Doing ML at a leading renewable energy company by Sean Otto (The AES Corporation)
Sean Otto of the AES Corporation goes over what AES does and their philosophy and the future of ML/AI.
Talking Points:
Speakers:
Sean Otto, Director of Analytics, AES
Read the Full Transcript
Sean Otto:
Good afternoon, everybody. Good, I can hear myself too. Sean Otto, director of analytics at AES. AES is probably a company you haven't heard of. We are a leading renewable energy company here globally. Let's go ahead and get a slide up, if you don't mind. There we go. Today I'll quickly talk about AES. A little bit of business, some AI projects, organization, team, some challenges that we have. Overall, this won't be a technical conversation, a discussion, but just to give you some ideas of what are the challenges that we have and honestly, the amazing amount of fun that there is in renewable energy. And I would not go work at a tech company. I'm happy to be at an energy company. There's lots of stuff to be going on in this space. Oh, I get this. Thank you.
So, thinking about a title here. Renewable energy and everything that you've heard in the world and politics and stuff. It's on the forefront of many different things with this idea of climate change and as I was putting this together, I've got interesting little quotes that sit at the bottom of it. And I may speak to those a little bit every now and then, but I think I'll start with this one here. So, when we think about AI, we really don't know how it's going to impact society. How will it direct change in our lives? How is it going to shape things? I really enjoyed the prior conversation today when he says, "We use, how can we have AI help democratize science? I'm like, "Oh, that's an interesting idea." So I don't know how AI is going to continue to change how we think and how we structure ourselves in this world, but I like that last statement. I mean, there may be some uncertainty on this, but it's going to be cool. Okay and I think as geeks, we can all agree with that. It's going to be cool and fun.
What is AES?
So AES, our whole goal is to accelerate the future of energy together and when it comes to the work that I do in the organization, it really is about how do we leverage and deliver AI/ML models to accelerate that future of energy. And even though we can leverage it to help in this world for social good and AI for good, it too, ironically has an energy problem. I mean, it takes computers, it takes processing power, and stuff to do. And I was learning about Nvidia and this digital twin model that they put together, and they ran it on 6,000 GPUs. All right, that's a lot of energy to think about what they were running on. So even though we talk about renewable energy and accelerating that future of energy in the back of my mind, how can we leverage AI to help us even solve the energy problem that we're creating at the same time?
So who is AES? A company you probably haven't heard of? We are a global organization. We're across four different continents today. We're about 14 countries. We generate 30 gigawatts of energy or larger. Now you think about back to the future and Emmett Brown saying 3.1 gigawatts. It's amazing. Well we're 10 times that. All right? And we're looking at literally growing six gigawatts a year. We've got ambitious growth goals. And let's see, going on to the next one here. What do we do? And this makes it challenging when I talk with individuals about AES and where does AI fit into it? And I think AES is a good company to understand the breadth and the landscape of where renewable energy is, how it impacts our world, and what we can do with it.
Power Generation
So first is about power generation. I'll get into some examples here. We generate power, as I mentioned. We own a couple utility companies. Two in the US and one in El Salvador. We have commercial development, trying to grow renewable energy, solar systems, battery energy systems, wind farms and then we also trade that energy.
First of all, power generation. What are the opportunities there? Solar. Lots of solar opportunities. Battery energy storage system, hydro plants, natural gas over on that right is a picture of a natural gas storage facility that we have in the Dominican Republic. And one of the challenges that we have is how do we forecast proper energy demand in the country and then also make sure that we're supplying that energy at the. Supplying liquid natural gas to that location at the right time. And that can be a challenge when you're looking at a variety of timetables and other things that impact it. Particularly with natural gas prices and how that's changing too. We own a variety of wind turbines out there. Roughly 1300 wind turbines today. So we have wind generation, we do have some coal. We have a strategic goal to get out of coal here very quickly and I think we're making good strides to that.
Hydropower. So that's on the power generation side of what we do and there's lots of interesting opportunities for AI inside of that. Battery energy storage systems. We have one out in Hawaii. It has 1,024 inverters. So I somewhat liken this challenge on the solar side and the battery energy storage system to the, to your data center problem. Which is not, when is a hard drive going to fail? When is a CPU or your system going to fail, but it's going to fail. What do you do about that? How do you proactively prevent and manage that failure? And battery systems, as we start building these gigawatt ones and larger, you've got a lot of equipment inside of there and battery degradation is an issue. Inverter performance is an issue and so those are all things that we need to monitor and be able to manage at a level of artificial intelligence and machine learning.
Utilities/Smart Grid
Whoops, go back one. Talked about Smart Grid. All right. Utility companies that we own. We have some models in place today around tree trimming. When do we go ahead and trim the trees? When's the right time? What are the opportunities for failure within the vegetation? How do you analyze vegetation growth? That's important. Reduce the outages. We all love electricity until it's not there anymore. So how do we reduce the outages? How do we maintain our grid and keep it reliant? Smart Grid, alright. How do we have smart systems? How do we engineer for smart systems? Electric vehicle penetration and usage? That's a very interesting use case for artificial intelligence. Trying to identify clusters and propensities of people who are willing to adopt EV. But at the same time, if I have a group of people willing to adopt EV, I as a utility company need to understand what's the impact on all of the equipment that I have on my transformers and my substations in the power systems there. Because that's going to be different than what it was originally planned for. And that too takes some planning.
New Innovation Services
New services, new innovation services. One of the things we're trying to do is how do you encourage people to get onto electric vehicles? Right? We have a company called Motor: M, O, T, O, R and it's designed to help people test out an electric vehicles. We have it in Indianapolis and in Salt Lake. Where we have one of our offices and trying to just rent one for a month, price, everything included. Go figure it out. Is this something you like? Is this something? How would it change your life? And how can you go ahead and transition into something like electric vehicles?
And I think this comment here from Christina Lund, our US president, our customers and vendors are trying to figure out how to do this too. And I think that's important that we're all in this together. Back to my statement about one of the key things AES is trying to do. And it was interesting and she said this just last Friday. Honestly, we were at a little coffee chat. She made this comment and I'm like, "You know that really is it? How do we do this together? But everybody's trying to figure this out. This is a new world for us and we don't have answers, but we have lots of data and we need to bring that data in and help leverage it in quick and efficient ways so that we can answer these questions.
AES AI/ML Philosophy
So what is our philosophy here? How did we go about starting this journey? So AES jumped on this journey about three years ago and I came in and started to build up a team. My background is a little unique. I am a behavioral scientist. I have a PhD in psychology, right? So I know some things about research. And as I transitioned through life, as we meander on our paths, I got into engineering. I got into the more physical sciences side of things and I had the opportunity to come into AES, which I consider more of a physics type of company with the power generation. And as I started working with the teams and the different peoples started to think about how we want to embody this idea of AI at AES? And I think these have been pretty much the core over the last three years.
Successful projects start with an idea of the community. So ideas always don't have to come from us. Conversations are good to start with and to. And sometimes you might have a good conversation, but you're not going to start a project for two years. All right, I've got something that I'm sitting on. I'm like, "That's so cool, I can't wait till we get there." All right? And some things that we can do. There has to be full trust on data and process. And that also means that you have to be fully transparent in everything. Our data engineering team runs on two week sprint cycles. Alright? Do we move as fast as the business wants? No, but we're consistent, reliable, and you know exactly what we're working on every day of the week. And that's that transparency that's needed and it's the trust that we continue to deliver at a consistent speed.
Minimize technology debt. This one I don't hear too much talked about. H2O is a nice tool in that it does minimize that technology debt. It allows people to get in and quickly do things. And we can search through 500 models or we can click a button and let H2O do some magic for us and get us started in the right direction. Like with the AT&T conversation that we had this morning. Knowledge is shared. I blame universities on this one in a unique way. We're design. We teach people how to give out reports, but we don't teach people how to document something so that somebody can come behind us and learn it and do it. People rotate through positions we rotate in our own lives. And so documentation has to be part of that deliverable that you have and part of your process.
I throw a documentation day on the calendar about once a quarter and the data scientists are like, "Thank you." Get rid of all of your other meetings for that day. The only thing you're going to do is just put on your favorite songs and just bang away on some documentation.
Change Management
Business either owns or partners on applications and visualizations. And I think that was talked about this morning too on the AT&T conversation is that when we start bringing in the business then they see the value. They champion that value. One of the individuals at our company was talking about change management. And I liked how we phrased it very simply. Very much from an engineer perspective. Change manage about getting people in on the front end. So they're with you on the back end, right? And that's the core of change management.
I find that we don't talk about change management with our businesses and the groups. One thing I'm prone to say in the organization is, "I can solve world hunger, but if nobody listens, it will never happen." All right? There's still a lot of things that we have to do, like conversation today about AI for good and what we do. How do we engage people to leverage some of the beauty and the value that we are creating and crafting? We know it's not perfect, all right? But we know it gets us closer to a destination. And I truly believe it's similar to the answer the Cheshire Cat gave. "Well, do you know where you're going?" "Not really, but we want to go somewhere." Okay? Doesn't matter how you get there, just start moving. Data products are a journey, not a destination. Reinforcing that idea. Once again, minimize technical debt. I got it there twice.
Transparency
And with that, I'm going to go ahead and talk to this comment or just say it here at the bottom. To reinforce this idea of transparency. Secrecy is the underlying mistake that makes every innovation go wrong in Michael Chrichton novels. If AI happens in the open, then errors and flaws may be discovered in time, perhaps by other wary AI's. And that's the idea there of responsibility and ethics and bias in AI. We need that transparency, we need that openness. And going back to my comment around universities, unfortunately we teach as students, we teach you to go grab something, do it all on your own, all right? Protect it from everybody else so that you can get a grade and then give it back out. But that's truly antithetical to how we want to work in the world. So I think there has to be some retraining as we bring people in to leverage and utilize data science.
What Does the Journey Look Like?
Little bit to that idea of what the journey looks like. We had a project that we started and this is one vegetation management. We have people in vegetation management. 30, 40 years. I know my trees, I know how to cut trees, I know when to cut my trees. Well, why are we still having problems? Well and they have all their excuses that they can in the world and there is that human expertise and it's vitally important. And as we go through this journey, I think you may agree here that we get to some type of data driven approach. It's a little bit more report based and there's some apprehension about this. Should I trust this? Should I not trust this? And you give them a tool and they're like, "This is great" and then tomorrow comes and they're like, "Well, things changed."
I'm like, "Well they did change cause that's life." So they're like, "Well, I'm a little reluctant to use it again this next time." And so then that comes into changing process. So you need to work with them and embed things into the process. Even with technicians on repairing wind turbines, right? When you have a technician who's seven days out and says, "I'm supposed to look at these 10 turbines, all of a sudden I'm going to throw in a new turbine because the alarms went off and that this one's important." How am I going to convince them to walk and climb up 80 meters and I've been on a wind turbine. It's not easy. You just climb on up. It could be hot, it could be cold. You've got other safety factors to consider at the same time too. There's some reluctance there.
You're changing the process when it gets to the real world and it's helping them understand that we are driving towards insight. And when you're able to flip that on their head and we work through this whole process with the vegetation management group, all of a sudden one of the guys goes, "Oh, this is so cool. I've done like eight of these in the past and they've all failed and you guys have all sucked, but this one's cool. You've helped me see the value of that at the end of the day." And I think that's true. That's what I really enjoy is being able to see people say there's value here.
What Tools Do We Use?
What are some of the tools that we use? We use Google Cloud with Python, Domino Data Lab, Rasgo, and yes, we love and use H2O. Are you using it as effectively as we can? No. To put a little comedy in here, one of the data scientists. We had a team meeting and complained how some of the business wasn't using something that he created. And I said it's about as easy as to get a data scientist to use H2O. And he goes, "Ha. Okay, touche Sean." Even then we have challenges of getting data scientists to leverage H2O and the value that is there too. And this is my little comment here. Everyone wants to do their own thing. Actually that doesn't work with data and for AI. AI and data forces you to come together.
Challenges
And we should embrace that idea that it forces us to come together. What are the challenges that we have in the few minutes that I have left here? Finding a new place for humans. That I think is a challenge. Where do humans sit in this world as we think about climate change and sustainability? How do we interact with our world? We're going to have to interact with it differently. And just like the vegetation management and even the technicians that we're bringing models to, they question their being a little bit, right? Get a little bit more philosophical here, but we are. We have an enjoyment in our life. We live our world, we enjoy what we do, and somebody comes along and changes that. That's not easy, right? And AI should be disruptive, but how do we make it disruptive in a way that pushes us forward?
Balancing this need for automation and non-auto automation. We need humans in the loop, but Hollywood's really good at selling this automation without humans. All right? And we need to figure out how to convey that balance to our teams. How do you manage the fuzziness of behavior with the rigor of controls? Behavior means people and how we respond and how we do things. Yet we need a rigid system because that's how AI responds best. Something formulaic, something process, something that never changes yet, people change, they do things differently. And we think about the weather. Weather's always changing and causing problems on models and stuff that we create. I think we need to take a look at how we redefine IT departments and how they support AI and ML. I think I've been very blessed in this company because I sit inside of IT, but I am not IT. Ironically, they come to me and ask questions a lot for things, which is nice. And I am in what's called innovation and solutions. So I'm there on the IT side to push the boundaries with the organization. To add value.
Democratizing Data
How do we get to this democratizing data mindset a little bit back to the idea of sharing. How do we get to that openness to democratize data also needs a little rethinking. How do we do it at universities, right? As a professor, here's your data. Everybody grabs it, tosses it onto their laptop, does their own magic with it? I mean, the idea of what we're talking about with the H2O future feature store is none of that. It's like, "Oh, go see what all of your other fellow students did in the features they created and what was going on in their minds." I mean, to create AI and ML more collaboratively and to teach it collaboratively is what I think is important. And driving that mindset is going to help adoption and help us lead a future where we can integrate it well into our world.
That point. Avoiding the sins of education. Things are taught to us that are not always what we should be doing in an enterprise. All right? From a university perspective, how do we help identify those and reteach people as they come in? Tool acceleration and I got 30 seconds here. Tool acceleration I think is a big challenge. Tools are coming out faster. Changes are happening faster every single day. Even in the conversation this morning from three, I'm like, "Well, I guess I got more learning to do." He's throwing out more things, which is great, but what do you think about it? Leverage it and put into your practice there. And then the last thing here, balancing turnkey: meaning something very rote and simple. Just plug it in and go. Versus the real scientific rigor that you would have in executing a solid project too. So with that, I think I am done.