244. AI Transforming Distribution Grids - Aug26

With Laurent Segalan from London and Gerard Reed from Berlin, this is redefining energy. Today on redefining energy, we're going to talk about how AI is going to transform the distribution grid. Yes, Gerard, and every day we get contacted by some new software company, found the solution for the grid, but we found the perfect one. Yes, and this is the CEO and founder of a company called Plexigrid Swedish Company. I've known that founder Alberto Mendes for many, many years used to be formerly with Vattenfall, like so he was on the board of Vattenfall at one point in time, and he stepped out of his nice, secure job to try and revolutionize the distribution grid. Yes, and the show today is a last show of our extraordinary partnership with the BABW Foundation AirBacquant. And just for those of you who don't know who the BMW Foundation Herbert Quant is, the foundation unites leaders across sectors to develop solutions that foster an innovative economy and a future proof society. A key focus of the foundation is energy transition and climate change, where the foundation drives international collaboration to accelerate the energy transition, with rising demand from AI and data centers, new partnerships, effective collaboration, and exchange of science-based solutions and strategy are essential. And that's really what their big push is. And yeah, we've been a lot of fun with BMW and we're looking forward to continue work with them in the future. So I guess today Alberto Mendes Rebolo, as you said, he had an incredible corporate career, Siemens Gammes on Vattenfall, big organization, and then now he's a co-founder and CEO of Plexigrid. Now Plexigrid and Alberto is going to explain what he does, but the only thing which is interesting is the operate SaaS model. Three years ago, he had a pipeline of 8 million, and now he has a pipeline of 150 million. So everybody is looking at his door because he finds some fantastic solutions. Yeah, and I think this is particularly the case in the grid area, and I said that because it's not easy. And just my experience over the last 15, 20 years, it's been very difficult for outsiders to get into that grid space and get into the market, so these guys are doing it. So yeah. Well, let's listen to Alberto. Alberto, welcome to the show. Thank you for having me. Alberto, maybe I let me kick off by asking, when we met, I think you were the chief operating officer in Vattenfall. The chief procurement officer. Chief procurement officer. Exactly. You're the chief procurement officer. A lovely, safe job, well paid, and then you decided to come to an entrepreneur. Why did you do that? Well, it's a combination of factors, but at this point, I would like to talk about my co-founder, Pablo. He's one of the best electrical engineers in the world, in the sense he literally got OSTAM in the electrical engineer of the year in 2022 from my three-poly. This is the electric Oscars and he's the real genius behind the Plexigree technology. How we know each other is that we study electrical engineering together. I always say, yes, who was the smart guy in the class and who wasn't anyhow, but so our relationship, besides being very good colleagues ever since university, is that anytime or engineers at Simmons, Gamessa or later on at Vattenfall couldn't figure something out electrically. We will send this to Pablo and his team and they will figure it out every time. Back to your question, why we decided his full professor of power systems doing very well, so he also had a kind of cushy job. He didn't need to get into all this trouble that we have got into, but through 25 years in energy transition, it became quite apparent to us a few years ago that the next big challenge is not to make a renewable energy cost competitive with fossil fuels, is the integration and the functioning of the grid, a grid that was designed for a complete different energy system, centralized and so on. I remember the day when we decided we will do something about it and it was along meeting, we spoke for hours, and then I asked Pablo who is working on this at Richard's level because he's also chief editor of the Elsevier Power System Journal which is alongside the IEEE journal where you get to publish most of the innovation in electricity networks. I said, "Well, these and these professors are working on this, but at a very theoretical level." And then on my side, since I was working in procurement in Baton Fagerard, I said, "Well, I'm going to ask to the big incumbent companies, right, the big names, and ask them what are they doing about this?" And I could see there was a big vacuum, right? At best, they have some power points or some ideas, but at the minimum, I will have expected to see dedicated research teams in these big companies trying to nail the problem, and I couldn't find it. So then I called Pablo and said, "There's no match, there is really here a vacuum, and it's got to be figured out." And we got increasingly excited about this and say, "Well, somebody's got to do it. So why not us?" And we have a good combination of skills, we have what it takes to take on this problem and nail it. So you start six years ago, what was your first product or software? And maybe you're going to explain how it has evolved over time. Initially, we developed, in university, a primitive version of a digital twin of the distribution network. Alongside one of the main European network operators, it was a night if, in the sense that certain tasks like cleaning topology or cleaning bad data or working with time series, we were using AI algorithms, machine learning algorithms, time series algorithms, but I will call it AI at application layer or a composite digital twin, if you will. It has a traditional solver, traditional modeling combined with some machine learning and inference for when the data is bad. And we thought that it will stay like this. Honestly, I mean, I'm a very practical person, don't invent the will. Surely there must be open models out there that can take care of the tasks that are needed to build a digital twin for the distribution grid. But as we went down the rabbit hole deeper and deeper, we realized that we tested and broke many nearly all open models in cladding physics based models. And we realized that the problem statement that we are addressing with hundreds of millions of nodes, with thousands of changes in topology every day with imperfect data from SmartMid is that break every now and then and noise data from, from the years, no existing model could do it. And most importantly, no existing model fulfills critical infrastructure requirements. So hallucination is not an option here because it could create blackouts. The data has to stay in the operator. So you need to introduce federated learning, explainable AI and fundamental and traceable AI and auditable AI is important. And for example, LLM's don't fulfill most of these things. The more we went down, the more we realized that we're going to have to become a foundational AI company. And we have to develop a foundation model. So you could see the core of Plexigrid is a foundation model for the electricity grid or a world model or a physics AI model of the grid designed to fulfill critical infrastructure criteria. So Alberto, that's all very technical. So could you tell me why the hell do we need this digital twin? We need to start from the beginning, grid 1.0 was designed for an energy system with large centralized dispatchable power plants with very predictable power flows from the large power plants to the cities to the households. You think about energy transition is one by one exactly the opposite. You have hundreds of millions of distributed DERs in terms of renewable, extremely complex by directional power flows, so the design criteria of the systems that run the grid today were designed for a completely different set of parameters. The answer is we need that because the environment and the prerequisites and everything that is connected to the grid is changing 180 degrees, from central to central, from dispatchable to intermittent, from unit directional to bidirectional. And we start to lose control of the grid. This confront system theory, if the speed of change in the environment of a system is going faster than the ability of that system to adapt to that changing environment, you will experience system instability and eventually system failure. And I think the Spanish blackout kind of epitomizes, for example, how bad things can get. But may I say that we see a huge amount in frequency on smaller outages. There's been a blackout in Berlin, out of sabotage, so resiliency is a problem. There's been a quite heavy incident in Italy, not recently. We have the Texas blackout, so Australia had a very similar blackout than the Spanish blackout because of very high overvoltage coming from rooftop solar and solar PV. So we see, besides full out blackouts like the one in Spain and Portugal, we see pretty large shortages coming. The speed of complexity of the system is growing faster than the ability of the operators to understand it and to manage it. This is why you need it. Alberto, sorry, I'm not as expert as you are, so I'm going to ask you a very simple question. So you have your system. What is the input and what is the output? output. And the output I guess is not a report per month. I guess it's online monitoring. So how are your clients using your output? Yeah. The best way to understand how the system works is have you been to a TSO control room and a DSO control room? Yes. Okay. So the TSO control room, that's a Plexigrid. So we define Plexigrid as a network with three superpowers. One is real time visibility. You can see what happens in each node of the network in real time. The second is real time analytics. You can compute power flows, state estimation, direction of flow, active reactive power voltages, the whole shebang. And then super power number three is orchestration of flexibility in real time, either in supply side or on the man side or on a storage side. So when you go to a TSO control room, you see, well, this is a Plexigrid but fulfills the definition. You see an interplay between these three superpowers in real time. They can see what happens. They compute power flows. And even when something is about to reach the limit, whether it is frequency or voltage or power, they will start to flex things, typically hydro power plants, gas pickers and so on. These days, they are also flexing batteries or you know, aggregators and other things. If you go to distribution, they are called surveillance centers is something else. You don't see this magic of three superpowers in the plane in real time. And from substation and below where you have 90 plus percent of the kilometers of network, nearly 100 percent of the DERs, the visibility is extremely limited or zero. So they are called surveillance center. They don't run the network actively. So the input, Lauren, is that the DSO of the future will need to have TSO superpowers. And we need to run the real time flexibly. How do we build that? We take as input the legacy systems. So we connect to ADMS. So at least we can see what happens up to substation level. Then we connect to the smart meters, which we use as instrumentation. But in a utility typically, this is used for metering and billing, not for real time control of the network. But we use the mass sensors, if you will. And then we connect to the GIS Geographic Information System. That's where you have a map of the grid where the cables go and so on. Normally, these are three silo systems in a utility. The ADMS is used for mid voltage control. The GIS is used for repair and maintenance and field operations. If there is an notice, you need to know where's happened, you need to know what cables to bring and so on. And the smart meter is used for billing. In a big utility, you can put people from these three departments in a hotel room. They don't know each other by the name because they are complete silo processes in complete silo monolithic systems. So what we do is to glue together all these systems. And if you know what happens in the substation and you know what happens on the smart meters and if you know the topology that connects everything together, in theory, you can transform this into a real time digital doing and then you have a control system that looks like a TSO. And it's able to sense things in real time and it's able to rescadial heat pumps or EVs to avoid bottlenecks or over voltages or under voltages. You transform the DSO network into a TSO. That's the output. And the impact is much lower congestion. You expedite the connection cues by 70, 80 percent shorter. You can offer your customer more affordable, flexible grid connections. The new EU grid packet has an entire section saying that if the operator cannot provide a firm connection, you're forced to offer the flexible alternative. So we can enable all these new functionalities and requirements that is the future of the grid. Alberto, how much commercial traction are you getting with this at this point in time? And where I'm coming from is listen, you know this better. Now I do grid operators are notoriously risk averse. In fact, they hate risk. Okay. And what you're doing is you're saying to them, these change the way you're doing stuff. And radically. And are you going from pilot to pilot to pilot forever? I love this question. One is the culture. In the DSOs, you have incredible operators. That's a very specific competence. I always say that what makes a good innovator doesn't necessarily make a good operator and what makes a good operator doesn't necessarily make a good innovator. In a sense, the challenge is to sell change to people who desperately need to change. Otherwise, the grid gets out of control. But who are used to a definition of engineering excellence which is the fight against variability. So an operator, you know, you want to go slow, you want to think things one, two, and three times before you move things because that drives safety and reliability. And that's their mandate. That's okay in a stable environment with the speed of change of energy transition. The risk is not moving fast enough and being overwhelmed that was happening on the grid. The first thing that we have to work with is to redefine engineering excellence as going fast enough to be able to cope with the incredible speed of change going on in your network. And I think this is starting to happen. We see this change. Going back to the commercial traction five years ago when we started, people were laughing about us because we built something for which nobody has asked. Nobody laughs about us anymore. And we definitely have moved from the pilot. I love this expression that by pilot, one pilot and another pilot, nobody can make it. But things are moving. Well, things are exploding largely driven by regulation. In a regulated market, the regulator is the market maker. And if you look at the European regulation, it's moving in full from cost-based remuneration to performance-based remuneration. This is one in a generation change of the rules for new connections in the network. It goes from first in first out to first come first served. And if you cannot offer a firm connection, you have to offer a flex connections. These are regulatory requirements that cannot be fulfilled with our technology. Just to give you some numbers, as recently as three years ago, or pipeline was 8 million euro. That's nothing. Then it went to 20 to 40. Now we are at 160 million euro of pipeline. And we see tenders. Three years ago it was at market for pilots. Now we see multimillion euro tenders coming from the big players. We are not only competing in these tenders. We are winning these tenders and outperforming and out-competing some of the big names in convincing with control technology. So we see the inflection happening, but it has taken five years. Bertu, if you look across Europe, give us a view of who countrywide, who's at the forefront of this? And who are the laggers? I call these the migrants of the grid. The more EBS, the more rooftops solar, the more batteries, the more distance you deploy, the more dysfunctional the grid becomes. So long as you still are on grid 1.0 in terms of grid controls. We look at countries, the Nordics, for example, they have extremely high digitalized networks, some of the most digitalized networks in the world. Why? Is that the Norwegians are smarter than others? Well, I think they're smart, but they are not necessarily more smart than the rest. But they deployed millions of EVs in a small country. They have passed the mark on that more than 50 percent of the fleet is fully electric. And they don't have gas boilers anymore. They move to heat pumps, same in Sweden. So imagine to what all of us does to the grid. So they have to deal with some of these problems earlier than other operators. If you think about solar, look to Spain and Australia, but Australia, so Australia, every third roof is a solar roof. And they have reverse power flow in 220 KB coming from rooftop solar in low voltage. It's madness what is going on. So they had to introduce dynamic operating envelopes, very innovative systems to really regulate these rooftop solar inverters and put a little bit of order on the network. Why? Because they sit hit the fan before there than in other countries, because they have a much higher penetration of these things. Follow the migraine. So I think the countries that are moving faster are the ones who have most advanced energy transition, high penetration of the ERs. And it also helps, of course, to have a well-diutilized network and a progressive and forward-looking regulator. But normally, you know, regulators who have put an agenda for advanced energy transition, they also have quite progressive in regulating other parts, including the grid. So that makes the first move. Funnily enough, you go by geographies. Northern Europe moves very fast. And then it always follows with southern Europe. Spain is very innovative, Portugal, Italy. They have very competent network operators. Some of the most digitalized and advanced, you will find them in southern Europe. And then on the laggard side, finally, you have UK, maybe Germany, still deploying smart meters as we speak, where the rest of Europe deployed them 20 years ago. Just going to say Germany. And I will say France, because not that they don't need your system. With EDF, we'll be proud to spend 20 times inside to try to replicate the fail. And they'll come to you in 10 years, pretending it was their choice in the first place. Because it's all about control. And they think that giving you data is not good for control. Don't think about this as something that runs on the cloud. where we do smart thing, Google style with the data. Think much more as an ADMS that runs on the operator. In that sense, if I harm to defend an ADS or EDF or any other European DSO, they are very responsible custodians of the information that they are meant to custody. And so our system is designed that it runs on the operator. So the data never leaves the operator or AI models learn in context within the operator and none of the data leaves. Any DSO who wants to work with us, they don't need to give away the data. The data stays with them. Exactly as it happens in any other of the core systems. The GIS system may be supplied by general electric or the ADMS may be supplied by neither electric. They are the technologists, but the custody and the operation of those systems stay on the operator. In that sense, we are no different than some of these other systems. When you open a new country, how much can you use what's existing and how much needs to be tailored for the way the country is organized? I've loaded this incredibly genius designs because one of the requirements from our investors from the beginning was that prove that this thing can scale globally, given the fact that electricity regulation is regulated at national level and in United States at the state level. So you have 27 regulators in Europe and 50 in US. So Pablo set up the platform so that it can be configured to any local regulation in a matter of weeks. For example, to prepare the machine for market entry in New Zealand, where we are operating in Auckland, it took us three weeks to set it up. And I call it the regulatory religions. You think about flexibility. You will find some countries that are going for direct control like Germany. Most European markets want to run this through local flexibility markets. Not the least because EU Directive 944 that set the scene for the flexibility market. Aggregators and all of that say that the use of flexibility should be down through markets. So we can accommodate that and we can integrate in local flexibility markets. And then in Australia, New Zealand and very popular now in United States is this dynamic operating envelopes. Where you use or detail to in on real time to understand what is the limit in each feeder in each transformer. And then set up limits on, for example, the PV inverters so that they don't inject more than the local grid can take. If you go around the world, you will see any combination of these three religions or platform cannot accommodate. All of them is a matter of configuration. And everything else behind is totally standard. At the beginning of the conversation, you talked about a foundational model that you're developing, right? Could you dig into that? Because when I hear that, I think, well, that's what Chachibit does and stuff like this. So why do you need to do that? - Chachibit, they're large language models. They're extraordinary, incredible at reading, writing, understanding natural language and programming and coding because programming and coding is done in human natural language. They're incredible at all those things. But they have problems. For example, they have hallucination problems. They cannot run what AI scientists call counterfactuals. It's what if a scenarios? What if the penetration of EVs here go from these to that? What will happen to the grid? They have big problems on doing this. What if a scenarios or counterfactual simulations which are key to operate an electricity network? And they don't understand physics, really. They can do some physics calculations indirectly, but there's no such thing as 100% physics rounding. Second, there is other kind of models called world models. Jan LeCoon, for example, has launched AMI. These are physics informed AI, Nvidia has, for example, Cosmos, which is an open model for physics AI. Even those models, the grounding of the physics is by injecting physics on a neural network. They mimic physics. Some of the experts in real physics in AI, they will tell you that they don't understand physics. They mimic physics. In the same way that Rafa Nadal does very advanced physics calculation when he was playing against a featherer because they can see he just hit with a top spin. The ball is not gonna follow a straight line. It's gonna do a curve and I need to get running to the exact right spot. Now, otherwise, I will be too late. It's a lot of physics calculations, but I don't think Nadal has the physics equations in his head. It's just that by repetition and repetition and learning, he can do all this math on his head. They learn like this and that's not okay for physical infrastructure because if something was wrong and you do a decision on your digital twin to switch this line to that and it results in a blackout, you will have to explain exactly how it is that you decided to do that. It needs to be traceable, it needs to be auditable and it needs to be explainable. And the grounding of the physics needs to be down at the core, not injected on top. We are a foundational company because critical infrastructure requires a new model class. It's a new model class that is very different from LLMs. It's very different from the JPS of Jan Lecun or the Cosmos from NVIDIA. It has a set of unique requirements for critical infrastructure that you will not find in these other applications for robotics. Love it. Alberto, when I listen to you, what I'm concluding is we don't need humans anymore going forward. Is that the way you see it as well? Not that all in distribution grid. In this case, we have humans that are being overwhelmed because certain tasks are in human. To run the distribution network as a TSO and it has hundreds of millions of notes in the network or in redola or in L. In transmission, you have thousands of notes. For example, you have a team that manually creates every single change of topology every day. To do that in distribution, you will need an army of people. Therefore, it doesn't get done. In transmission, you need to dispatch hundreds of power plants. Now these guys in distribution will have to dispatch millions of power plants, tens of millions of virtual power plants. It's unhuman. It cannot possibly be done by human. So here we see is not that the jobs will disappear. New jobs will have to be recruited of young engineers and generations and talent and AI scientists to build and operate these new control systems that will be able to run network with hundreds of millions of notes, hundreds of millions of DRs in real time. These are new requirements that are unhuman. It cannot be done by humans. Therefore, they have never been done because the technology up to now isn't allowed for these things to be done. What we're going to see is a recruitment of a new generation of engineers, the control engineer or network control engineer 2.0, a generation said digital native that is going to work alongside the traditional operators and they're going to be the humans in the loop. Programming the agents, setting out the algorithms, tweaking everything, following up that everything is correct, and being there as the human in the loop to make sure that nothing goes wrong. - Well, listen, I just want to say thanks because now I know all my kids need to study. So thank you very much. Great having you. - Thank you very much, guys. It's been a pleasure. Always stimulating discussions with you too. - Thank you very much. But, Jard, I'm glad we have this podcast because we can really bring extraordinary people, extraordinary, there's so much claim going around. And I think you really crack the code because our distribution grid's the cost of fortune. They are totally blind. And there's going to be renewable energy, consumption, EV charging, everything. That's going to go through the distribution grid and he brings a brain. He brings a brain. - Yeah, listen, oftentimes you talk about investing in the grid, most people think of big transmission grid lines and interconnectors and stuff like that. And that's part of it. But actually, I think increasingly the most important part is sorting out the distribution grid. And as you use the phrase, there are other grid operators are blind. They don't see what's going on. And there's definitely huge room and space for optimization. And for getting more out of existing grids, the need to actually go and put new wires in which takes years and years and years. - And I think the reason of his success is not only technical and of course he talk about AI and financial model and the fact that even open AI can do it that they have to do their own models. But it's the cultural part. The cultural part, you know, distribution grids, they are very conservative person and the fact that he comes from this industry makes the whole exercise much easier. 'Cause you can arrive with the best software in the world if you're like a young, aggressive tech guy, it's not gonna work. So it's great to have like real professionals who can talk as peers. - Yeah, yeah, they speak the same language. You're absolutely right. And trust is there straight away 'cause they understand where the other side is coming from, so totally which is absolutely wrong, yeah. So we'd like to thank the BMW Foundation Airbus Quant for our collaboration. So we had an episode with them at the end of last year called AI and Energy, where they're not the one about European and Neil Cloud. So that's the last of the trilogy. Like the good, the bad and the ugly, maybe it's the best of the Sergio Leone trilogy. Although the Godfather number three sucked. So Godfather number two was better. and terminator or two was better than terminator three but overall it's a good trilogy. Exactly and then this and finally let's talk let's firstly thank Albert for coming out on the podcast and I wish them all the best over the next two years because their solution is important and it can help us really manage existing grids better than they've been managed a present and also save the customer money exactly child always a pleasure I talk to you next week the portrait thank you for listening to redefining energy don't forget to rate the show and subscribe on Apple podcast Spotify or the platform of your choice