December 13, 2024 • 8 mins
Can you imagine a world where artificial intelligence revolutionizes the way we power our lives? Discover how AI is fundamentally reshaping the energy sector by enhancing grid efficiency, reliability, and resilience. We explore the challenges faced by our energy infrastructure—from rising demand to weather-related vulnerabilities—and reveal how AI-driven solutions like predictive maintenance, dynamic line rating, and topology optimization are enabling smarter use of resources. Learn about the promising developments in streamlining interconnection times for new projects and the ways AI is optimizing generation operations using precise weather forecasts and sensor data.
Uncover the significant role of generative AI in revolutionizing power grid modeling and price forecasting. Discover groundbreaking work by the National Renewable Energy Laboratory, where advanced neural networks are accelerating the design of wind turbine blades and simulating complex grid scenarios. These innovations could predict the impact of electric vehicle adoption on energy loads and infrastructure, marking a step towards a more sustainable future. As we embrace these technological advances, we also address the need for balance, considering the additional demands AI places on the grid. Stay tuned for our next session, where we'll focus on AI's transformative effects on the distribution grid.
Uncover the significant role of generative AI in revolutionizing power grid modeling and price forecasting. Discover groundbreaking work by the National Renewable Energy Laboratory, where advanced neural networks are accelerating the design of wind turbine blades and simulating complex grid scenarios. These innovations could predict the impact of electric vehicle adoption on energy loads and infrastructure, marking a step towards a more sustainable future. As we embrace these technological advances, we also address the need for balance, considering the additional demands AI places on the grid. Stay tuned for our next session, where we'll focus on AI's transformative effects on the distribution grid.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Speaker 1 (00:00):
In the past two weeks , since I put up the last of
five videos on AI datacenter-driven demand on the grid
, there have been a number ofcritical updates, with perhaps
the most interesting being thatMeta recently announced they
will issue an RFP seeking 1 to4,000 megawatts of new nuclear
generation to come online in the2030s.
Winning that bid wouldcertainly boost the
(00:21):
investability of any number ofstartups in the space, although
shares of the publicly tradedmodular nuclear reactor
companies haven't been doing toobadly lately.
But let's turn our attentionaway from the data center's
near-insatiable appetite forjuice and discuss ways that AI
could actually help to make thegrid more efficient, reliable
and resilient.
A good rule of thumb about AIin this space is to think about
(00:44):
the following questions whatcould I do in various areas
across the grid if I had accessto instantaneous, accurate and
highly detailed information?
What if I had the ability toask any question about the grid
and grid-connected assets that Iwanted?
And what if I could reliablyand instantly act upon that
information and receivereal-time feedback in response
(01:05):
to my actions?
To better understand AI'spotential value proposition here
, let's start with the basicchallenges and opportunities to
be addressed.
First, our grid is bursting atthe seams both at the
transmission and distributionlevels, with infrastructure
increasingly hitting thermallimits at certain times, largely
when ambient temps are high andthere's a lot of current
(01:26):
flowing.
Second, there's an enormouspent-up demand for new supply,
with difficulties in gettinginfrastructure permitted and
built.
And third, the grid isincreasingly vulnerable to
severe weather.
And fourth, it's still carbonintensive.
Now let's look at howAI-assisted tools can help.
Today, we'll tackle somepromising use cases on the
(01:47):
supply side in the bulk powersystem, with transmission, ai
holds significant potential.
For example, it can help withpredictive maintenance, with
algorithms scanning data sets tolook for anomalies that
indicate potential loomingequipment failures.
Operationally, ai can helpboost the performance of
transmission lines by assistingcertain grid-enhancing
(02:08):
technologies known as GETs.
These don't involve buildingout new infrastructure, but
rather making more efficient useof the equipment we already
have.
The first of these is dynamicline rating, in which historical
method of limiting lines basedon static capacity ratings is
ditched in favor of an approachlooking at actual ambient
conditions.
Lower temps and higher windspeeds, for example, pull heat
(02:32):
from transmission lines,allowing them to move more power
, in some cases as much as 50%or more.
That helps limit congestionbottlenecks and aids with the
interconnection of moregenerating assets, rather than
being subject to a firm 24 by 7,365-day line rating that would
(02:52):
preclude their interconnection.
They might be able to operatemost of the time and only be
curtailed a few hours during thehottest and stillest days.
Absent the situationalawareness afforded by AI, you
could not adopt this flexibleand pragmatic approach.
This may be most helpful towind assets since, logically,
during the same periods whenwind turbine output is high,
that same wind is dissipatingheat from transmission lines.
(03:14):
Then there's so-called topologyoptimization, a fancy phrase
for directing electron trafficby opening and closing breakers
to route power differently,facilitating higher utilization
of assets.
Ai can help here as well bymore quickly assessing a wider
variety of scenarios across thegrid, and this can also help
with interconnecting new assets,speaking of which
(03:38):
interconnection itself is a bigproblem.
In its latest report, lawrenceBerkeley Labs noted that the
median time for interconnectingnew assets from the study
request to transmissionoperators to the time power
actually flows is north of fiveyears.
In the year 2000, it only tooktwo years to interconnect
because planners were dealingwith fewer and far larger
(03:59):
projects, mostly big coal andgas plants, the elephants.
In large part, today's delay isa result of a changing resource
mix and a numbers game.
The elephants have beenreplaced by cats and dogs, and
rats and frogs, with largenumbers of projects as small as
10 megawatts joining the queue.
In the 2000 to 2004 period,about 300 projects waited
(04:21):
patiently in the queue.
That number grew to about 850annually from 2005 to 2014, and
has since soared to over 3,000requests per year in recent
years, with now 10,000 projectsimpatiently languishing in that
queue.
The challenge posed by theselarge numbers is that there are
many more moving parts toevaluate.
(04:41):
Each new project or potentiallyapproved cluster of projects
changes the situation for thoseprojects waiting downstream,
creating a constantly shiftingdynamic that dramatically
increases the need for planningresources and sheer
computational capability.
This is an area where AI holdsenormous amount of potential in
both cutting the time requiredand increasing the number of
(05:03):
scenarios that can be assessedin that interconnection planning
process.
The DOE reports, thetransmission owners and software
developers have started todeploy these newer models, but
substantial work remains to bedone in this area.
At the same time, a lot'shappening on the generation side
of the equation, with numerousopportunities and digitalization
(05:23):
of various assets that willhelp operation of the gen fleet
and dispatch strategies.
Gas generators, for example,can be run more efficiently
based on actual operatingconditions rather than
prescribed schedules.
For example, algorithms appliedto information derived from a
variety of sensors can tell theplant operators how hard they
can run a turbine, perhapsoverfiring and exceeding
(05:46):
nameplate ratings on a frigidday when demand is high and
power prices are soaring.
They can also better understandwhen to take turbines out for
maintenance rather than relyingon fixed schedules.
Perhaps an apt analogy herewould be changing the oil in
your car, not based on the milesdriven or months since your
last visit to Jiffy Lube, butbased on the sensors in the oil
(06:07):
actually telling you how dirtythat oil is.
Ai, married to more powerfulcomputers, is helping to
generate more accurate weatherforecasts.
Longer term and moregeographically precise
locational forecasts can helpgreat operators refine their
output projections and dispatchstrategies, while optimizing
utility-scale battery storageand dispatch as well.
(06:27):
Within a wind farm, ais canalso help with so-called wake
steering, which has nothing todo with directing mourners
towards an open casket.
No, by orienting wind blades afew degrees, computers can
minimize the disruptions in windflow affecting downwind
turbines, thus optimizing output.
Among other benefits.
(06:47):
The National Renewable EnergyLaboratory, nrel, suggests that
integration of wake steeringstrategies into the siting and
planning process could cut landrequirements for future wind
plants by an average of 18% andup to 60% in some instances.
Ai can also help newsustainable technologies such as
advanced geothermal projectsthat extract heat from solid
(07:09):
rock miles underground and usethat to generate power.
Here machines and algorithms doall sorts of things, from
telling operators where to drillto physically guiding the drill
bits through the subsurfacehard rock, predicting reservoir
behavior and determining howmuch heat to extract from a
given area over a specificduration.
In addition, ai has the abilityto speed up environmental
(07:33):
reviews and completion ofdocumentation for
interconnection requests theboring but necessary stuff
associated with these projects.
Some of these applications arealready happening with AI
related to machine learning.
Here, computers take existinginformation and make sense of it
using algorithms that identifypatterns to make decisions.
But as the large language modelsbecome increasingly powerful
(07:56):
and more sophisticated, theability to develop generative AI
to understand the patterns ofexisting data and then generate
new data to improve decisionmaking well, that will take us
to the next level.
Here we'll see increasedabilities to model the nation's
power grid, predict future powerprices, develop better
technology.
Nrel reports, for example, it'salready employing advanced
(08:17):
neural networks to improve thedesign of wind turbine blades
100 times faster than throughprevious methods, and model
countless what-if scenariosacross the grid.
For example, what if EVadoption increased by X
percentage in a specificlocation?
What would the impacts be onhourly loads and associated
supply infrastructure?
That's where we're headed.
So not only should we get usedto AI, we should lean into it.
(08:41):
If these data centers and theirlarge language learning models
are going to stress the grid onthe demand side, we might as
well get as much value out ofthese new capabilities as we
possibly can.
In our next session, we'll talkabout the impacts of AI on the
distribution grid.
Thanks for watching and we'llsee you again.
In the past two weeks , since I put up the last of
five videos on AI datacenter-driven demand on the grid
, there have been a number ofcritical updates, with perhaps
the most interesting being thatMeta recently announced they
will issue an RFP seeking 1 to4,000 megawatts of new nuclear
generation to come online in the2030s.
Winning that bid wouldcertainly boost the
(00:21):
investability of any number ofstartups in the space, although
shares of the publicly tradedmodular nuclear reactor
companies haven't been doing toobadly lately.
But let's turn our attentionaway from the data center's
near-insatiable appetite forjuice and discuss ways that AI
could actually help to make thegrid more efficient, reliable
and resilient.
A good rule of thumb about AIin this space is to think about
(00:44):
the following questions whatcould I do in various areas
across the grid if I had accessto instantaneous, accurate and
highly detailed information?
What if I had the ability toask any question about the grid
and grid-connected assets that Iwanted?
And what if I could reliablyand instantly act upon that
information and receivereal-time feedback in response
(01:05):
to my actions?
To better understand AI'spotential value proposition here
, let's start with the basicchallenges and opportunities to
be addressed.
First, our grid is bursting atthe seams both at the
transmission and distributionlevels, with infrastructure
increasingly hitting thermallimits at certain times, largely
when ambient temps are high andthere's a lot of current
(01:26):
flowing.
Second, there's an enormouspent-up demand for new supply,
with difficulties in gettinginfrastructure permitted and
built.
And third, the grid isincreasingly vulnerable to
severe weather.
And fourth, it's still carbonintensive.
Now let's look at howAI-assisted tools can help.
Today, we'll tackle somepromising use cases on the
(01:47):
supply side in the bulk powersystem, with transmission, ai
holds significant potential.
For example, it can help withpredictive maintenance, with
algorithms scanning data sets tolook for anomalies that
indicate potential loomingequipment failures.
Operationally, ai can helpboost the performance of
transmission lines by assistingcertain grid-enhancing
(02:08):
technologies known as GETs.
These don't involve buildingout new infrastructure, but
rather making more efficient useof the equipment we already
have.
The first of these is dynamicline rating, in which historical
method of limiting lines basedon static capacity ratings is
ditched in favor of an approachlooking at actual ambient
conditions.
Lower temps and higher windspeeds, for example, pull heat
(02:32):
from transmission lines,allowing them to move more power
, in some cases as much as 50%or more.
That helps limit congestionbottlenecks and aids with the
interconnection of moregenerating assets, rather than
being subject to a firm 24 by 7,365-day line rating that would
(02:52):
preclude their interconnection.
They might be able to operatemost of the time and only be
curtailed a few hours during thehottest and stillest days.
Absent the situationalawareness afforded by AI, you
could not adopt this flexibleand pragmatic approach.
This may be most helpful towind assets since, logically,
during the same periods whenwind turbine output is high,
that same wind is dissipatingheat from transmission lines.
(03:14):
Then there's so-called topologyoptimization, a fancy phrase
for directing electron trafficby opening and closing breakers
to route power differently,facilitating higher utilization
of assets.
Ai can help here as well bymore quickly assessing a wider
variety of scenarios across thegrid, and this can also help
with interconnecting new assets,speaking of which
(03:38):
interconnection itself is a bigproblem.
In its latest report, lawrenceBerkeley Labs noted that the
median time for interconnectingnew assets from the study
request to transmissionoperators to the time power
actually flows is north of fiveyears.
In the year 2000, it only tooktwo years to interconnect
because planners were dealingwith fewer and far larger
(03:59):
projects, mostly big coal andgas plants, the elephants.
In large part, today's delay isa result of a changing resource
mix and a numbers game.
The elephants have beenreplaced by cats and dogs, and
rats and frogs, with largenumbers of projects as small as
10 megawatts joining the queue.
In the 2000 to 2004 period,about 300 projects waited
(04:21):
patiently in the queue.
That number grew to about 850annually from 2005 to 2014, and
has since soared to over 3,000requests per year in recent
years, with now 10,000 projectsimpatiently languishing in that
queue.
The challenge posed by theselarge numbers is that there are
many more moving parts toevaluate.
(04:41):
Each new project or potentiallyapproved cluster of projects
changes the situation for thoseprojects waiting downstream,
creating a constantly shiftingdynamic that dramatically
increases the need for planningresources and sheer
computational capability.
This is an area where AI holdsenormous amount of potential in
both cutting the time requiredand increasing the number of
(05:03):
scenarios that can be assessedin that interconnection planning
process.
The DOE reports, thetransmission owners and software
developers have started todeploy these newer models, but
substantial work remains to bedone in this area.
At the same time, a lot'shappening on the generation side
of the equation, with numerousopportunities and digitalization
(05:23):
of various assets that willhelp operation of the gen fleet
and dispatch strategies.
Gas generators, for example,can be run more efficiently
based on actual operatingconditions rather than
prescribed schedules.
For example, algorithms appliedto information derived from a
variety of sensors can tell theplant operators how hard they
can run a turbine, perhapsoverfiring and exceeding
(05:46):
nameplate ratings on a frigidday when demand is high and
power prices are soaring.
They can also better understandwhen to take turbines out for
maintenance rather than relyingon fixed schedules.
Perhaps an apt analogy herewould be changing the oil in
your car, not based on the milesdriven or months since your
last visit to Jiffy Lube, butbased on the sensors in the oil
(06:07):
actually telling you how dirtythat oil is.
Ai, married to more powerfulcomputers, is helping to
generate more accurate weatherforecasts.
Longer term and moregeographically precise
locational forecasts can helpgreat operators refine their
output projections and dispatchstrategies, while optimizing
utility-scale battery storageand dispatch as well.
(06:27):
Within a wind farm, ais canalso help with so-called wake
steering, which has nothing todo with directing mourners
towards an open casket.
No, by orienting wind blades afew degrees, computers can
minimize the disruptions in windflow affecting downwind
turbines, thus optimizing output.
Among other benefits.
(06:47):
The National Renewable EnergyLaboratory, nrel, suggests that
integration of wake steeringstrategies into the siting and
planning process could cut landrequirements for future wind
plants by an average of 18% andup to 60% in some instances.
Ai can also help newsustainable technologies such as
advanced geothermal projectsthat extract heat from solid
(07:09):
rock miles underground and usethat to generate power.
Here machines and algorithms doall sorts of things, from
telling operators where to drillto physically guiding the drill
bits through the subsurfacehard rock, predicting reservoir
behavior and determining howmuch heat to extract from a
given area over a specificduration.
In addition, ai has the abilityto speed up environmental
(07:33):
reviews and completion ofdocumentation for
interconnection requests theboring but necessary stuff
associated with these projects.
Some of these applications arealready happening with AI
related to machine learning.
Here, computers take existinginformation and make sense of it
using algorithms that identifypatterns to make decisions.
But as the large language modelsbecome increasingly powerful
(07:56):
and more sophisticated, theability to develop generative AI
to understand the patterns ofexisting data and then generate
new data to improve decisionmaking well, that will take us
to the next level.
Here we'll see increasedabilities to model the nation's
power grid, predict future powerprices, develop better
technology.
Nrel reports, for example, it'salready employing advanced
(08:17):
neural networks to improve thedesign of wind turbine blades
100 times faster than throughprevious methods, and model
countless what-if scenariosacross the grid.
For example, what if EVadoption increased by X
percentage in a specificlocation?
What would the impacts be onhourly loads and associated
supply infrastructure?
That's where we're headed.
So not only should we get usedto AI, we should lean into it.
(08:41):
If these data centers and theirlarge language learning models
are going to stress the grid onthe demand side, we might as
well get as much value out ofthese new capabilities as we
possibly can.
In our next session, we'll talkabout the impacts of AI on the
distribution grid.
Thanks for watching and we'llsee you again.
Popular Podcasts
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
Therapy Gecko
An unlicensed lizard psychologist travels the universe talking to strangers about absolutely nothing. TO CALL THE GECKO: follow me on https://www.twitch.tv/lyleforever to get a notification for when I am taking calls. I am usually live Mondays, Wednesdays, and Fridays but lately a lot of other times too. I am a gecko.