's a relativelyrecent entrant into the
zeitgeist, but you know, ai as afield has been going strong for
many, many years I think Idon't know the full history, but
dating back at least to the 60sand before and there's been
(06:13):
lots of different types ofartificial intelligence.
So what is what is AI reallymean?
It means teaching computers howto reason about the world and
make decisions sort of on theirown is a large part of it, and
so that's sort of the broadconcept, and it's had many
different forms over the years,with tons of subdisciplines.
(06:34):
Things like fuzzy logic was abig one at one point.
Expert machines is anotherpiece, and AI is composed of all
of these different ideas.
I think machine learning is amuch newer, maybe newer entrant
into that sort of overall bucketof what AI is, and machine
(06:57):
learning is really primarilyabout feeding the computer lots
of data and letting the computerlearn on its own what that data
means, and oftentimes machinelearning results in models.
It creates models that can thenbe used to do inference on the
world.
So prediction, if you will.
(07:17):
And those models are typicallynot human, understandable,
meaning that we don't know howthe machine necessarily came up
with that decision.
It's not.
It's not reasoned about in thesame way that we reason about
things necessarily.
But that's the core idea ofmachine learning and, again,
machine learning is a subset ofartificial intelligence.
Machine learning typically usesthings called neural networks.
(07:41):
Things called neural networksand the AI chatbots are sort of
a massive version of thattechnology.
If you will, implemented oninstead of data, like virtual
peakers using, like, which aretypically numbers, but these AI
chatbots are implemented on,obviously, words and language.
But it's all stem from the samecore technology idea of feeding
(08:05):
the machine lots of data andletting the machine come up with
, effectively, a model of theworld.
Jen Szaro, AESP (08:12):
Right.
So I mean, I think today mostpeople think about AI, machine
learning, and they think aboutchat, gpt, right, because that's
pervasive.
Everyone's kind of seen it andplayed with it.
I've used it myself to craft afew outlines and come up with
some snazzy session descriptions, but those are generally
different kinds of models thanyou would probably be using in
(08:33):
our industry, right?
Because those are more focusedon gathering knowledge from
what's out there in the internet.
So they're what they consideropen models.
Is that correct?
Bill Burke, Virtual Peaker (08:42):
Yeah .
So I think that those modelsare typically being trained on
like I said a couple of minutesago, being trained on human
language, and they're using thewords written on websites and
books and whatnot in order tocreate a model for how people
talk and how people think.
Now that sort of technology issomewhat applicable to our field
(09:07):
, but it's generally moreapplicable in the form of
customer engagement sorts oftechnologies.
How do you make it easier forusing something like ChatGPT,
like you said, creating outlines?
How could you use it to helpease the support burden when
you're handling a support issuewith a customer?
(09:28):
Or how can you use orpotentially using that sort of
technology to write marketingcopy that you would then send
out to customers in order tobuild support for whatever
you're doing?
So that's definitely still used.
What we do is somewhat differentin that we use the same sort of
general ideas of neuralnetworks and then feed it
(09:51):
numbers typically right.
We feed it data streams thatwe've captured or the utilities
captured and learn about theworld through numbers, and then
use those numbers to doprediction, and those
predictions usually take theform of some sort of maybe a
prediction on the energyconsumption tomorrow, maybe a
forecast for the total systemload over the next several days.
(10:14):
Maybe it's forecastingindividual device behavior over
the next several days, so it'sreally predictions on what
machines are going to do and howthey're going to consume energy
, and that's the way we're usingartificial intelligence or
machine learning today.
Jen Szaro, AESP (10:35):
Right, because you've got so much data coming
in.
I can't even imagine a humanbeing trying to make heads or
tails of all of that datawithout the assistance of
something like machine learningand AI.
Bill Burke, Virtual Peaker (10:47):
Yeah , I think that machine learning
what it doesn't.
So humans, you know over youknow, all the years of people
running grids, humans havegotten pretty decent about
knowing when peaks are going tohappen or sort of the general
load shape tomorrow.
You know thinking about theweather and all this stuff.
But that's, you know, thinkingabout the weather and all this
stuff, but that's, you know,typically stored in some human's
(11:08):
mind and it's more of anintuition about what's going to
happen.
Of course, people have donesome more rudimentary style
modeling using regression andthings like that, but machine
learning really takes it up anotch and allows us to get a lot
more accurate and performpredictions that are much longer
term than would otherwise beable to be done in an accurate
(11:31):
fashion.
But data is key.
Jen Szaro, AESP (11:33):
Data is key, absolutely All right.
Well, now that we have thatunderstanding, let's start high
level and then kind of get moregranular into the applications
of AI that you're seeing wholepromise, especially in the area
of demand flexibility andvirtual power plants.
So, as I alluded to at the topof the podcast, we've been now
over 150 years of pop culturepainting a captivating tale of
(11:57):
gloom and doom If AI takes over.
We've all seen the Terminator,at least most of us.
If you hear a lot of buzz thatAI is going to make this or that
job obsolete, what is thefuture role of AI going to be in
day-to-day work for the averageenergy professional?
You know I see it as a toolpersonally, more so than
(12:18):
something that's going toreplace me.
Bill Burke, Virtual Peaker (12:21):
Yeah , I see it the same way.
Honestly, jen, I think that,having actually never seen
Terminator, you know, so maybeI'm not as doom and gloom about
AI as a lot of other people inthe world.
There's a long backstory aboutme not being able to watch
R-rated movies when I was a kid,so anyway, yeah, it's a real
(12:42):
shame, but what I see for ourindustry is AI helping in, you
know, forecasting load and thenbeing able to then control
(13:08):
demand in order to meet thatload in the future.
That's a key component of it.
A few other things that I'dlike to mention there for sure
is providing great customersupport.
You know, having utility,having utility customer
engagement.
Technologies that help peoplein the customer support center
(13:31):
provide great customer supportto the end-use customers.
Those utilities are notparticularly well-known for
having great customer support,but if we can get good AI
technologies to help increasethat customer support, it can
really help across the boardwith utility customer engagement
and making customers happier.
(13:51):
So I see that as one area ofinterest with AI.
I think that using AI to helppredict places in the grid that
are going to be constrained isanother promising application of
AI technology.
So as EVs and as PV grow in thegrid, it puts strains on the
(14:16):
distribution system, strainsthat have never been seen before
, and if we can use AI to reallypredict those ahead of time,
potentially years in advance.
I've seen some interesting toolsfor doing that which will help
the utilities get ahead of thedistribution system build out,
help them, you know, deploytechnologies to really handle
that.
And then, of course, you know,what we're doing today is really
(14:39):
trying to make that demand ascontrollable as a gas turbine
generator.
That's, you know's, what we'vebeen focused on from the
beginning, and using AI topredict load, both in aggregate
and individual device level load, is critically important to
that.
So having a good model of theworld really smooths that, and
we're deploying this newtechnology that we call top-line
(15:01):
demand control.
It's a new technology categorythat marries artificial
intelligence, look-aheadoptimization.
It uses digital twin technologyfor modeling individual devices
and then gives the utility theability to have dependable and
precise control over that totalload shape.
So marry that with predictionof future load and you now have
(15:27):
a great tool to use instead ofgas speaker blinds.
Jen Szaro, AESP (15:31):
You know I'm completely sold on this approach
.
I think it's so long overdue tobe able to pick up on these
patterns and really understandhow customers are going to
behave or how loads are going tobehave, and all of us have been
stuck in the dreaded IVR cycles.
You know they're no fun, so Ilove the idea of being able to
(15:51):
apply AI in that way as well, tomake the customer experience
more enjoyable and moreefficient.
Right, we've all got our dayjobs, we all have things to do,
and no one wants to spend timeon the phone waiting to get a
response.
So you know, I think most folkswho are in the industry now are
growing comfortable with theidea of communications and
(16:15):
controls being handled bymachine learning.
But how do you really get yourutilities and customers to trust
this technology?
You know what kind of pushbackare you seeing so far and how
have you been able to respond tothat?
Bill Burke, Virtual Peaker (16:30):
Yeah , so I think for customers like,
if you think about theutilities customers this sort of
technology should be relativelytransparent to them.
The customer interactionsshould seem natural.
The way their devices arebehaving if they're in an
(16:52):
AI-controlled load managementprogram, the way their devices
are behaving should seem naturaland really that customer
experience should be preserved.
That should be the goal ofusing AI in any of these
technologies is to effectivelymake it transparent for the
customer and just make themdelighted by what the AI is
(17:12):
doing for them.
And we're used to that with ourphones, we're used to that with
lots of differentrecommendation engines from
Netflix or whoever.
And providing a great customerexperience should be the key
thing.
And if we can do that, then Ithink the utilities customers
(17:33):
are going to be happy to adoptthese sorts of technologies.
I think that the utilitiescustomers are going to be happy
to adopt these sorts oftechnologies.
I think that getting theutility behind the concept of
using AI is something that isgoing to be challenging in some
respects, but at the same time,there's a real desire to adopt
(17:53):
this sort of technology atcertain points in the
organization, a real desire toadopt this sort of technology at
certain points in theorganization.
So you know, the innovationcenters at utilities are really
excited about AI and we getreached out to quite a lot to
just talk to utilities, aboutour use of AI.
So that's fantastic.
Where it gets harder to push inAI tools is when you talk about,
or when you get to, the reallyoperational components of
(18:17):
utilities.
So when you think aboutdistribution operations or you
think about power purchasing,those sorts of areas, are
they're so critical to literallykeeping the lights on that?
There's a heart.
There's a higher barrier ofentry into that for AI, because
the AI someone, has to proveitself.
(18:39):
The AI has to operate similarlyto the way they're used to
operating and it has to betransparent so that they can
understand what's happening.
And so that's really where thebarrier is, is really getting
down to the nuts and bolts ofhow do we keep the lights on,
and we think frankly, we thinkour top line demand control
(18:59):
technology creates the metaphorsbetween, or the analogies, if
you will, between, generationand demand.
That will help all those folksreally understand and then be
able to operate it.
Of course, they'll have toprove it to themselves that it
works, but we're pretty excitedabout that step in and of itself
.
Jen Szaro, AESP (19:20):
I think you're spot on with that.
I think it is the operationsfolks.
They're the ones responsiblefor the reliability scores of
the utility right and get theblowback if things don't go well
.
So you know, I guess it'sunderstandable that they would
be a little reticent to jumpinto something new and change
management.
I think within the utilitysector has always been a
challenge, at least from when Iworked there, and it's hard,
(19:43):
especially when you don't, Ithink at this point, have
necessarily the right regulatoryconstructs in place to help
mold utilities and allow them tochange and be a little bit more
flexible.
Are there areas of the countrythat you see are really ready
for this, more so than others?
Bill Burke, Virtual Peaker (20:01):
Yeah , definitely.
You know there's places withmuch higher cost electricity
than others, and California, newEngland, places like that are,
I think, typically more apt totake on these technologies
because they see the price goingup and they want to do
something about that price andso they're willing to look at
(20:22):
these efficiencies and look attechnologies to improve the sort
of overall system efficiency inorder to keep costs down.
But even utilities throughoutthe United States, even places
that have relatively low cost ofenergy, are recognizing that
we're moving into ademand-constrained paradigm, how
(20:51):
power ramps because of demandover the course of the day is
going to create a situationwhere you don't have enough
instantaneous power, if you will, you don't have enough capacity
in order to meet the needs, andthat's going to be the cost
driver over the energy, the baseenergy need, and even places
(21:11):
with, like I said, even placeswith low energy costs are
starting to recognize that.
So we see interest basicallyall across North America today,
which is where we're operating,in these sorts of tools.
Now some of that capacityconstraint, if you will, for
some utilities is much fartheroff than it is for places like
(21:32):
California and Texas and NewEngland, but even in spite of
that they recognize that it'scoming.
Jen Szaro, AESP (21:41):
And I know that there are a lot of places in
the US that are really heavilyinvesting in renewables and wind
and solar we know haveintermittency issues, and so I
would think that that would alsobe a huge driver for in making
these investments to help sortof fill in those gaps.
Are you seeing that where Iknow I've heard here in Florida
there's a lot of investmenthappening in utility scale solar
(22:02):
.
They just haven't started.
Really, I don't think thinkingabout filling in the gaps yet.
So are you seeing other places,other regions where that's
happening, where they're likeyep, we've got days where we
have 60% plus penetration and weneed to fill the gaps in?
Bill Burke, Virtual Peaker (22:20):
Yeah , definitely.
We see that a lot of utilitiesin places with a lot high
renewable penetration arebeginning to think about both
reducing the peak and consumingmore energy when renewables are
producing.
So it really goes bothdirections, because they both
(22:42):
create ramp rate issues, andwhen I say ramp rate I mean how
quickly the apparent system loadchanges really defines how they
turn on generators and howquickly they turn on generators
and how quickly those generatorsneed to increase their power
output.
And so if you think about thesomewhat nightmare scenario of
(23:04):
everybody driving EVs cominghome from work around five
o'clock and plugging in this isabout the same time that the sun
is starting to go down you seea massive increase in demand at
the same time, a massivedecrease in power output from
the solar panels.
This creates the scary scenario.
So how utilities are absolutelythinking about, how do we
(23:26):
charge cars in the middle of theday when the solar is going,
and how do we then defer whenthey turn on or when they start
charging in the afternoon whenpeople come home.
So these are all things that AIand the sort of technologies
that we're working on can helputilities with.
Jen Szaro, AESP (23:43):
Yeah, absolutely.
As an EV driver myself, I'mhypersensitive to it being in
the industry, but all myneighbors are starting to get
EVs now too, on the sametransformer.
So it's going to be interesting, with that sort of clustering
effect, to see what happens ifthey don't start to embrace
things like demand flexibility.
Bill Burke, Virtual Peaker (24:05):
So you know, that's the sort of
scenario where utilities areparticularly nervous.
A single service transformerwith multiple EVs.
It takes a long time to get newtransformers.
Overloading them reduces theirlife, causes them to blow up.
Blow up probably not the rightword, but to fail more quickly
(24:25):
and using demand flexibility,you can really shift that around
and save those transformers ifyou can control when the EVs
charge, and that's the sort oftechnology we work on Absolutely
.
Jen Szaro, AESP (24:38):
So tell me, once we get past the trust
issues, what are the other areasof promise that you see for AI
in our industry and what aresome things you think AI
probably would not be able tosolve for us in our industry.
Bill Burke, Virtual Peaker (24:52):
Yeah , that's a great question.
So you know we've talked alittle bit about the customer
engagement side of it.
You mentioned the long IVRqueues, which is painful.
Ai can help with those sorts ofthings.
They can help with the customerinteraction, they can help
write and copy, they can helpwith just knowledge right, they,
they?
There's a lot of complexquestions that customers ask and
(25:13):
if the AI can source thecorrect answer, that's super
helpful.
Energy efficiency is another oneUnderstanding how consumers use
energy, recommending energysaving measures.
These can be applicable forboth consumers and utilities.
I think energy efficiency andAI have a lot of promise, for
(25:34):
sure.
And then, of course, gridmanagement, which is what we're
doing.
Ai can optimize the balancebetween supply and demand,
enhance the efficiency of thegrid operation centers again, if
we get past the trust issuesand then some challenges that AI
might not be able to solve,Obviously, at least today and in
(25:59):
the near future, AI can't buildinfrastructure right.
It can't help us modernize theinfrastructure by actually
physically building these things.
It can help predict where weneed to make those
infrastructure modernizations.
So it's a great promise there,but the actual building that's
all humans today and forprobably quite a long time.
Jen Szaro, AESP (26:25):
Yeah, I think so.
I also see my line peoplefixing the lines as they go down
as well.
So maybe, to your point, we canget there faster and find the
faults more quickly, but I thinkwe do still need people in this
industry for sure.
Bill Burke, Virtual (26:41):
Absolutely .
It can't solve the regulatoryand policy issues because it
can't really today navigate thecomplex regulatory environment.
It can't really create policies.
What AI can do is help peopleunderstand the policies more
quickly.
I recently heard a utilityexecutive talking about feeding
(27:04):
AI a policy document and gettinga quick summary of it in order
to understand the high points.
So AI has promise in helping usunderstand these complex issues,
but it doesn't have the abilityto actually create them yet.
That's still going to requirehumans and there's a lot of
human decisions that have to bemade because there's tradeoffs
between cost and people andequity and all this other stuff
(27:27):
that AI is not there for yet.
Jen Szaro, AESP (27:30):
Yeah, I mean I think that idea of kind of
poring over regulatory filingsand historical filings to try to
understand what a greatapplication for sure.
I mean, that's such a painpoint I think our regulatory
bodies have to face.
Bill Burke, Virtual Peaker (27:45):
Yeah , it's a massive challenge, even
for us as a company selling toutilities, understanding where
the policies are coming out thatbenefit us or don't benefit us.
Just understanding that in aquick way, because there's a lot
of different ways utilities areregulated across North America.
Without a doubt, At least 50different ways probably.
Jen Szaro, AESP (28:06):
At least, and then some right.
Bill Burke, Virtual Peaker (28:08):
Yeah for sure.
And then it can't, today, solvethe cybersecurity threats.
It can help us with threatdetection.
It can help us, in certainscenarios, understand what's
happening and maybe even predictwhat might happen in the future
.
But cybersecurity threats are alittle bit like war.
(28:30):
Right, there's a human wagingit on one side and you really
need another human to match witswith that person.
So it's not quite there yet,but it's definitely still
helpful.
Jen Szaro, AESP (28:41):
I've got a son who's working in that space and
he's like yeah, mom, I'm stillvery much needed.
So that made me comforted.
Considered to be paid for hiscollege education.
Bill Burke, Virtual Peaker (28:51):
Yeah , absolutely.
I think that cybersecurityissue is going to be powered by
humans for quite some time tocome.
Jen Szaro, AESP (29:00):
So it's been such a blur.
So much has happened in thelast three to five years in our
industry.
What do you think the state ofadoption for AI is going to look
like over the next five years?
Bill Burke, Virtual Peaker (29:13):
It's really hard to know exactly
what's going to happen in thefuture.
Right, there's no AI crystalball just yet.
If there were, I would probablynot be here, yeah.
(29:52):
So in the next five years, theadoption of AI to facilitate key
objectives across manydifferent spheres.
Right In the utility sectorspecifically, we're thinking
about how to managetransportation.
Electrification this is a bigconcern and, like I've said a
few times already, predictingwhere distribution system
constraints are is going to be abig part of it.
(30:12):
We're going to enhance supplyand demand predictions as
renewable energy adoptionincreases.
Ai has a huge capability thereand it's already begun taking
hold at utilities today.
That prediction is going toaccelerate.
We're really going to be focusedon shifting peak consumptions
(30:32):
using AI in order to delay andeven avoid infrastructure
investments.
This technology that will allowus to control demand and
predict demand very accurately,you know hosts a lot of
different benefits and you knowsome of it is reducing that peak
(30:53):
sort of globally.
But really getting down intothe distribution system and
adjusting demand at verygranular levels allows utilities
to save a lot of money oninfrastructure upgrades.
Delay those infrastructureupgrades.
They'll eventually have to dothem, but hopefully they can do
(31:13):
them at a time that's most costeffective for them and most
beneficial to the customers.
Jen Szaro, AESP (31:19):
Yeah, just allowing for the time to plan
for those in a meaningful way, Ithink, rather than having to be
reactive, is where I think it'sgoing to really help us.
I mean, we all know how long ittakes to build the average
natural gas plant.
We know what it's like to tryto build a transmission and get
(31:39):
that cited and permitted.
These are things that take areally long time.
So if we can delay those in ameaningful way with demand
flexibility and virtual powerplants, I think we've got a
chance to kind of ensure thatwe're getting it right and
making the best use of our ratepayer dollars too.
Bill Burke, Virtual Peaker (31:56):
I agree entirely.
Jen Szaro, AESP (31:59):
So what should utilities and energy users be
doing now to sort of be readyfor AI entering the landscape in
a meaningful way?
Is there anything like I know?
I was just out with the folksfrom the Flex Lab over at
Lawrence Berkeley Labs.
Wow, that's an impressive groupout there.
I see all this work happeningwith grid interactive efficient
(32:22):
buildings.
That's really exciting.
I see so much happening in theinteractive efficient buildings.
That's really exciting.
I see so much happening in thehome automation space.
How do we tap into this?
How do we support that in ameaningful way?
Bill Burke, Virtual Peaker (32:34):
Yeah , that's a great question.
So I mentioned data is reallykey and utilities upgrading
their systems to collect and beable to analyze that data, I
think is the key first step.
So if the utility doesn't havesmart meters, they need them
yesterday.
(32:54):
They need good telemetry off ofall their infrastructure.
They need to start storing thatdata.
They need to work onaccelerating getting that data
into a computer right, reducingthe time lag between when the
data is collected and when itgets in, because AI doesn't work
(33:14):
without data in most situations.
So the more data they can get,the better.
And that goes across the board,across their infrastructure in
every way, across the board,across their infrastructure in
every way.
And then the second.
So that's sort of primary.
The second step is reallystarting to think about that
change management inside theutility.
How do we get everybody onboard with being very
(33:36):
analytically focused andthinking about adoption of AI
technologies where it makes themost sense?
Change management is hard.
It's hard for mostorganizations and having the
leadership in place and theleadership helping with that
change management, I think isthe second massive step that
needs to be taken.
Jen Szaro, AESP (33:57):
Yeah, I could not agree more.
I think that level ofleadership to take the fear out
of change management and helppeople see the benefits can
really make or break theutilities.
Adoption right now of gridmodernization technology
specifically.
Bill Burke, Virtual (34:13):
Absolutely .
Jen Szaro, AESP (34:15):
And going back to the customer, the user, end
user.
For me, this is always one ofthe biggest challenges and the
hardest pieces of the puzzle tomake fit, because it has to be a
win-win for them as well.
Right, they have to understandwhat's in it for them to
participate in these kinds ofprograms, and so what sort of
(34:37):
communication do you seeutilities really needing to do
with their partners, to ensurethat you know this isn't just
something being driven by theutility, but also something
that's really going to benefitthe customer partners in all of
this?
Bill Burke, Virtual Peaker (34:50):
Yeah , that's a great point,
Absolutely.
I think in many places we haveto recognize that the customers,
the end-to-use customers,aren't adopting distributed
energy resources, things likesmart thermostats and whatnot
for utility benefit.
They're adopting them forthemselves.
They want this technology,right.
(35:12):
They want to use a heat pump,water heater because it lowers
their electricity bill or lowerstheir overall energy bill.
They want to use EVs becausethey're fond of drive and they
love the cars that they'rebuying.
They're not buying them forutility benefit, right.
They're buying them for theirown transportation.
And so making sure thatcustomers understand the benefit
(35:35):
of adopting these sorts ofprograms is critical, and that
is through good customerengagement technologies, good
customer engagement mechanisms.
I should say not technologiesTechnology definitely has a role
to play in that, but helpingthose customers understand we're
able to reduce overall energycosts because you're adopting
(35:57):
these technologies.
How much are we able to reduceit?
I think Fremont Power has beena fantastic example of that.
I think almost annually theystate publicly how much money
(36:20):
their virtual power plant hassaved their rate payers, which
is huge for customers tounderstand what the benefit is
and really adopt this sort oftechnology.
But a lot of customers areadopting the technologies
because they want it and they'reactually going to their utility
.
We see this in a lot of co-opsand we hear this from a lot of
our cooperative customers.
They say we want to stand up aprogram because our customers
are asking us to.
So a lot of customersunderstand that by doing these
sorts of things they're going tohelp save the world effectively
(36:41):
.
But that doesn't get you out oftelling them how much going to
help save the world effectively,but that doesn't get you out of
telling them how much they'rehelping save the world as well.
Jen Szaro, AESP (36:50):
Absolutely.
I think you need to remindpeople of the impact that they
could have themselves and then,if we can, as an industry,
empower that behavior.
I mean I think that's thewin-win that I see for us.
Absolutely, Me too.
Bill Burke, Virtual Peaker (37:03):
Me too.
Jen Szaro, AESP (37:04):
Well, thank you , bill.
This has been a reallyincredible conversation.
I'm just so intrigued by allthe work that you and your team
are doing right now at VirtualPeeker, and I cannot wait to see
what comes next with AI overthe next years, and so thanks
for spending time with us todayon the Energy Beat podcast.
Bill Burke, Virtual Peaker (37:24):
Jen, this has been fantastic.
I really enjoyed it as well.
Wide-ranging discussion, wewere a bit all over the place,
but hopefully that's the goodpart.
Yeah, I think so too.
I think hopefully we helpedeverybody who's listening
understand AI a little bit morehow it could help utilities
control demand better, helputilities engage customers
better.
There's a lot of excitement inthe world.
(37:45):
Thanks a lot for having me.
Jen Szaro, AESP (37:47):
Thank you, and if you'd like to learn more
about Bill and his company, justhead over to virtualpeakercom
and check them out.
Thank you so much for beingwith us today.
We'll see you next time.
Welcome back to the Energy Beat podcast.
Picture it.
New Zealand 1863, author SamuelButler released an article
Darwin Among the Machine whichis credited by many to be the
beginning of the pop culturetrope of artificial intelligence
we know today to be thebeginning of the pop culture
trope of artificial intelligencewe know today.
Fortunately for us, butler'svision of us creating our own
(00:28):
successors has not yet come topass.
In fact, today AI shows thepromise of helping us solve some
of our greatest generationalchallenges, and without the
threat of global annihilation.
And today we are joined by DrBill Burke, ceo of Virtual
Peeker, to talk about how AIwill help us tackle some of our
(00:50):
greatest contemporary energychallenges.
Welcome, bill.
Bill Burke, Virtual Peake (00:54):
Hello Jen.
Thanks so much for having meback on the Energy Beat podcast.
I'm excited to be here.
Jen Szaro, AESP (00:59):
We're excited to have you.
You're doing such great stuffwith your organization.
I can't wait to dive in alittle deeper.
So before we do that, tell us alittle bit about yourself and
tell the audience all aboutVirtual Peeker.
Bill Burke, Virtual Peaker (01:10):
Yeah , thanks a lot, Jen.
I'm Bill Burke.
I'm the founder and CEO ofVirtual Peeker.
I started the company back in2015,.
Really with the key vision ofhow do we make demand as
controllable as a gas turbinegenerator, and today we're
working to really revolutionizehow we manage demand on the
electricity grid, making it moreefficient, more sustainable and
(01:33):
ensuring a prosperous andsustainable legacy for future
generations.
Jen Szaro, AESP (01:39):
So tell us.
How did you come up with thisidea for the company?
What drove you to launch suchan amazing and innovative
approach?
Bill Burke, Virtual Peaker (01:47):
Yeah , thanks for that, jen.
So I've been in this space fora very long time.
I started back in 2006 workingon a project, doing my graduate
work at UC Berkeley, working ona project to create the first
programmable communicatingthermostat project to create the
first programmablecommunicating thermostat and
ever since then I've beenfocused on this area of how do
(02:08):
we help utilities get behind themeter and control devices.
I focused my dissertationresearch on it.
I worked on that at GEAppliances, really applying this
idea of getting behind themeter to help utilities control
load across the entire home.
And then this was very earlydays of Internet of Things, you
(02:29):
know, I think, 2010.
So Nest Thermostat hadn't evenentered the market yet, and so
Internet-connected products werevery new.
Spent some time there at GE andthen left to start Virtual
Peeker, really with this bigidea of making demand as
controllable as a gas turbinegenerator by helping utilities
(02:49):
control those internet-connectedproducts.
Things like thermostats andwater heaters and batteries and
EVs and that big vision ofdemand as controllable as a gas
turbine generator has taken sometime to really come to fruition
, but with our latest technologydeployments that include, you
know, some innovative AIapproaches, some innovative
optimization approaches, wethink we're getting very close
(03:13):
to really bringing this toreality.
I'm pretty excited about it.
Actually, it's something I'vebeen working on literally since
2006.
Jen Szaro, AESP (03:21):
So I'm not going to lie, you're kind of
fulfilling my college dreams bywhat you're doing.
I remember starting out as aphysical chemist in this
industry and working in thephotovoltaic side of things and
watching the changes happen intechnology and it was never fast
enough for me and I alwaysenvisioned this whole idea of
distributed generation, kind ofreplacing power plants.
(03:44):
And now you're making thathappen.
Did you have any doubts whenyou're going through that
process of building the companyand how did you push past that?
Bill Burke, Virtual Peaker (03:54):
Yeah , yeah, starting a company is
not for the faint of heart, forsure.
Of course, doubts along the way.
So why do people care aboutthis concept of demand is
controllable as a gasregenerator.
It really comes back tophotovoltaics, which you were
working on as a physical chemist, right?
The wind doesn't blow all thetime, the sun doesn't shine all
(04:16):
the time, and so renewableenergy is intermittent.
It's intermittent in nature,and keeping the grid in balance
is something that has to be donein nature, and keeping the grid
in balance is something thathas to be done, and today that's
primarily being done by usinggas peaker plants or some sort
of fossil fuel energy source.
And so if we can turn thedemand side of it into a
(04:37):
controllable resource thatutilities can utilize to help
balance the grid, which is whatwe're working on, we think that
we smooth the path toward atotally green grid.
That's the reason we're reallydoing all this, and it's taken
some time.
Everything has taken some timein this energy transition, but
we're pretty excited to be, Ithink, at the forefront of
(04:59):
making this a reality.
Jen Szaro, AESP (05:01):
Yeah, some of your projects are just really
outstanding in this reality.
Yeah, some of your projects arejust really outstanding.
So, as we move forward with thelatest and greatest in
innovation in this area ofmanaging demand and using it to
kind of fill in those peaks andvalleys, we've got the advent of
AI.
It's relatively new, fairlywell misunderstood, I think, by
(05:22):
any non-technical crowd and eventhose who are in some technical
spaces, and the terminology is,you know, very confusing at
times.
I think so for our listeners,can you quickly spell out what
you see as the relationshipbetween AI and machine learning,
just for those who might notunderstand that terminology and
(05:42):
what the differences are betweenthe two or how they're
technically, how they're relatedand kind of what that is going
to mean for our industry?
Bill Burke, Virtual Peaker (05:51):
Yeah , yeah, yeah, absolutely.
I'd be happy to talk about that.
So I would first maybe pushback a little bit on the concept
of AI being new.
I think it's a relativelyrecent entrant into the
zeitgeist, but you know, ai as afield has been going strong for
many, many years I think Idon't know the full history, but
dating back at least to the 60sand before and there's been
(06:13):
lots of different types ofartificial intelligence.
So what is what is AI reallymean?
It means teaching computers howto reason about the world and
make decisions sort of on theirown is a large part of it, and
so that's sort of the broadconcept, and it's had many
different forms over the years,with tons of subdisciplines.
(06:34):
Things like fuzzy logic was abig one at one point.
Expert machines is anotherpiece, and AI is composed of all
of these different ideas.
I think machine learning is amuch newer, maybe newer entrant
into that sort of overall bucketof what AI is, and machine
(06:57):
learning is really primarilyabout feeding the computer lots
of data and letting the computerlearn on its own what that data
means, and oftentimes machinelearning results in models.
It creates models that can thenbe used to do inference on the
world.
So prediction, if you will.
(07:17):
And those models are typicallynot human, understandable,
meaning that we don't know howthe machine necessarily came up
with that decision.
It's not.
It's not reasoned about in thesame way that we reason about
things necessarily.
But that's the core idea ofmachine learning and, again,
machine learning is a subset ofartificial intelligence.
Machine learning typically usesthings called neural networks.
(07:41):
Things called neural networksand the AI chatbots are sort of
a massive version of thattechnology.
If you will, implemented oninstead of data, like virtual
peakers using, like, which aretypically numbers, but these AI
chatbots are implemented on,obviously, words and language.
But it's all stem from the samecore technology idea of feeding
(08:05):
the machine lots of data andletting the machine come up with
, effectively, a model of theworld.
Jen Szaro, AESP (08:12):
Right.
So I mean, I think today mostpeople think about AI, machine
learning, and they think aboutchat, gpt, right, because that's
pervasive.
Everyone's kind of seen it andplayed with it.
I've used it myself to craft afew outlines and come up with
some snazzy session descriptions, but those are generally
different kinds of models thanyou would probably be using in
(08:33):
our industry, right?
Because those are more focusedon gathering knowledge from
what's out there in the internet.
So they're what they consideropen models.
Is that correct?
Bill Burke, Virtual Peaker (08:42):
Yeah .
So I think that those modelsare typically being trained on
like I said a couple of minutesago, being trained on human
language, and they're using thewords written on websites and
books and whatnot in order tocreate a model for how people
talk and how people think.
Now that sort of technology issomewhat applicable to our field
(09:07):
, but it's generally moreapplicable in the form of
customer engagement sorts oftechnologies.
How do you make it easier forusing something like ChatGPT,
like you said, creating outlines?
How could you use it to helpease the support burden when
you're handling a support issuewith a customer?
(09:28):
Or how can you use orpotentially using that sort of
technology to write marketingcopy that you would then send
out to customers in order tobuild support for whatever
you're doing?
So that's definitely still used.
What we do is somewhat differentin that we use the same sort of
general ideas of neuralnetworks and then feed it
(09:51):
numbers typically right.
We feed it data streams thatwe've captured or the utilities
captured and learn about theworld through numbers, and then
use those numbers to doprediction, and those
predictions usually take theform of some sort of maybe a
prediction on the energyconsumption tomorrow, maybe a
forecast for the total systemload over the next several days.
(10:14):
Maybe it's forecastingindividual device behavior over
the next several days, so it'sreally predictions on what
machines are going to do and howthey're going to consume energy
, and that's the way we're usingartificial intelligence or
machine learning today.
Jen Szaro, AESP (10:35):
Right, because you've got so much data coming
in.
I can't even imagine a humanbeing trying to make heads or
tails of all of that datawithout the assistance of
something like machine learningand AI.
Bill Burke, Virtual Peaker (10:47):
Yeah , I think that machine learning
what it doesn't.
So humans, you know over youknow, all the years of people
running grids, humans havegotten pretty decent about
knowing when peaks are going tohappen or sort of the general
load shape tomorrow.
You know thinking about theweather and all this stuff.
But that's, you know, thinkingabout the weather and all this
stuff, but that's, you know,typically stored in some human's
(11:08):
mind and it's more of anintuition about what's going to
happen.
Of course, people have donesome more rudimentary style
modeling using regression andthings like that, but machine
learning really takes it up anotch and allows us to get a lot
more accurate and performpredictions that are much longer
term than would otherwise beable to be done in an accurate
(11:31):
fashion.
But data is key.
Jen Szaro, AESP (11:33):
Data is key, absolutely All right.
Well, now that we have thatunderstanding, let's start high
level and then kind of get moregranular into the applications
of AI that you're seeing wholepromise, especially in the area
of demand flexibility andvirtual power plants.
So, as I alluded to at the topof the podcast, we've been now
over 150 years of pop culturepainting a captivating tale of
(11:57):
gloom and doom If AI takes over.
We've all seen the Terminator,at least most of us.
If you hear a lot of buzz thatAI is going to make this or that
job obsolete, what is thefuture role of AI going to be in
day-to-day work for the averageenergy professional?
You know I see it as a toolpersonally, more so than
(12:18):
something that's going toreplace me.
Bill Burke, Virtual Peaker (12:21):
Yeah , I see it the same way.
Honestly, jen, I think that,having actually never seen
Terminator, you know, so maybeI'm not as doom and gloom about
AI as a lot of other people inthe world.
There's a long backstory aboutme not being able to watch
R-rated movies when I was a kid,so anyway, yeah, it's a real
(12:42):
shame, but what I see for ourindustry is AI helping in, you
know, forecasting load and thenbeing able to then control
(13:08):
demand in order to meet thatload in the future.
That's a key component of it.
A few other things that I'dlike to mention there for sure
is providing great customersupport.
You know, having utility,having utility customer
engagement.
Technologies that help peoplein the customer support center
(13:31):
provide great customer supportto the end-use customers.
Those utilities are notparticularly well-known for
having great customer support,but if we can get good AI
technologies to help increasethat customer support, it can
really help across the boardwith utility customer engagement
and making customers happier.
(13:51):
So I see that as one area ofinterest with AI.
I think that using AI to helppredict places in the grid that
are going to be constrained isanother promising application of
AI technology.
So as EVs and as PV grow in thegrid, it puts strains on the
(14:16):
distribution system, strainsthat have never been seen before
, and if we can use AI to reallypredict those ahead of time,
potentially years in advance.
I've seen some interesting toolsfor doing that which will help
the utilities get ahead of thedistribution system build out,
help them, you know, deploytechnologies to really handle
that.
And then, of course, you know,what we're doing today is really
(14:39):
trying to make that demand ascontrollable as a gas turbine
generator.
That's, you know's, what we'vebeen focused on from the
beginning, and using AI topredict load, both in aggregate
and individual device level load, is critically important to
that.
So having a good model of theworld really smooths that, and
we're deploying this newtechnology that we call top-line
(15:01):
demand control.
It's a new technology categorythat marries artificial
intelligence, look-aheadoptimization.
It uses digital twin technologyfor modeling individual devices
and then gives the utility theability to have dependable and
precise control over that totalload shape.
So marry that with predictionof future load and you now have
(15:27):
a great tool to use instead ofgas speaker blinds.
Jen Szaro, AESP (15:31):
You know I'm completely sold on this approach
.
I think it's so long overdue tobe able to pick up on these
patterns and really understandhow customers are going to
behave or how loads are going tobehave, and all of us have been
stuck in the dreaded IVR cycles.
You know they're no fun, so Ilove the idea of being able to
(15:51):
apply AI in that way as well, tomake the customer experience
more enjoyable and moreefficient.
Right, we've all got our dayjobs, we all have things to do,
and no one wants to spend timeon the phone waiting to get a
response.
So you know, I think most folkswho are in the industry now are
growing comfortable with theidea of communications and
(16:15):
controls being handled bymachine learning.
But how do you really get yourutilities and customers to trust
this technology?
You know what kind of pushbackare you seeing so far and how
have you been able to respond tothat?
Bill Burke, Virtual Peaker (16:30):
Yeah , so I think for customers like,
if you think about theutilities customers this sort of
technology should be relativelytransparent to them.
The customer interactionsshould seem natural.
The way their devices arebehaving if they're in an
(16:52):
AI-controlled load managementprogram, the way their devices
are behaving should seem naturaland really that customer
experience should be preserved.
That should be the goal ofusing AI in any of these
technologies is to effectivelymake it transparent for the
customer and just make themdelighted by what the AI is
(17:12):
doing for them.
And we're used to that with ourphones, we're used to that with
lots of differentrecommendation engines from
Netflix or whoever.
And providing a great customerexperience should be the key
thing.
And if we can do that, then Ithink the utilities customers
(17:33):
are going to be happy to adoptthese sorts of technologies.
I think that the utilitiescustomers are going to be happy
to adopt these sorts oftechnologies.
I think that getting theutility behind the concept of
using AI is something that isgoing to be challenging in some
respects, but at the same time,there's a real desire to adopt
(17:53):
this sort of technology atcertain points in the
organization, a real desire toadopt this sort of technology at
certain points in theorganization.
So you know, the innovationcenters at utilities are really
excited about AI and we getreached out to quite a lot to
just talk to utilities, aboutour use of AI.
So that's fantastic.
Where it gets harder to push inAI tools is when you talk about,
or when you get to, the reallyoperational components of
(18:17):
utilities.
So when you think aboutdistribution operations or you
think about power purchasing,those sorts of areas, are
they're so critical to literallykeeping the lights on that?
There's a heart.
There's a higher barrier ofentry into that for AI, because
the AI someone, has to proveitself.
(18:39):
The AI has to operate similarlyto the way they're used to
operating and it has to betransparent so that they can
understand what's happening.
And so that's really where thebarrier is, is really getting
down to the nuts and bolts ofhow do we keep the lights on,
and we think frankly, we thinkour top line demand control
(18:59):
technology creates the metaphorsbetween, or the analogies, if
you will, between, generationand demand.
That will help all those folksreally understand and then be
able to operate it.
Of course, they'll have toprove it to themselves that it
works, but we're pretty excitedabout that step in and of itself
.
Jen Szaro, AESP (19:20):
I think you're spot on with that.
I think it is the operationsfolks.
They're the ones responsiblefor the reliability scores of
the utility right and get theblowback if things don't go well
.
So you know, I guess it'sunderstandable that they would
be a little reticent to jumpinto something new and change
management.
I think within the utilitysector has always been a
challenge, at least from when Iworked there, and it's hard,
(19:43):
especially when you don't, Ithink at this point, have
necessarily the right regulatoryconstructs in place to help
mold utilities and allow them tochange and be a little bit more
flexible.
Are there areas of the countrythat you see are really ready
for this, more so than others?
Bill Burke, Virtual Peaker (20:01):
Yeah , definitely.
You know there's places withmuch higher cost electricity
than others, and California, newEngland, places like that are,
I think, typically more apt totake on these technologies
because they see the price goingup and they want to do
something about that price andso they're willing to look at
(20:22):
these efficiencies and look attechnologies to improve the sort
of overall system efficiency inorder to keep costs down.
But even utilities throughoutthe United States, even places
that have relatively low cost ofenergy, are recognizing that
we're moving into ademand-constrained paradigm, how
(20:51):
power ramps because of demandover the course of the day is
going to create a situationwhere you don't have enough
instantaneous power, if you will, you don't have enough capacity
in order to meet the needs, andthat's going to be the cost
driver over the energy, the baseenergy need, and even places
(21:11):
with, like I said, even placeswith low energy costs are
starting to recognize that.
So we see interest basicallyall across North America today,
which is where we're operating,in these sorts of tools.
Now some of that capacityconstraint, if you will, for
some utilities is much fartheroff than it is for places like
(21:32):
California and Texas and NewEngland, but even in spite of
that they recognize that it'scoming.
Jen Szaro, AESP (21:41):
And I know that there are a lot of places in
the US that are really heavilyinvesting in renewables and wind
and solar we know haveintermittency issues, and so I
would think that that would alsobe a huge driver for in making
these investments to help sortof fill in those gaps.
Are you seeing that where Iknow I've heard here in Florida
there's a lot of investmenthappening in utility scale solar
(22:02):
.
They just haven't started.
Really, I don't think thinkingabout filling in the gaps yet.
So are you seeing other places,other regions where that's
happening, where they're likeyep, we've got days where we
have 60% plus penetration and weneed to fill the gaps in?
Bill Burke, Virtual Peaker (22:20):
Yeah , definitely.
We see that a lot of utilitiesin places with a lot high
renewable penetration arebeginning to think about both
reducing the peak and consumingmore energy when renewables are
producing.
So it really goes bothdirections, because they both
(22:42):
create ramp rate issues, andwhen I say ramp rate I mean how
quickly the apparent system loadchanges really defines how they
turn on generators and howquickly they turn on generators
and how quickly those generatorsneed to increase their power
output.
And so if you think about thesomewhat nightmare scenario of
(23:04):
everybody driving EVs cominghome from work around five
o'clock and plugging in this isabout the same time that the sun
is starting to go down you seea massive increase in demand at
the same time, a massivedecrease in power output from
the solar panels.
This creates the scary scenario.
So how utilities are absolutelythinking about, how do we
(23:26):
charge cars in the middle of theday when the solar is going,
and how do we then defer whenthey turn on or when they start
charging in the afternoon whenpeople come home.
So these are all things that AIand the sort of technologies
that we're working on can helputilities with.
Jen Szaro, AESP (23:43):
Yeah, absolutely.
As an EV driver myself, I'mhypersensitive to it being in
the industry, but all myneighbors are starting to get
EVs now too, on the sametransformer.
So it's going to be interesting, with that sort of clustering
effect, to see what happens ifthey don't start to embrace
things like demand flexibility.
Bill Burke, Virtual Peaker (24:05):
So you know, that's the sort of
scenario where utilities areparticularly nervous.
A single service transformerwith multiple EVs.
It takes a long time to get newtransformers.
Overloading them reduces theirlife, causes them to blow up.
Blow up probably not the rightword, but to fail more quickly
(24:25):
and using demand flexibility,you can really shift that around
and save those transformers ifyou can control when the EVs
charge, and that's the sort oftechnology we work on Absolutely
.
Jen Szaro, AESP (24:38):
So tell me, once we get past the trust
issues, what are the other areasof promise that you see for AI
in our industry and what aresome things you think AI
probably would not be able tosolve for us in our industry.
Bill Burke, Virtual Peaker (24:52):
Yeah , that's a great question.
So you know we've talked alittle bit about the customer
engagement side of it.
You mentioned the long IVRqueues, which is painful.
Ai can help with those sorts ofthings.
They can help with the customerinteraction, they can help
write and copy, they can helpwith just knowledge right, they,
they?
There's a lot of complexquestions that customers ask and
(25:13):
if the AI can source thecorrect answer, that's super
helpful.
Energy efficiency is another oneUnderstanding how consumers use
energy, recommending energysaving measures.
These can be applicable forboth consumers and utilities.
I think energy efficiency andAI have a lot of promise, for
(25:34):
sure.
And then, of course, gridmanagement, which is what we're
doing.
Ai can optimize the balancebetween supply and demand,
enhance the efficiency of thegrid operation centers again, if
we get past the trust issuesand then some challenges that AI
might not be able to solve,Obviously, at least today and in
(25:59):
the near future, AI can't buildinfrastructure right.
It can't help us modernize theinfrastructure by actually
physically building these things.
It can help predict where weneed to make those
infrastructure modernizations.
So it's a great promise there,but the actual building that's
all humans today and forprobably quite a long time.
Jen Szaro, AESP (26:25):
Yeah, I think so.
I also see my line peoplefixing the lines as they go down
as well.
So maybe, to your point, we canget there faster and find the
faults more quickly, but I thinkwe do still need people in this
industry for sure.
Bill Burke, Virtual (26:41):
Absolutely .
It can't solve the regulatoryand policy issues because it
can't really today navigate thecomplex regulatory environment.
It can't really create policies.
What AI can do is help peopleunderstand the policies more
quickly.
I recently heard a utilityexecutive talking about feeding
(27:04):
AI a policy document and gettinga quick summary of it in order
to understand the high points.
So AI has promise in helping usunderstand these complex issues,
but it doesn't have the abilityto actually create them yet.
That's still going to requirehumans and there's a lot of
human decisions that have to bemade because there's tradeoffs
between cost and people andequity and all this other stuff
(27:27):
that AI is not there for yet.
Jen Szaro, AESP (27:30):
Yeah, I mean I think that idea of kind of
poring over regulatory filingsand historical filings to try to
understand what a greatapplication for sure.
I mean, that's such a painpoint I think our regulatory
bodies have to face.
Bill Burke, Virtual Peaker (27:45):
Yeah , it's a massive challenge, even
for us as a company selling toutilities, understanding where
the policies are coming out thatbenefit us or don't benefit us.
Just understanding that in aquick way, because there's a lot
of different ways utilities areregulated across North America.
Without a doubt, At least 50different ways probably.
Jen Szaro, AESP (28:06):
At least, and then some right.
Bill Burke, Virtual Peaker (28:08):
Yeah for sure.
And then it can't, today, solvethe cybersecurity threats.
It can help us with threatdetection.
It can help us, in certainscenarios, understand what's
happening and maybe even predictwhat might happen in the future
.
But cybersecurity threats are alittle bit like war.
(28:30):
Right, there's a human wagingit on one side and you really
need another human to match witswith that person.
So it's not quite there yet,but it's definitely still
helpful.
Jen Szaro, AESP (28:41):
I've got a son who's working in that space and
he's like yeah, mom, I'm stillvery much needed.
So that made me comforted.
Considered to be paid for hiscollege education.
Bill Burke, Virtual Peaker (28:51):
Yeah , absolutely.
I think that cybersecurityissue is going to be powered by
humans for quite some time tocome.
Jen Szaro, AESP (29:00):
So it's been such a blur.
So much has happened in thelast three to five years in our
industry.
What do you think the state ofadoption for AI is going to look
like over the next five years?
Bill Burke, Virtual Peaker (29:13):
It's really hard to know exactly
what's going to happen in thefuture.
Right, there's no AI crystalball just yet.
If there were, I would probablynot be here, yeah.
(29:52):
So in the next five years, theadoption of AI to facilitate key
objectives across manydifferent spheres.
Right In the utility sectorspecifically, we're thinking
about how to managetransportation.
Electrification this is a bigconcern and, like I've said a
few times already, predictingwhere distribution system
constraints are is going to be abig part of it.
(30:12):
We're going to enhance supplyand demand predictions as
renewable energy adoptionincreases.
Ai has a huge capability thereand it's already begun taking
hold at utilities today.
That prediction is going toaccelerate.
We're really going to be focusedon shifting peak consumptions
(30:32):
using AI in order to delay andeven avoid infrastructure
investments.
This technology that will allowus to control demand and
predict demand very accurately,you know hosts a lot of
different benefits and you knowsome of it is reducing that peak
(30:53):
sort of globally.
But really getting down intothe distribution system and
adjusting demand at verygranular levels allows utilities
to save a lot of money oninfrastructure upgrades.
Delay those infrastructureupgrades.
They'll eventually have to dothem, but hopefully they can do
(31:13):
them at a time that's most costeffective for them and most
beneficial to the customers.
Jen Szaro, AESP (31:19):
Yeah, just allowing for the time to plan
for those in a meaningful way, Ithink, rather than having to be
reactive, is where I think it'sgoing to really help us.
I mean, we all know how long ittakes to build the average
natural gas plant.
We know what it's like to tryto build a transmission and get
(31:39):
that cited and permitted.
These are things that take areally long time.
So if we can delay those in ameaningful way with demand
flexibility and virtual powerplants, I think we've got a
chance to kind of ensure thatwe're getting it right and
making the best use of our ratepayer dollars too.
Bill Burke, Virtual Peaker (31:56):
I agree entirely.
Jen Szaro, AESP (31:59):
So what should utilities and energy users be
doing now to sort of be readyfor AI entering the landscape in
a meaningful way?
Is there anything like I know?
I was just out with the folksfrom the Flex Lab over at
Lawrence Berkeley Labs.
Wow, that's an impressive groupout there.
I see all this work happeningwith grid interactive efficient
(32:22):
buildings.
That's really exciting.
I see so much happening in theinteractive efficient buildings.
That's really exciting.
I see so much happening in thehome automation space.
How do we tap into this?
How do we support that in ameaningful way?
Bill Burke, Virtual Peaker (32:34):
Yeah , that's a great question.
So I mentioned data is reallykey and utilities upgrading
their systems to collect and beable to analyze that data, I
think is the key first step.
So if the utility doesn't havesmart meters, they need them
yesterday.
(32:54):
They need good telemetry off ofall their infrastructure.
They need to start storing thatdata.
They need to work onaccelerating getting that data
into a computer right, reducingthe time lag between when the
data is collected and when itgets in, because AI doesn't work
(33:14):
without data in most situations.
So the more data they can get,the better.
And that goes across the board,across their infrastructure in
every way, across the board,across their infrastructure in
every way.
And then the second.
So that's sort of primary.
The second step is reallystarting to think about that
change management inside theutility.
How do we get everybody onboard with being very
(33:36):
analytically focused andthinking about adoption of AI
technologies where it makes themost sense?
Change management is hard.
It's hard for mostorganizations and having the
leadership in place and theleadership helping with that
change management, I think isthe second massive step that
needs to be taken.
Jen Szaro, AESP (33:57):
Yeah, I could not agree more.
I think that level ofleadership to take the fear out
of change management and helppeople see the benefits can
really make or break theutilities.
Adoption right now of gridmodernization technology
specifically.
Bill Burke, Virtual (34:13):
Absolutely .
Jen Szaro, AESP (34:15):
And going back to the customer, the user, end
user.
For me, this is always one ofthe biggest challenges and the
hardest pieces of the puzzle tomake fit, because it has to be a
win-win for them as well.
Right, they have to understandwhat's in it for them to
participate in these kinds ofprograms, and so what sort of
(34:37):
communication do you seeutilities really needing to do
with their partners, to ensurethat you know this isn't just
something being driven by theutility, but also something
that's really going to benefitthe customer partners in all of
this?
Bill Burke, Virtual Peaker (34:50):
Yeah , that's a great point,
Absolutely.
I think in many places we haveto recognize that the customers,
the end-to-use customers,aren't adopting distributed
energy resources, things likesmart thermostats and whatnot
for utility benefit.
They're adopting them forthemselves.
They want this technology,right.
(35:12):
They want to use a heat pump,water heater because it lowers
their electricity bill or lowerstheir overall energy bill.
They want to use EVs becausethey're fond of drive and they
love the cars that they'rebuying.
They're not buying them forutility benefit, right.
They're buying them for theirown transportation.
And so making sure thatcustomers understand the benefit
(35:35):
of adopting these sorts ofprograms is critical, and that
is through good customerengagement technologies, good
customer engagement mechanisms.
I should say not technologiesTechnology definitely has a role
to play in that, but helpingthose customers understand we're
able to reduce overall energycosts because you're adopting
(35:57):
these technologies.
How much are we able to reduceit?
I think Fremont Power has beena fantastic example of that.
I think almost annually theystate publicly how much money
(36:20):
their virtual power plant hassaved their rate payers, which
is huge for customers tounderstand what the benefit is
and really adopt this sort oftechnology.
But a lot of customers areadopting the technologies
because they want it and they'reactually going to their utility
.
We see this in a lot of co-opsand we hear this from a lot of
our cooperative customers.
They say we want to stand up aprogram because our customers
are asking us to.
So a lot of customersunderstand that by doing these
sorts of things they're going tohelp save the world effectively
(36:41):
.
But that doesn't get you out oftelling them how much going to
help save the world effectively,but that doesn't get you out of
telling them how much they'rehelping save the world as well.
Jen Szaro, AESP (36:50):
Absolutely.
I think you need to remindpeople of the impact that they
could have themselves and then,if we can, as an industry,
empower that behavior.
I mean I think that's thewin-win that I see for us.
Absolutely, Me too.
Bill Burke, Virtual Peaker (37:03):
Me too.
Jen Szaro, AESP (37:04):
Well, thank you , bill.
This has been a reallyincredible conversation.
I'm just so intrigued by allthe work that you and your team
are doing right now at VirtualPeeker, and I cannot wait to see
what comes next with AI overthe next years, and so thanks
for spending time with us todayon the Energy Beat podcast.
Bill Burke, Virtual Peaker (37:24):
Jen, this has been fantastic.
I really enjoyed it as well.
Wide-ranging discussion, wewere a bit all over the place,
but hopefully that's the goodpart.
Yeah, I think so too.
I think hopefully we helpedeverybody who's listening
understand AI a little bit morehow it could help utilities
control demand better, helputilities engage customers
better.
There's a lot of excitement inthe world.
(37:45):
Thanks a lot for having me.
Jen Szaro, AESP (37:47):
Thank you, and if you'd like to learn more
about Bill and his company, justhead over to virtualpeakercom
and check them out.
Thank you so much for beingwith us today.
We'll see you next time.
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
Crime Junkie
Does hearing about a true crime case always leave you scouring the internet for the truth behind the story? Dive into your next mystery with Crime Junkie. Every Monday, join your host Ashley Flowers as she unravels all the details of infamous and underreported true crime cases with her best friend Brit Prawat. From cold cases to missing persons and heroes in our community who seek justice, Crime Junkie is your destination for theories and stories you won’t hear anywhere else. Whether you're a seasoned true crime enthusiast or new to the genre, you'll find yourself on the edge of your seat awaiting a new episode every Monday. If you can never get enough true crime... Congratulations, you’ve found your people. Follow to join a community of Crime Junkies! Crime Junkie is presented by audiochuck Media Company.