Gridium: Welcome to Finely Tuned. In each episode here, we’re speaking with people who care about our built environment. This podcast is built by Gridium.
Hello everyone, and welcome to this conversation with Ram Rajagopal, Associate Professor of Civil and Environmental Engineering at Stanford University and Principal Investigator of the Powernet program. Professor Rajagopal has been studying and researching electricity for 20 years and has commercial experience at IBM on Watson, at National Instruments, and entrepreneurial experience here in Silicon Valley.
My name is Millen, and I’m with Gridium. Buildings use our software to fine-tune operations.
Professor Rajagopal and I will be discussing Powernet, an ARPA-E program to develop an end-to-end, open-source system that enables real-time coordination of utilities’ centralized assets with millions of distributed energy resources. There are some very helpful videos of the technology this lab is producing at powernet.stanford.edu.
The future of distributed, integrated, and automated electricity grid is an exciting one, as such I am really excited to be here with you here today Professor Rajagopal. Thank you!
Professor Rajagopal: Hi Millen. It’s wonderful to have the chance to talk to you about Powernet, it’s a project that’s exciting to me as well.
Gridium: Great.
Before asking about Powernet, can you tell us a little about your path into the built environment, the smart grid, and what got you interested in electricity infrastructure?
Professor Rajagopal: Sure. My background is in electrical engineering and computer sciences. You know, that’s what I did in my undergraduate and in my initial grad school. And I started working at a company called National Instruments looking into how do we take measurements from the real world, process them—whether it was cameras, sensors of different types—and then control things. This was mostly done in the manufacturing context, but also for things like cars and so on. That was exciting to me—to see that interaction between the real world and this digital information world.
So when I went to do my PhD, I was looking for something like that and my advisor, Professor Pravin Varaiya—he’s kind of one of the world’s leading experts in control systems design and theory. And he had been working for a while around transportation systems because he said, “Well, it’s a challenging problem. It’s an infrastructure network. Once you have a network, now we have to think in layers and it’s not just about measurements, it’s not just about learning; but it’s also figuring out how to architect things correctly.
So, that was my first foray. That’s what I did for my PhD: figured out how to use sensors and the data from sensors to better manage or traffic networks. We design sensors. Here in the podcast they can’t see one, but I have them here in my office. And that was a success story.
As I was finishing my PhD, I was hearing a lot about the challenges of renewable energy and global warming. And I went to inform myself to the extent of this issue, the importance of this issue. And it seemed really important. And I thought, “What can I do as an engineer?” And one of the things I realized is as much like in transportation, the energy system, particularly the electric power system, is a system that generates a significant amount of data. The data is underutilized and the transformations are changing dramatically the way we use the system. So, we will need to think about these layers again. So I thought, maybe this is a good area to dive into—and that’s how I started out.
Gridium: How has your entrepreneurial and industry experience shaped your research, or your approach to research?
Professor Rajagopal: I think I learned two major things from industry. One is to find and narrow-down your problem statement, what you’re trying to solve. Because if I have to solve something, deploy it and scale it, it has to be narrowed down. Of course, from that, many things can grow. So that was one thing that has really shaped how we did the research around Powernet. But though the idea is very broad, we tried to narrow down that problem, that was the first element.
I think the second element that was very important was, in industry you don’t do anything on your own. You have partners, customers, you have advisors. So, creating and participating on an ecosystem to solve problems. So, I brought that to this project as well: the idea that we will have an external board of advisors to the project, which is not in name only. So, we meet with them. They determine and help us determine what is a good research agenda to go for the next year. And we also partnered with various organizations and wherever we could, we utilize ideas or we build upon them and we found that these same partners is a good way to disseminate the innovation that we’re bringing through this project.
Gridium: You mentioned this project in Powernet—can you tell us what Powernet is all about in an overview? I would like to know how Stanford, SLAC, the University of Florida, and Google are involved. I also noticed in the lab, in a tour, earlier this morning, that you have a Tesla powerwall…
Professor Rajagopal: Yeah. So, Powernet is a project that is aimed at understanding how can we take homes and buildings that today are mostly passive consumers of energy. When I say passive it means, you know, these buildings and homes make decisions on their own of when to consume and how much to consume based on the activities they’re having. How do we take that environment and now enable a scalable coordination so that all these homes and buildings can coordinate over a power network and achieve goals that are good for themselves, but also good for the network.
So these goals can be, you know, as a building owner, I want to minimize my electricity bill. But at the same time, I may also want to not have to upgrade the transformer. In order for me not to have to upgrade a transformer, perhaps not just this one building has to reduce the consumption—if also my neighbor was able to do it, it would be great. Or maybe there’s moments in which my solar is producing a lot; I’m not using it, my neighbor uses it. That’s a good way to match up the bandwidth that way. So, I think that was a goal we started with.
And what is the challenges there? We have heard a lot about this idea and we have seen many attempts at getting to it. The challenges are four. One, you want this thing to be scalable, so what does that mean? That means that it has to work with readily-existing technology. So one of the things we wanted to do was, can we use the cloud with this? The second thing is, for this to be scalable, we have to make things work as autonomously as possible. If it requires a lot of time as human operators, it’s almost like creating a new infrastructure. But if it is more autonomous, and it can just work on it’s own, it’s really helpful; you can just intervene when necessary. What that meant is, we need to learn the preferences of buildings and homes automatically somehow, rather than asking thousands of questions. Of course you can go in and change those options by yourself, or do you use the data to learn. Third, we need to understand and incorporate the power network. Typically, we think of a building in isolation; but now we are coordinating, we have to think about the network. And this is a kind of this black box that people don’t want to touch. But I think we are showing, and now many other projects are also starting to show that, “Well, you can cooperate in that way”. And fourth, it has to be something easy to deploy. So, those are the four aims of Powernet.
One of the things we did to start off the project was, we said, “Who are our partners that could go and make a really strong case for building something like this?” If you think of cloud, of course Google is one of the companies that does that. So, we got very fortunate that Google paired up with us and they had been developing some technology that was around managing a single home and optimizing it, and they loved it…
Gridium: Google PowerMeter.
Professor Rajagopal: …PowerMeter. And then they said, “Oh, it’ll be great figuring out how to do this on a network.”
Then, we wanted to have a partner that understood how to do these deployments and experimentation and SLAC is an expert on that today; in fact, this project actually helped us start a group in SLAC specifically looking at this problem.
And finally, for the Navy, we needed a place to run these experiments that’s like the real world. And the Navy immediately jumped up and said, “You know what? We love to try innovations and we want to be a test bed for that.” And by the way, this is one of a funny tidbit, but one of the things these commanders brought to us was, “Look… you know, you don’t even have to worry if your algorithm is not working that well.” And we said, “Why?”. “Because, these soldiers and officers are in the Navy. They will follow what the commanders say.” So, there is a chain of command, so it’s a little bit easier to manage the process.
But from that time to now, we also added the City of Fremont—so, we are recruiting homes in Fremont.
Gridium: Great.
Professor Rajagopal: And as well as rural areas, so we’re working with a facility that produces milk, and that’s where we are.
Gridium: The electricity grid is changing, indeed we’ve had the president of the future grid coalition, Mark Shahinian, join us on the podcast to discuss net-metered energy consumption. When you think about the future of the electricity grid, what does it look like?
Professor Rajagopal: I think it looks very different than what we were used to thinking about the electricity grid. I think there is going to be a host of new entities; we are already starting to see. You know, we didn’t have a charge point before. We didn’t have a Tesla before. And we didn’t have companies like OhmConnect that engage directly with the consumer and circumvent, in some ways, the utility. And the utilities themselves are getting very sophisticated.
So, the first thing I fully expect, is in this grid of the future you’re going to interact with it through many different pathways. The second thing that I expect is, a lot of the data that we produce today is going to start to get used to the benefit of the consumer, to the benefit of the building owner. But, even more, they’re going to start to invest more and more in monitoring and more importantly, the ability to control. So if you think about a lot of the grid today, the paradigm is: supply follows demand. We’re moving to a paradigm where demand and supply will try to balance each other. So demand will be active. In order for that to be realized, we need the controls; and how does that happen? It cannot be manual, so we are going to start to see a lot of it.
The other last important aspect, I think, is storage and EV charging are going to bring a dramatic change on the system. In the beginning, I used to think, “Well, these are going to be very important and they’re going to be, you know, 10%, 20% of the grid capacity. It’s going to be around that, which is great, but it’s not game changing.” Right now, I think, it’s safe to say, we are going to surpass. The adoption trends I’m seeing and also the various issues that are popping up and the change in the sector is such that, I think—and the costs, are such that I think we’re going to see a change on that. And once that changes, the paradigm on the grid of supply and demand balancing each other at every point in time, now you have a buffer. With a buffer, it changes a whole lot.
So, what does it mean if you’re a building owner? I think it means that you’re going to be able to get much cheaper electricity if you make the right choices and investments for your buildings. And you’re going to have the opportunity to be much more sophisticated in how you manage and think about power for your facility.
Gridium: I don’t want to ask you to pick a favorite, so I realize this question might be a little bit tough, but what do you think is most exciting piece about the work that your team is doing here
Professor Rajagopal: I think there’s two parts to this. One thing that is super exciting for me was forming the team itself. It’s an incredible group of students and postdocs that chose; you know, they could work on so many other problems. We are in Silicon Valley. But they chose to use their talents and skills to look at this problem because they’re passionate about helping really, you know, improve the electricity system and make the world more cleaner, a better place. I know this all sounds a little fluffy, but it took a lot of commitment from them because this is not the area where you’re going to earn the most if you finish this project. So, I should put that first. You know, it makes me want to come to work every day.
Gridium: Sure.
Professor Rajagopal: In terms of the technical contributions, I think the most exciting thing is that we have been able to design kind of this modular architecture where there is some amount of local intelligence, there is global intelligence in the cloud and there is local intelligence at the building interacts with the cloud that represents the network, the power network. And through that exchange, they can settle and cooperate. And we found that cooperation signals don’t need to be: I get a signal from this cloud coordinator every minute. In fact, we can even send you a coordination signal every 3, 4 hours. And you can choose how much of it to follow and the overall system really behaves nicely.
Beyond that, I think the idea of using data and kind of automating the processes of how a home or a building optimizes itself was also very exciting
Gridium: What are some of you key technology concerns
Professor Rajagopal: I think there are four technology concerns. One is the ability to incorporate the different types of cyber and physical constraints. So, what I mean by that? The power network is a physical constraint. The cloud and its ability to process data, and the amount of signalling it can do, it’s a cyber constraint. So, bringing those two together.
The second key technology challenge is we don’t have today high-resolution monitoring and active control at the panel of your home. You know, these are fairly dumb today. So, if we could enable that—and that’s one of the things you’re doing with the smart fuse, which is a part of this project. It’s a smart fuse. Then now, you open a whole host of applications, and the concern there is how they design that thing. You need it to be fast, it needs to sample a lot of data, it needs to work with a high amount of power. So that’s a power/electronics challenge.
Third challenge is how to design intelligence for the coordination itself that accounts for this cyber/physical constraints.
And the fourth challenge, if I’m going to make your home smart and say, now I can schedule things. I can time when they happen, I can decide when an EV should be charged. I can decide when to turn on and off your water heater. That has to happen in a way that’s fairly transparent; it cannot override your preferences. Initially, when we thought about this, the first idea was let’s collect the preferences of people, so you would send a questionnaire of 40-50 questions; who wants to do that? To program their home, answer 50 questions… it doesn’t make any sense. Your building, it’s going to be thousands of questions, if you think about it—a commercial building.
So, can you learn from the data what these preferences are going to be? And what we discovered, and what we are trying to do now—and we are going to test it in the real world; so far, we tested it in the lab here—is that if I look at the data of how you’re using the different appliances, the different equipment, I can infer something about a preference. I can see when you turn things on and off. I don’t know exactly if that is your preference because maybe every day you put your thermostat at 70, but you will be okay at 72, you’re just putting 70 in the summer because it’s a little colder. So, what we tried to do is, we observe the sequence of decisions you made, then learn a statistical model from it. Then we try to play a little bit of a game with you in that, we try to say, “Oh, let me go to the border of what you tried.” And through that, without asking too many questions, I can keep changing things until you complain, and then we know, that data point can now reset the system. So, this is a form of reinforcement learning…
Gridium: Right.
Professor Rajagopal: …but it involves the residents in that.
Gridium: That’s very cool. As I understand it, the research is bucketed into four groups: behind the meter, a home hub, consumer preference feedback loops, and power markets. Can you describe how these four elements work together?
Professor Rajagopal: Yeah, so in the behind the meter element, it’s all about understanding what to measure; it’s our smart fuse and the ability to measure those circuits and give us a little bit of controllability of the circuits as well. We can connect and disconnect them, or throttle the power using power electronics. And then it’s also about understanding how to digitally connect to these devices—that’s where the home hub comes. The home hub is really the local intelligence that sits in your home. Why do we need such a hub? Why can’t everything be directly in the cloud? Sometimes, the Internet might not be there, but your things have to continue to work and be scheduled and so on. Nobody will be happy as getting service not found or my things are not turning on properly. So, having that local intelligence allows to give that reliability, because even when my connectivity is not there, I can continue to operate.
The second thing it allows us to do is to give a little bit of privacy, because now the data that’s collected inside a home can stay in the purview of a home hub and it doesn’t need to be shared with a coordinator; maybe only the things that you allow to be shared, shared. So, that is the purpose of the home hub. It can kind of calculate and compute and think and locally optimize, for yourself.
And the third element, as you were mentioning, was about markets. And I think there it’s more about, now, I have these individual homes and they optimize themselves to minimize their bill. But I have now another layer, the cloud coordinator, that allows me to coordinate. What do I coordinate for?
One of the things you can coordinate for is to supply services to the wholesale market, like shaving a peak or following particular market signal. So, that’s where we investigate a little bit, how should this whole system interact with that market?
And I think there was also something about consumers and the feedback loops. So, what we are doing there is, now you’re home hub is talking to all the devices in your home, it’s trying to calculate and optimize for your home. It has to learn about you and like I described before, you get that feedback loop to learn preferences without too much interference, questioning, etc. And also enables us to forecast and various other things that are services of this platform.
Gridium: Sure. I recommend that our audience visit the Powernet website and review these videos, but for those folks that haven’t yet had that chance, can you describe what it might look like when a home reacts autonomously to changes in the network condition, you know, driven by Powernet; what that might look like?
Professor Rajagopal: Sure. So, you know, a couple of examples would be really nice. One example is, let’s suppose that in a particular network we have a transformer go out because it blew up or there was some problem. Normally, what would happen is your home has to shut down; this is what they told you to do. It would isolate that section and shut down power to your home. With Powernet, what it can do is now, first of all, each of the homes could go on it’s own, power-independent mode. So, I could disconnect from the network and manage my own power, locally, and optimize and reoptimize to keep going based on the fact that I have storage or a solar panel, or even discharging from my EV… I can keep going. But at the next level, it’s even more interesting. Now, let’s suppose that we got the inability for power to flow through the transformer, but we can at least exchange power amongst ourselves—like in the Stanford campus. There is multiple buildings, they’re all connected to some larger centralized transformers. That guy goes out, we could still exchange power amongst ourselves. So, how to do that? So, Powernet enables that as well.
Another kind of scenario, which is also exciting, is in systems where we have a lot of solar power… so, lots of homes adopted solar panels. There is moments in the day at noon, maybe that a lot of people are outside of home, lots of power is getting produced, now what are you going to do with that power? So if you have something like Powernet, it can take homes that have storage, or have a pool pump, or they have control of their cooling, and it can do things like I can store the energy, I can run my pool pump autonomously, and I can even precool the home so that when you come things are good and you got the cool with all of the free power that was floating around this network. So, that is kind of another scenario of what we can do.
Gridium: We were speaking a little bit earlier about the situation when there might be excess supply on the network. How can Powernet help the grid handle excess supply?
Professor Rajagopal: So, as I mentioned, I think one way is that there is a certain part of the consumption that your home that’s flexible, of your building that’s flexible. I could schedule that consumption to happen in different points—that’s one way that Powernet helps you manage, it can help you do that schedule.
Second, if they have storage, you could potentially charge your batteries with that free power. In fact, when there is excess power and that power is too much, in the wholesale market, they set the prices to be negative, so they’re basically paying to consume.
Gridium: Yeah.
Professor Rajagopal: So, Powernet can now consume and do actually useful things with it because it doesn’t require necessarily just your presence there.
Gridium: Can you describe what role Artificial Intelligence plays in Powernet?
Professor Rajagopal: Yeah, AI has a few different roles. The first role of AI in Powernet is that we use it extensively to do forecasting for all of our scheduling algorithms. The way these algorithms work is because we have storage and we are scheduling things in the future, we need to understand what might happen in the network in the future. So we use machine learnings specifically to do that kind of forecast. The second place where we use AI is really on learning the preferences of the consumer. So, we use models driven by data and interaction with the consumer. And the third place is the architecture itself and the way the scheduling algorithm works. So, there is this local intelligence, there is this global intelligence and the way we designed it, it uses the principles of AI even though it’s much more of a kind of modern optimization problem. But those are the places where we are using AI.
Gridium: Generally speaking, how has progress been? How far along is the team now?
Professor Rajagopal: We have managed to build our proof of concept in the lab, so we are able to do some demonstrations. We have integrated storage with the fuel loads and being able to follow signals. We have architected the software, designed the algorithms… and the next step is really taking it from this lab where we have real batteries, as you see, and we have appliances and so on. We are taking it from that environment to the real world environment. I think the critical change and difference there are two: one, it’s not infrastructure that’s in the university anymore. So the network, the Internet and all that… their reliability is going to be different. And second, there’s actually real people living in real homes, whereas here, we don’t have a real person living in our homes upstairs. So, it’ll be exciting because there’s kind of this education component. And the third step that I was really happy is that when we started demonstrating this today in the street, one of our industry partners—actually they were not a partner, they were attending a session—they came to us and said, “You know what? This is exactly what I need to do much more efficient cooling and management of our farms. The farms have a lot of electricity consumption.” And they design a lot of these smart equipment that you can attach to things in the farm, but they didn’t have a cloud system to manage and coordinate and so on. So, that’s something we’re helping them set up and test…
Gridium: In an industrial setting.
Professor Rajagopal: …in an industrial setting. So, we are working with a milk production facility and we’re going to control the whole smart barn actually. Because it started with let’s set up the control room, the cooling. And then we noticed, “Well, but we can also add storage.” And then we found a partner to provide the batteries. And then we said, “Oh, but we can also add all kinds of sensors.” And now it’s become this project on it’s own.
Gridium: Well, thank you very much Professor Rajagopal for your time today. I really had fun today on the tour of your lab and it’s been great and generous of you to host me in your office. This has also been a blast to do our first in-person conversation for the podcast. So again, thank you
Professor Rajagopal: Thank you for inviting me. I invite everyone to visit our website. They can also drop by and visit at Stanford and we would be happy to show the lab and give a demonstration. And I have to point out that all of this work has been supported by RPE, one of our sponsors, Google, and the National Science Foundation. So, I’m happy with that. And also particularly with the amazing set of students and postdocs that have been really carrying the load and doing everything. They’re going well beyond what you expect for a graduate students. You know, they’re developing software, putting it in the real world, testing it… it’s almost like a small startup. It’s amazing.
Gridium: It’s inspiring work. Keep it up!
Professor Rajagopal: Thank you.
Gridium: Thank you.
