The grid needs an automation upgrade

When the Marshall fire ripped through the front range of Colorado in late 2021, many Boulder County residents were left without power for more than a day. The governor declared a state of emergency as the wildfire wreaked havoc on the power grid.

Kyri Baker, a professor and researcher at the University of Colorado Boulder, is an expert in power systems. She researches how to use artificial intelligence to make the grid more resilient and efficient. During the Marshall fire, Baker lost power in her home for thirty six hours, something that could have been shortened, or prevented, with better automation.

“We have a lot of consumers who still, when they have a power outage, have to call the utility to inform them of the power outage,” she said during an interview on the With Great Power podcast. “That shouldn't be the way things work. It should be automated. It should be as efficient as possible to restore power to customers.”

Baker started researching grid automation at the National Renewable Energy Laboratory almost a decade ago. In 2016, she helped build a home energy management system called Foresee, which used machine learning to track energy usage patterns and grid conditions. Four years later, the software was deployed in a small mountain town just north of Aspen where simulations showed the platform brought energy savings up to 7.6%. 

Baker is still surprised by how much of the grid is still manual, given that utilities have been talking about digitization for decades. At the University of Colorado Boulder she researches how machine learning can replace manual operations across the transmission and distribution grids. 

In particular, she studies a subset of machine learning called neural networks, which “do really well with extreme amounts of data and uncovering patterns” and can help reduce outage restoration time. Using millions of data points about the grid, these models can predict when outages will occur and the most efficient way to dispatch repair crews.

Baker is also exploring how machine learning models can optimize power flow. “The way that we currently operate transmission grids, one of the most complex human made systems in existence, is via these approximations of power flow physics,” she explained. 

With recent advancements in computing power and artificial intelligence, Baker says operators can move away from approximations and use higher fidelity machine learning models to understand how power moves across the grid. “A core part of my work…is training machine learning models to solve extremely complicated grid optimization problems that if we were to use existing techniques would just take too long or be too hard to solve.”

While Baker continues to train power flow models, she hopes to see utilities rely more on neural networks to predict equipment failures and reduce outage times. “A lot of utilities and ISOs are already doing these things,” but she wants them to trust the models more and expand beyond their comfort zones.

She says the next step is for the models to “co-pilot” with humans, not replace them. “AI can help us use data…and make those decisions that are otherwise very complex for a human to make manually,” she added.

As someone who works with this technology every day, she admits that giving machine learning more control over the power system is scary: “At the end of the day, I still believe that the human behind these models needs to fully understand what's happening…and fully understand how to override anything that doesn't make sense.”

In this episode of With Great Power, host Brad Langley talks with Kyri Baker about applications of AI for the power system and where the technology is headed next.

With Great Power is a show about the people building the future grid, today. It's a co-production of GridX and Latitude Studios. Subscribe on Apple, Spotify, or anywhere you get your shows.

Transcript

Brad Langley: It's been more than a decade since a team of former Apple engineers launched an ambitious plan to make smart thermostats a coveted consumer product.

News anchor: Since about 2012, the Nest Learning Thermostat has been promising energy savings, money in your pocket and easy smart home control.

Brad Langley: With its modern design digital screens and learning algorithms, the Nest thermostat promised to lower energy bills by automatically adjusting the temperature in a home. It was so popular in fact that Google bought the company for $3.2 billion in 2014.

News anchor: Now owned by Google and onto its third generation, the latest model adds a few additional features. So do they make this pricey thermostat worthwhile?

Brad Langley: The thermostat ignited interest in home energy automation, and it was an early success story for machine learning at the edge of the grid. But in 2015, a researcher at the National Renewable Energy Lab wanted to go further. She wanted to build a home energy management system that went beyond Nest's optimization algorithm.

Kyri Baker: Smart home energy management systems weren't very prevalent, so there was a Nest thermostat that did some basic optimization, basic cost savings and comfort optimization. But a lot of what we aim to do was make it more holistic. So link with the broader grid, actually look at the carbon intensity of the grid, the cost of electricity in real time, and design a system that was very intuitive.

Brad Langley: That's Dr. Kyri Baker, and she was one of the lead developers working on the platform. With the help of some behavioral scientists, Kyri and her team developed a survey to understand the various ways people use energy in their home and how they balance financial savings, carbon savings, and comfort.

Kyri Baker: These kind of comparison questions were designed by these behavioral scientists. People ranked what they cared about, and then the system adjusts the optimization objectives accordingly. So it was customized to each household, each person's individual preferences.

Brad Langley: They combine that customization with machine learning to track usage patterns and grid conditions. And simulation showed the platform brought energy savings up to 7.6%. But Kyri also noticed something else. People actually preferred to have some level of control over the system.

Kyri Baker: I think at the end of the day, people like having control over their technology and that's why programs like these direct load control programs where Xcel Energy or other utilities control your thermostat directly aren't super popular. Just because at the end of the day, we love knowing that we can ultimately have control over our technology.

Brad Langley: This is With Great Power, a show about the people building the Future grid. Today, I'm Brad Langley. Some people say utilities are slow to change, that they don't innovate fast enough. And while it might not always seem like the most cutting edge industry, there are lots of people working really hard to make the grid cleaner, more reliable and customer-centric. This week I'm speaking with Dr. Kyri Baker, adjunct professor and researcher at the University of Colorado Boulder. For almost a decade, Kyri has been looking at how human behavior and automation interact on the grid, and it all started back at NREL with the home energy management system she helped build called FORESEE. The name FORESEE is a bit of a play on words. It's spelled F-O-R-E-S-E-E. But it also stands for the four Cs of energy efficiency: costs, comfort, convenience and carbon. And while she was working on this project inside the National Renewable Energy Laboratory, Kyri realized she loved developing technologies that make people's lives easier.

Kyri Baker: And that really broadened my horizons. I saw the impact that my work could have rather than just writing papers and being in a classroom. I saw that I could actually build systems that impacted people. I could mentor younger researchers, I could really change the world. And so that inspired me and I got really, really into energy after that. I live, breathe, and eat energy every day.

Brad Langley: Kyri is an engineer and she's an expert on power system optimization, but she also thinks people should be at the heart of the electric grid, a view she picked up while working in the government lab. Our conversation begins with her experience at the National Renewable Energy Laboratory and how it shaped her career.

Kyri Baker: Well, we did have one project that I was put on really early that was working with utilities in California, and that project was basically introducing me to concepts like the duck curve. So they explained some areas of California have so much solar that there's all these voltage issues, there's this dip in the middle of the day and the net load. And that was something I had never really thought about at that stage. People didn't talk about the duck curve as much in 2013. I looked at this and they were telling me that there's all these restrictive permitting processes for people to install just solar on their own rooftop. It was called the California Rule 21. And part of that project was revamping that and making it faster. And if our findings were accurate, we could basically change the way that thousands of people get to install solar on the rooftops. We can make it way faster, way more cost-effective. That made me think, "Wow, we're actually doing stuff here that's going to impact renewable energy and the average person." And that was really cool to me.

Brad Langley: So in 2017, you started teaching at CU Boulder. Why did you decide to start teaching?

Kyri Baker: I had always wanted to be a professor. My dad was an electrical engineering professor, and I grew up around that. I went to his classes, I saw that he gets to help students learn and we did activities with his research group. Things you do in Idaho, you go to the side of the road and shoot debris. That was what we did with his research group. And I thought it was really neat. I was like, "My dad is respected in his field. He's helping people. He's inventing new things. He has over 200 patents or something crazy like that." And so I wanted that with my own life. And also seeing that he had a very flexible schedule and got to travel and that seemed cool as a kid. I knew I wanted to teach. I just didn't anticipate it to be in energy, and I think it was a pleasant surprise.

Brad Langley: What kind of courses are you teaching now?

Kyri Baker: Right now I'm actually teaching a basic intro to circuits class. It's the basics of electricity, Ohm's law, voltage, current resistance. That's probably my favorite class to teach. I think you would expect me to teach in the power grid class my core area of research. But I think starting even more fundamental really helped me realize and comprehend these basic concepts, which I then apply to more complex systems. I really thoroughly understand all these basic laws that when I took them as an 18-year-old, I didn't care about or think much about later. And I also enjoy teaching undergraduates. I think they're more lively sometimes, and sometimes it's really nice to see how excited they get to learn new stuff.

Brad Langley: And how is teaching different from doing research in a lab?

Kyri Baker: Oh, it's totally different. I can guarantee some of the students in my class don't even know I do research. They just think of me as a teacher. But the majority of my day is definitely research. Teaching is being able to explain existing concepts well and research is being able to explore new concepts. Research allows me to sort of exercise the creative part of my brain, which I love. And then teaching allows me to feel like I'm giving back and helping my knowledge impact younger people.

Brad Langley: Last year you were featured in the New York Times article, and I love the title, Do You Even Decarbonize, Bro? And you were characterized as a self-described decarb bro. First of all, what is a decarb bro, and why do you consider yourself one?

Kyri Baker: I didn't anticipate being contacted for that article. It was a huge surprise and a huge life experience to have a sunrise photo shoot for the New York Times outside a wind farm. That was really, really cool. The term bro has some negative connotations. It's usually guys that are drinking beer and watching sports and being womanizers. But the term bro, I think is actually gender agnostic. It's not really tied to just men. It really means somebody who's excited about life, who wants to not take things too seriously, but who wants to pursue things that they care about. And so I consider myself a decarb bro because I'm not a climate doomer. And you see that a lot, especially with the younger generation where they're so depressed about climate change. This one student at our university said sometimes she has a hard time getting out of bed because she's so depressed about climate change. This mentality is not going to solve climate change.

I prefer the alternative tactic of getting excited about the technology and the things we can do to help make the world a better place rather than the things that are currently wrong about the world. That's what the decarb bro aims to do is to get people pumped up about decarbonization technologies.

Brad Langley: Let's talk about AI a little bit. AI is obviously permeating the industry. I was just at Distribute Tech, a major event and a lot of the sessions are focused on AI, whereas a year ago we weren't quite there yet. Obviously it's just accelerated hugely in the past 12 months. And you research how AI can be used to speed up energy optimization across the transmission and distribution grid. Let's start with current challenges of energy optimization. Can you describe the problems you're trying to solve across the whole energy value chain?

Kyri Baker: One of the problems I'm trying to solve is the home energy management system problem. I do still think that there's a lot of work to be done with individual consumers, how we understand energy, how we use it, and how it coordinates with the power grid. But some of the bigger scale projects I'm interested in are at the grid level, so distribution grid and transmission grid control. And I would say the biggest one is the way that we currently operate transmission grids, one of the most complex human-made systems in existence, is via these approximations of power flow physics. The way that power flows throughout the grid is highly nonlinear. It's very complicated. We've approximated it for markets, we've approximated it to dispatch generators and to determine prices. I think the computing power, especially with AI, is at the point where we don't need to make those approximations anymore. We can speed up those computations, solve those problems faster and actually operate the grid more efficiently by using the high fidelity model of the power lines of the power flow of actual voltages in the grid.

That's what a core part of my work does is training machine learning models to solve extremely complicated grid optimization problems that if we were to use existing techniques would just take too long or be too hard to solve.

Brad Langley: You're working on machine learning and neural networks to address these problems. Can you maybe give a brief description of those technologies for those that may not be as aware of what they are?

Kyri Baker: Machine learning and AI in general are basically fancy terms for just saying you're going to utilize a bunch of data to uncover relationships. In this case, you're taking a bunch of historical grid operation data or you're generating it with a model offline. Let's say you're an independent system operator, you have a model of your grid. You just run a bunch of different loading scenarios in the grid. You look at what the voltages ended up being, you look at what the generator outputs ended up being. And the model in this case, often a neural network because they do really well with extreme amounts of data and uncovering patterns, the complex patterns, is the neural network reads in this data, you train it to learn the relationships between inputs and outputs. For example, the current loading status of the grid is X. The neural network should be able to produce what the best solution for the generators to do just based on the millions of data points it's learned from.

You don't have to run that physical model and wait two hours for it to finish running and solving. You run the neural network, which has just been trained to learn these things. And the name, it comes from a basic modeling of the human brain or brains in general. There's neurons firing in particular patterns that have been learned since we were babies absorbing information from our environment. And eventually our brains understand, for example, different patterns that are really hard to articulate with social interactions, for example, or other things that have just been training our brains for decades to understand but would be hard to write in math. Neural networks help learn those relationships.

Brad Langley: And are you focused on any kind of specific applications for these technologies right now as it relates to energy optimization? Where do you see this really helping in transmission and distribution?

Kyri Baker: Well, the main purpose of speeding up these computations is basically if we can do that, we can use a higher fidelity model for the power flows, and then that will do things like cut costs, cut carbon, improve the reliability of the grid. We've done some small studies that show if ISOs changed the model they use to clear the market or determine power flows, we could cut a significant amount of carbon emissions just from that. It would be using generators more efficiently. It would be lowering losses and therefore lowering generation to begin with. People don't normally think about that when they think about grid enhancing technologies or building out new transmission. But I like to say that emissions can be lowered today with a software upgrade. We just need to be operating these assets more effectively.

Brad Langley: And you pay a lot of attention to the distribution grid because that's where a lot of the bottlenecks to electrification will occur. Can you describe what those bottlenecks are?

Kyri Baker: I think the distribution grid is totally underestimated right now. I think finally it's starting to get the attention that it deserves, but if we're going to have everyday consumers help solve these climate change and decarbonization problems, we're going to need to eventually electrify. We're going to need people to replace their gas vehicles with electric vehicles, replace their gas furnaces with heat pumps. If you see something where let's say your neighbor gets an EV, it blows out the distribution transformer and they have to pay for a large part of that upgrade, you're going to be really turned off from buying an EV. These psychological aspects and societal aspects are a huge bottleneck to adoption. There's a lot of studies out there that show that EV adoption, for example, is very clustered in neighborhoods where people are adopting EVs, then the neighbors start adopting EVs. People are very, very social animals who react a lot to what they see around them. And so if the distribution grid is failing, if there's more outages, if there's transformer shortages, electrification is not going to happen.

I think we need to be proactively upgrading these components and being ready for electrification, not just using the run to failure model that utilities have been using in the past.

Brad Langley: AI can be used to infer information about the grid. How does it do that and what kind of information is it inferring?

Kyri Baker: One example is utilities may not have models of their own distribution grids because when these things were built, distribution wasn't as relevant. It's just you need a connection to consumers. We have a lot of programs that are upgrading dumb electricity meters to smart electricity meters. We have GPS coordinates of where the meter's located. We have more granular measurements of power and voltage, but we still don't necessarily have the connections between these houses because the meters don't know where that line is going that's connected to them. Machine learning can do things like take limited measurements and infer what houses are connected to what transformer, for example. It's not always perfect, but it is better than having an intern drive around and manually mark these things, which I've had a student say that that was one of their internships, driving around noting which transformer is connected to which houses. That is one way to do it, but it's not scalable. AI helps close that gap and make it less manual.

Brad Langley: When I hear stories like that, just need to chuckle that such a modern marvel electric grid can still have so many flaws yet still be so vital and productive, it just feels like a disconnect, no pun intended.

Kyri Baker: It truly is. It's really amazing how much of the grid is still pretty manual. Even controlling large transmission switching or generator decisions is still pretty manual, which is also why I care about education. Because if we have workforce issues where there's not enough people going into power systems because the pay is low and working on LLMs pays much better, then we're going to have a huge workforce issue in the future with this big piece of infrastructure that nobody knows how to operate.

Brad Langley: What do you see as the benefits of using AI in this capacity? Is it better grid management, deferring infrastructure investments, better customer service, all of the above? Where do you see the real value coming from AI and the work that you're doing?

Kyri Baker: All those things. I think one of the big thing is getting rid of the manual aspect of this. We have a lot of consumers who still, when they have a power outage, have to call the utility to inform them of the power outage. That shouldn't be the way things work. It should be automated, it should be as efficient as possible to restore power to customers. In Colorado, we had this extremely terrible fire, the Marshall fire a couple of years ago that destroyed a thousand homes and caused multi-day power outages and including our house, we didn't have power for 36 hours and the temperature was below freezing. That kind of thing, if we have better predictive models for when outages will occur, better models for how to dispatch repair crews in the most efficient way, AI can help us use data model those things and make those decisions that are otherwise very complex for a human to make manually

Brad Langley: And recognizing utilities have done very successful work for a long time doing it, how they've done it and that some say they're kind of slow to change. Do you see this transition away from manual operations happening at scale anytime soon, or is this a decades-long effort?

Kyri Baker: I think it's already happening. I have talked to people. For example, I have a friend who's a data scientist at PG&E and she uses machine learning to predict transformer failures. A lot of the data and measurements that utilities have access to can be used to train AI models that then help them diagnose issues or do load forecasting that's better. And a lot of utilities and ISs are already doing these things. It's just trusting the models a bit more, expanding it a little bit beyond their comfort zone and having the models co-pilot with humans, I think is the next step. Not replace the humans, but co-pilot with the humans.

Brad Langley: AI obviously has a ton of benefits, a lot of that potential upside, but there are concerns about AI, there's some potential pitfalls, especially as it pertains to the power grid. What are some of those in your mind, using AI to manage the power grid? What are some of the potential issues with doing that?

Kyri Baker: I mean, one of the huge issues I see is honestly the workforce development. This is happening to my knowledge in software too. You log on to LinkedIn and there's all these jobs that are now like, "AI-assisted software engineer," or, "LLM prompt engineer." And so we're moving away from people actually knowing how to code themselves or engineer themselves to things that are very, very reliant on models that were trained on data that we don't have access to. These black box models, having increased control over our systems is something that really scares me. So even though I do the work of training AI models and having them operate grids, at the end of the day, I still believe that the human behind these models needs to fully understand what's happening and fully understand how to override anything that doesn't make sense.

Brad Langley: I know you said that automation is happening now through AI. If we look forward five years and 10 years, maybe paint a picture for us in where you see those developments happening or maybe some of those milestones that you'll be really happy to see, to know the ES is happening now, but it's really progressing how I think it needs to progress.

Kyri Baker: I think hopefully we're going to see more proactive strategies for equipment failures, like predicting when a transformer is going to be overloaded, for example, or predicting when we need to upgrade a line or a circuit breaker or fuse. Having increased measurements in the grid, which we're having every year, more sensors, more measurements, more wireless communication and ability to remote sense the status of the grid. All that data can be used to help diagnose problems more effectively and help preemptively perform maintenance. I'm hoping that even though the climate's getting more intense and we're getting increases in demand and aging components, I'm hoping AI can help reduce outage times in the more immediate future. And at the same time, hopefully lowering costs for consumers by operating the systems more efficiently. You may not have to pay some expensive fee to import a new transformer because you've already preemptively bought 20 and put them in your warehouse. Things like that where a lot of the AI is used for forecasting, I think is where the next step is.

Brad Langley: We call the show With Great Power, which is a nod to the power industry. It's also a famous Spider-Man quote, "With great power comes great responsibility." I'm curious, what superpower do you bring to the energy transition?

Kyri Baker: That's a good question. I think certainly I'm a very creative person and I'm not afraid to fail. And I think that's actually pretty rare, especially a lot of students who are starting their career in research are just so afraid to do the wrong thing or say the wrong thing, and it's like, "We're not going to get anywhere unless you sometimes look stupid." I'm past the point where I'm embarrassed to look stupid. I've gained the confidence that I look smart enough at the time that I can look stupid some of the time.

Brad Langley: Well, Kyri, this was a terrific conversation. I really appreciate your time. Thank you very much.

Kyri Baker: Thank you.

Brad Langley: Kyri Baker is an adjunct professor at the University of Colorado Boulder. With Great Power is produced by GridX in partnership with Latitude Studios. Delivering on the clean energy future is complex, GridX exists to simplify the journey. GridX is the enterprise rate platform that modern utilities rely on to usher in our clean energy future. We design and implement emerging race structures and we increase consumer investment in clean energy all while managing the complex billing needs of a distributed grid. Our production team includes Aaron Hardick and Mary Catherine O'Connor. Anne Bailey is our senior editor. Stephen Lacey is our executive editor. The original theme song is from Sean Marquand. And Roy Campanella mixed the show. The GridX production team includes Jenny Barber, Samantha McCabe, and me, Brad Langley. If this show has providing value for you and we really hope it is, please help us spread the word. You can rate and review us at Apple and Spotify, or you can share a link with a friend, colleague, or the energy nerd in your life. As always, thanks for listening. I'm Brad Langley.