Podcast Details


Rajeev Singh | Chief Technology Officer at AutoGrid

2022-07-29
Rajeev Singh is the chief technology officer of AutoGrid, a company which is known as a pioneer in the energy management space—and the first company that was backed by Arpa-E to pursue research in smart grid applications. Mr. Singh was AutoGrid’s first engineer, and in this conversation, we discuss AutoGrid’s work with helping grid operators integrate renewable energy. Hope you enjoy my conversation with Mr. Singh! Topics covered in this podcast: ​ The story behind AutoGrid and Mr. Singh's initial attraction to the company How AutoGrid leverages Cloud servers to help manage supply and demand flows Why is elasticity important for AutoGrid's functionality? What is AutoGrid Flex and how do utilities benefit from this service? How is AutoGrid able to manage renewable energy generation at such a wide scale? The variables AutoGrid takes into account when forecasting generation and consumption What is the energy storage management system that AutoGrid has developed?


Transcript


00:07 Karan Takhar
Hello everyone. This is Karan Takhar, and welcome to the Zenergy podcast. Over the past decade, India has done an impressive job of integrating renewable energy into its energy mix. For this Fulbright podcast series, I sought to investigate the enabling factors and potential of India's global leadership in renewable energy, with a focus on solar. This Fulbright series is broken down into Four Seasons. This season, we look at the next set of key technologies and regulations integral for unlocking India's continued renewable energy success at the system level. It includes conversations with leading regulators and thought leaders across energy management. Storage, transmission, and distribution. In this episode, I will be speaking with Rajiv Singh, who's the Chief technology officer of Autogrid, a company which is known as a pioneer in the energy management space and the first company that was backed by the US government body RPE to pursue research. In smart grid applications. Mr. Singh was Autogrid's first engineer. And in this conversation, we talk about auto grids, cutting-edge work with renewable energy management, and grid integration. I hope you enjoy my conversation with Mr. Singh. So since being. Founded in 2011, I know that auto grid was. One of our pics is like First-backed smart grid companies, and then since then, of course, auto Grid has grown significantly in running production. On at least. Four different continents, and I also read that you were auto goods, one of like the very first engineers. Yes, as to what attracted you to the company in the initial days, and for those listeners who may be unfamiliar, could you briefly talk about like what led to the auto grid?

02:18 Rajeev Singh
First, let me talk about the auto grid, and then let me, let me talk about myself. So, the automated company, the CEO, and the founder and CEO was coming out of exiting, uh, startup, which was in the electronic design automation tooling space. So, you know this is software that designs chips that go into less than the iPhone, right? So having done that, I mean that he exited that caught up, and then he was kind of looking for. The next thing too, so he got into dour Stanford as a director of Smart grid research. So. That's kind of where he came up with, and I think the problems are fairly similar at the kind of the micro level and the chips and then the macro level on the grid. So, he figured out that, you know, hey, we can what, what will be, but what he had been doing in the in the macro level on for chip design, then those concepts could be applied now using distributed computing and machine learning to the electricity grid. So, he wrote a proposal. Stanford said. Hey, come and join us as a record of sponsored research, he did that, and that's where the concept came from now, and his background is purely, you know, kind of chip. Sign. So obviously, building a cloud-native massively distributed system is not his thing. So, I mean, he wanted somebody to do that. So, he pinged me and it was kind of coincidental because I was at that time, you know, I was just coming off building a large platform again from zero lines of code in my previous company to run, you know, systems in the pharmaceutical. Area, and this is not for. Or anything to do with drug means. So, it's probably done by back-end systems. All of you know transactions flow through the distance. Customers are like Glaxo, Smith Kline, Pfizer, Merck, etc. So, I was brought in to build that system from scratch, and I've been there for eight plus years, and that system had been running, we had a, you know, number of customers. We had like seven applications running on the platform. So, I said, hey, my work here. It's done now. We don't have something else to do. So, it's kind of coincidental that you know when the founder was looking for somebody to build this. I was there, so I said. Hey, let's do that. Because the other thing that attracted to me is that I'm also an engineer, a mechanical engineer from IIT Kanpur. Obviously, engineering systems attract me more, you know, because these are physical systems, and they are easier for me to understand rather than something abstract like, you know, options trading or, you know, puts and gets on the in the finance. Mention markets because those are kind of generally not tied to anything not titled reality. Right. So, so yeah, yeah, so, so an electrical system being a very large machine, in fact, the electrical grid is the largest machine that humans have built. Yeah. So, as you know, right? So, this was very exciting. The dealers were coming online, you know, you could see that there was a lot of activity around solar, although there was some bad press, Fitbit, Solyndra, and whatnot. But the point is we could see the trend of shifting to renewables. And Caliban being in California was also kind of fortunate because California is at the forefront of renewable penetration and acceptance, and our research as well, there are multiple factors. So, it's. Number one is. Physical system. Very interesting problem large-scale problem. It requires cloud computing. It requires a massively distributed system, requires real-time computing. Ingestion of, you know, in millions of endpoints concurrently. So, it is all of those things combined. I said, hey, this is really, really exciting. Oh yeah. And one thing is we'll also be applying data science in real-time at scale. So, in my, not in my last previous company, but my complete before that, I built systems, Ford Motor Company and Continental Airlines where we were. Training data science, too, let's say, price or seat or price cargo. So, or, you know, figure out what's the optimal amount of incentive to be put on a forward car or truck. So, the system that I had built, I had that background as well. So this was just the perfect combination of applying machine learning at scale cloud natives. So that was pretty awesome. So, you know, I kind of jumped at it and then. I came on that I built a team.You know, I was fortunate enough to hire my friend and colleague from who I've known for at that time for about 12-13 years. And with him, I've had built these systems at Ford Motor Company, and there's, there's an article about that in Forbes. Magazine, I can forward it to you. At some point, that. The system that they've built, so I mean, he joined pretty early. On and he. Yeah. So he, but he's running. He's a chief data scientist. He's currently the chief data scientist. His name is TK Schwartz. He's on our website, so he built the data science team. I built the engineering team and the kind of cloud system engineering team, and away we go.

06:38 Karan Takhar
Mean? It's interesting.

06:40 Rajeev Singh
Penny is there, yeah.

06:40 Karan Takhar
So what exactly is the cloud system team is that were like all of the different, yeah. Could you just talk a little bit, give some details around what the cloud system is?

06:53 Rajeev Singh
 Right. So I mean the cloud system means is that, you know, with a WSI had been OK. The other thing was that I had been pushing the previous company that I was to move our systems to the cloud, and this was around the year 2008, so. Cloud was pretty. Very high you; only AWS was present. There was. I mean, I don't think Azure was there. It might have been there, but nobody knew about it. Google Cloud was certainly not there, so there was only one cloud provider, you know, which was a double OS. So that's what we mean is that prior to that, I'm not sure if you're familiar with that when people ran systems, they would either rent space or Rackspace in Acolo and then, you know. Rack up servers and manage it themselves. When a WS came in and changed that by offering, you know, big virtualization, you could actually they would manage their servers, they would manage everything. And they would rent you virtual machines by the hour. I see, like that's pretty much. What is? Uh, what's cloud and that's. What cloud is even today?

07:45 Karan Takhar
In this context, in the auto grid context. So essentially, yeah, how does that apply in terms of auto grid?

09:00 Rajeev Singh
Does the automated systems we have, uh, we have at the end of the day, we have to deliver software to our customers and. We follow the software as a service approach, So what we do is we set up our software systems on AWS so, which means that all of the servers, all of the all the data is being ingested into the servers on the. But by servers, I mean virtual machines. So they are being ingested into the virtual machines. We are using all of the AWS technologies, which are EC2, EBS, and EFS. You know that we have used a lot of technology. Of course, now, in the latest generation, we're using Kubernetes extensively, so that's, but let's not get to that. The point is all of the computing, storage, and networking. It's a memory. All of that runs on AWS hardware. Runs out. And we rented by the hour.
 
08:20 Karan Takhar
OK, very interesting. That's, so that means that we don't have to physically procure servers or worry about electrical Connexions, network Connexions, etc. It's all taken care of by Amazon, and we just obviously and we pay Amazon for the time we use those virtual machines and if there's a very.

08:57 Karan Takhar
I see.

09:00 Rajeev Singh
Interesting concept. Doing so allows us what it's called plasticity, and we'll talk about why it is important to auto grid you want. OK, yeah, I'd love to hear so. Oh, OK, yeah. So obviously, so if you're running on physical servers. First of all, you have to kind of layout the. Capital up front. You have to, you know, purchase all the servers upfront. It's a big capital expense, right? Whereas with the cloud when we started small. We could only. We would only rent, let's say, our systems ran on, you know, five virtual VM virtual machines when we were very, very small, right? And then, as our number of customers grew. We could scale up that by that system by adding more virtual machines on an as-needed basis. 

09:38 Karan Takhar
Oh, so that's. How it was elastic?

09:39 Rajeev Singh
It's called laughing so, so now, in fact, it is elastic that our number of. Servers go up and down, you know, several times an hour, so based on the instantaneous load on the system.

09:50 Karan Takhar
Oh, wow. Yeah.

09:52 Rajeev Singh
But it is that elastic. So that is very cost-effective because we don't have to be running kind of server 100% for the entire year. If you're so. You're only going to be using it for, you know, 30% of the time, and in the end, the wait we have architected the system is that because it is fairly it's microservices-based, so only the services that need load at that. For example, we need to do a dispatch and let on the electrical grid, right? So when we do dispatching electrical grid, the dispatch subsystem. Scales up and takes up more compute and CPU and thereby more EC2 VMs. Of course, I'm kind of simplifying this. This actually runs on Kubernetes. So Kubernetes does that under the hood, but the point is only the dispatch subsystem scales up, not the ingestion system or not the web applications or not we run, you know, one of everything pretty much in the platform right now. So, it is just fine-grained scaling to make very efficient use of abilities because if you don't keep an eye on the cost. Iblis can get very expensive, very fast. People don't realize this, so we have something to be cautious about, especially for us. You know, a fledgling startup. In the early days, we had to be extremely. And now we are very conscious of the cost. Early days we had to be very conscious of how much we had our AWS extended.

11:05 Karan Takhar
And just to give some more context around what exactly auto Grid does, says reading online that seven out of 10 of the largest utilities I think globally are customers of auto grid, specifically. To the auto grid flex, yeah, service, So could you talk a little bit about what auto Grid Flex is and how utilities benefit from this service, like what is in it for them meaning? Yeah. Thank you.

11:32 Rajeev Singh
Yeah, yeah. So the main premise of part of it is that we have a software platform, what you call flex, which manages and optimizes renewable and distributed energy resources. They called doctors in real-time and at scale. So what is the distribution? So if you talk to energy people in senior bills, everybody knows you have, you know, been in power. You have you have lithium-ion stationary storage, so these things are PV solar panels. So what is happening is that generation, especially in states like California, where it with be 100 by 2045 been mandated that 100% of generation will happen from renewables, and we are already over 50% at this point. I believe I haven't checked the numbers last, but I think we're over 50% of the load comes from renewables. It will be. You know, when solar now lithium-ion batteries are coming into the mix at a very, very. That's great. So what happens is that instead of this traditional model of, you know, centralized generation, either it is, you know, nuclear or coal, you know, natural gas, whatever else, right? Do you do power generation in one spot on a power plant, and then you have a transmission system, and then you distribute it to wherever it is required that model is being appended and his is got nothing to do with auto grid, this is just happening across. Below, in the last 20 years, people are generating power in their homes, right? So, you know, you I can put a solar on my roof, I can put Tesla Powerwall batteries in my garage, stored the, you know, generate power when, especially in California, it's very sunny multiples at the time. So you can just, you know, capture that energy, store it and use it. Now, all these systems are connected to the grid, they are addressable, and they are they can be sent commands to act upon it, right? I can ask the battery to charge. I can. Ask the battery to Discharge, I can. Have the solar to, you know, curtail, etc., by sending offer commands. So think of this. Each house becomes a node on this massive network, where each of them is generating or consuming power, and flex is kind of like the. It's like the operating system, which can add an. Aggregate collection either asks all the houses to consume energy. Or ask all the houses to effectively reduce consumption of energy, and thereby creating what is called a virtual power plant, I see, and that's almost. So sorry for that. So yeah, yeah, that's, that's and that that requires Real-time ingestion, real-time telemetry, real-time command and control, and this is what orbit Flex does. So all of your flex sits in the cloud, and then. Interacts with a. Lot of these controllers and tell them what to do, and then, on the other hand, it aggregates their capacity of flex. That's why it's called flex. The flexibility in the, in the, in the grid, flexibility in terms of, you know, consumption, flexibility in terms of, of generation across, you know. Ten thousand houses, let's say, and you aggregate that up and then offer it to our customers who are either aggregators or they are utilities or somebody else, right, or any energy traders?

14:20 Karan Takhar
I see. That's super interesting. So, uh. Course one of like the main challenges in terms of being able to manage renewable energy is its intermittency and variability. So I know auto Grid has developed really like shine capabilities and being able to like forecast renewable energy generation, and I'm just curious as to like how? Is auto grid able to do this? But such a great. It's scale. Are there any specific tools that are allowed?

14:50 Rajeev Singh
So, I mean, without giving too much of far sticker sauce, this is what we've been doing for at scale in other industries, and This is why my friend and colleague Kiki shorts is at the auto grid, is that we have done this in other industries by forecasting less in demand for a seat in the airlines, right? Airlines, so right now, they're pretty hard. Hard hit, but, you know, just a few more. So, though you're running like 3000 flights a day, right and so and there's a lot of data, historical data that you can you get, and then you using that you're able to forecast how much demand is going to. Be for proceeding, so similar concepts is not. I mean, obviously, there's at a high level, they're similar, but the devil is in the details. We do the same thing because we're getting telemetry a problem from every point. Let's say you have an implanted house, so we get, you know, let's say meter data, so we get consumption data from the house, and the oftentimes the modern, you know, let's say there's a Tesla Powerwall, it's submitting the state of charge from the batteries that we. You know, one minute or 15 minutes or, you know, whatever time or five. Minutes. So, we're getting this historical time series time series of these metrics from every connected node to our system, and that allows us to run machine learning on those and then figure out the load profile of the house for the next 48 hours or, you know, seven days. This is what the energy consumption pattern of that house is going to be. I mean by a house; I mean a generic term. It could be residential interest commercial. We've been hooked up to a very large. The brewery which can swing a couple of megawatts, you know, up and down based on one command from us. So, I mean by the house is just a generic term over some power-consuming entity that has solar and batteries, etc. So, the point is that we are able to learn the behavior of that connected entity that houses the platform, and so we can figure out how much energy is going to consume. And more importantly, we also know the sensitivity of that house to weather 2 two tariffs, and there are a number of other external inputs which our models figure out that this house is energy conversion and is sensitive to that also track how truly the facility or the house is, following our command. So, when we give it, we give it a command to shed load. How have they historically? Performed, and we can score them based on that. And that gives us a kind of confidence. Interval of OK, if I ask this house to give me. Do you know one kW of load shed, or if I ask about this facility? To give me 300 kilowatts of of of load shed. How? Reliably? Can they do that? And then our models factor that and say, OK, now qualify aggregate 10,000 of these facilities, then I can say by the confidence that I can give you these many megawatts, 50 megawatts to the upstream entity, which could be the aggregator. Interesting. So that's where the machine learning comes down. That gets extremely complicated, and it took it has taken us ten years. It's to define those algorithms there is, there is. It's a whole other universe.

17:51 Karan Takhar
I can imagine, yeah. Well, can you just quit? Or unless if this is confidential, but like some of the other variables that you take into account, such as, like you said, weather and then tariffs. Uhm, other ones? Or is that?

18:05 Rajeev Singh
Yeah, I'm going. I'm going by memory. Let me see whether tariffs then of. Course there's there's other vertical egressors like, you know, day of the week, hours of the day. There's some seasonality, of course, to this may or may not be a system. It depends on the type of facility. There's there's a lot of nuances here, just industrial facility. If it's 24/7 operations, they may not be 2 minutes. I mean, there's some. Seasonality, but maybe it's at the hour level, and I see, and I'm just trying to look at the slide here that has that information, but I can find it, but those are the things. I can think of off the top of their head there, there. There are others.

18:42 Karan Takhar
Thank you and I this is my last question because I know you have to go. It has to do with recent development. I was reading that AUTOGRID has partnered with Sunrun to help manage their batteries on the battery systems. And yes, I was also reading old documents released by auto Grid which was very helpful for me because kind of newly learning about the space. It was called grid flexibility for dummies, and in it, Mr. Narayan states that a good storage solution it connects to and manages a variety of energy storage devices. And enables you to easily swap in and swap out specific battery types. Yeah, I mean. Like it's it's. Able to like and adapt to different bets. So just wondering if you could talk a little bit about the energy storage management system that auto grid-like has developed as well as this recent development in terms of its project with Sun Ray?

19:42 Rajeev Singh
Yeah, so not that great. So let's doesn't unpack this. So the first one, we are very excited about our partnership with Sunrun because, obviously, someone is doing some really, really cool things in this space. They actually offer a full solution for residential homeowners where they can install solar and battery in the home and hook it up to their platforms which are running on their cloud and but what we bring water grid brings is that so someone does this kind of project development software. This is essentially just deploying the batteries and solar, and everybody homes ask, yeah right, I'm you know, it's growing very, very rapidly, and the number of homes that are adopting you knows this solar and PV, but what order it comes in is like it comes in, and then we bolt on to Sunrun systems. It's a very unique multi-cloud solution which, at some point, we will publish a paper on that, but whatever, it's cloud. Talks to Sundance Cloud and optimizes this fleet of batteries, which is which are being rolled out too. You know 10s of thousands of homes in so so that's that's why the partnership, uh, Sunrun kind of owns the assets if you will and has the battery controllers, and they may have the downstream systems that they were not aware of the abstract that out for us we are kind of overall flex provider that is in real-time able to steer those batteries. Using some shady. State-of-the-art distributed optimization technique. Because these optimizations have to happen extremely fast in real-time, so so for example, the end customer is so Sunrun, and Auto Grid will jointly offer the capacity to, let's say, Southern Cal Edison, right? Or Hawaiian Electric or somewhere some other it ability. So that's the end customer or just the customer of the system of the joint solution, and ut the customer only said, hey, give me, you know, 75 megawatts at this time they don't care about. How we do it, So what we do, what flex does it flex has a view into every connected battery and storage system on Sunrun's portfolio, and we don't have machine learning on it. And then we can figure out, as you have described, you know, the consumption patterns and the response to signals to each of those points, and we can then send a signal. And auto-correct that and deliver those 75 megawatts up to. So, the utility.

22:00 Karan Takhar
Very interesting. So it kind of is an added layer where now the batteries become another asset which auto grid can help like the utility kind of use as a resource to balance the system. Is that correct?

22:13 Rajeev Singh
Yeah, that is correct. I think because of the batteries, you can it's charge or discharge, and you can avoid you know by by by by discharging the battery and by and by doing self-consumption, you can effectively take the load off of the grid, and if you do that for 10s of thousands of houses, then. And you can deliver that excess capacity to the to the grid. I mean, the concept sounds simple, but doing this at scale and in real-time is key because that's where a lot of the technology comes in, and we leverage in-memory computation. There's a there's a lot of techniques that we use to be able to deliver on this and because it has to be done at scale and at some point. Is going to go to, you know, you know, hundreds Of thousands or the ends of homes, and we have a similar system running in Japan, by the way, with visionaries. I think they have there was last year. This is not only the US. The only thing it's a global phenomenon where people are putting in lithium-ion storage with, with, with solar, and so the opportunities for something like flexes are worldwide.

23:11 Karan Takhar
Very, very interesting. Thank you, Mr. Singh. For your time. I hope you enjoyed that episode, and do check out the show notes For more information on my guest. See you next time.


Related Podcasts



Barbara Humpton | CEO of Siemens | Creating the Energy Future

Read More


Ram Ramanan | Vice President at Bloom Energy

Read More


Nicola De Blasio | Senior Fellow at Harvard University

Read More


Alex Laplaza | Partner at LowerCarbon Capital

Read More