Energy Renewed: Data Utilization

In episode 1 of Energy Renewed, Katie Janik of ICF speaks with Lynsey Tibbs of Southern Power Company and Jason Kaminsky of kWh Analytics to discuss operational project data collected from renewable energy projects. The conversation centers around reasons to collect operational project data, ideas for how to use the data, and challenges to organizing and using operational project data. We also discuss the ways in which investors and project owners differ in their use of operational project data.

Katie: Welcome to "Energy Renewed," a podcast by ICF. A meeting of the minds on renewable energy, where people come together to discuss ideas and synergies to propel the industry forward.

Full transcript below:

I'm Katie Janik from ICF, and the host of Energy Renewed. ICF provides technical advisory services to lenders, investors, and project owners for renewable energy technologies and processes. In this podcast series, we will consider varying viewpoints, ranging from policy topics to equipment components.

Hi, there. In this episode, we're discussing operational data for generation and renewable energy projects. Most project owners and investors have large datasets collected from their projects. As a consultant, almost every client asks for support around data, from data sourcing to data metrics indicative of performance. Today, we will discuss some ideas for data utilization, and we are joined by two industry experts to help with the discussion: Lynsey Tibbs from Southern Power Company and Jason Kaminsky from kWh Analytics. Hi, Lynsey and Jason. Lynsey, will you introduce yourself?

Lynsey: Hi. Good morning, Katie. My name is Lynsey Tibbs and I'm a project manager at Southern Power Company, which is an independent power producer. I'm part of our solar operations group. We have a solar fleet of about 2,300 megawatts across 6 states in the U.S., which totals about 28 sites. So, our day-to-day operations are really making sure that those sites stay online, stay healthy.

Jason: Yeah. And I'm Jason Kaminsky. I'm a former solar investor, I think I'm here with the investor perspective today, and before that, worked as a solar developer. But today, I'm here in my capacity as the chief operating officer at kWh Analytics.

So, in kWh, we're very close with the financial investment community on risk management, and really, data management. That's really the two primary avenues. So, on the first, we work with investors, think of lenders or [inaudible 00:02:06] equity investors that have grown portfolios from...really, the industry wasn't much 10 years ago, to now, tens of thousands or hundreds of thousands of projects and managing the data for internal risk management purposes, and we'll talk a little bit about what that means.

And then second, we've taken that data and developed a unique and kind of proprietary underwriting approach to have a view on production risk, and we use that to really help lenders take that risk and move it to a different market, the insurance market, and we can talk a little bit more about how that works as well. That's through a project we call the solar revenue [inaudible 00:02:43].

Katie: Great. Thank you for being here. I'm excited that your backgrounds are so well-rounded, because I think it will give us a diverse set of ideas for data utilization. I realize that data is a large topic and we could spend hours talking about it, but for now, I'd like to focus on reasons to collect data and some ideas on what to do with operational data. Let's start with perspectives. Jason, what perspective do you bring to the discussion and what is a reason to utilize project data?

Jason: Yeah, it's a great question. So, I'll represent the investor view, so think of this as a lender [inaudible 00:03:15] equity investor and operating in a highly regulated environment where you're making very long-term investments in what's considered project finance. So in order to get paid back, these projects need to work, they need to be collecting revenues from the project in order to really pay off the investment.

So there's a large scope of work that bankers are going through to really keep their internal management and their internal credit team educated about the portfolio, keeping an eye on the health of the portfolio, and at least in the banker community, we call that asset management. It's a little bit different than what Lynsey does on the sponsor side, but asset management is a really core focus for the bank and financial investor community to know, is this asset working as I expected? Do I need to write it off? Do I need to restructure it, or am I gonna get paid back just fine? But I'd say that's probably the overarching goal, and associated with that is really a lot of compliance work. Banks have internal audit groups. There's a lot of different groups that have an interest in knowing how these assets are performing.

Second is really underwriting, so if you know how your existing fleet is performing, it will use that data and use those insights to make better investments in the future. And the third category that we're really seeing emerging is buyouts. So we're at year five, six, seven, on some of these very early investments and knowing how really your solar project has operated to date is now an input into the negotiations about what it's worth and how much does the sponsor need to pay in order to buy the investor out of the project.

Lynsey: For us, Katie, Jason's right, our perspective is slightly different, but really, they're in alignment. We're looking at asset management from the perspective of optimizing the assets we already own, and so the assets that we have on the ground had a business case behind them, we had a revenue assumption attached to that. Are we actually meeting that revenue assumption? Are our plants at the optimal place to be producing the energy we predicted? And so, if we don't have data, if we don't have the insights that we can pull from data, we're really flying blind. We don't have a way to know if we're meeting those goals.

In addition to that, the way we maintain our plants really goes into how we optimize them, and so, these data insights help us prioritize that maintenance. It helps us go into the field with a purpose, really, on a condition base maintenance program rather than through traditional 100% field checks where you would go out in the field and test every combiner box. Here, we can really target different areas of the site and use the data to tell us where to go first.

And then lastly, we've been, from an internal perspective, trying to make sure we're the best at what we do. We're optimizing our fleet to the best of our ability, but you really don't know if you're doing that or not unless you compare with your peers. And so, we've begun the process of understanding, from a benchmarking perspective, how we compare with our peers, and really looking to groups like Jason's and others who are providing benchmark analysis to help us understand where we stand in terms of optimizing our assets.

Jason: Yeah, I'd echo that, Lynsey. I think that's a great point. We're seeing a lot more activity on the benchmarking side [inaudible 00:06:27] in the last 12 months, and the core question is, I have a project, I see 5 other projects nearby, and I don't know if my project's doing what it should do. Is it performing to expectations? Was the estimate right, going into it? So, we've been approached by a number of people that own projects and said, "Hey, can you help me answer those questions?" And we said yes, and we went back and looked at it. And there's a lot of different ways you could benchmark a project. You could try to normalize for weather, you could try to do a lot of...call it fancy analytics. But where we shook out is there's really two key metrics that we should be looking at the start to really answer that initial question: is my project performing to expectations, and how's it doing relative to...I'll call it the industry or other projects?

So the two metrics that we're looking at on the benchmarking side, one is specific yield, the variant for capacity factor, but it's basically, how much electricity am I producing per unit of capacity? And the second is the performance index, so like, what is the site...or how is the site producing relative to the estimate going into the project? And with those two, we're able to really then discern what is going on with the project and help provide richer insights to [inaudible 00:07:38] like Lynsey about how is a project or a portfolio performing and what else might be going on with it.

Katie: Okay. Thank you for providing specific metrics. I think that will be helpful to listeners. Beyond benchmarking, a recent buzzword in the industry is probabilistic model. Jason, will you describe why probabilistic models are helpful and how you use these models?

Jason: Sure. Yeah. Probabilistic model is really just a fancy word for a statistical model. It's kinda interesting because when you look at a financial model, for those of you on the finance side, you know it's an Excel model where you put in a number, so you're kind of forced to say, "This is my best guess, my P50 of what's going to happen."

A probabilistic model gives you a whole range of outcomes. So, in our world, we model solar production. That's how we use solar probabilistic models, and the reason we do that is that at a high level, pretty much every solar project has a fixed price agreement, so think of it, PPA, but the revenues are still gonna fluctuate with the quantity of production, and when lenders look at that they're still looking at the volatility and giving a big discount to the forecasted volumes that are being produced.

So we use probabilistic models in our group to determine the likelihood that a given project or given portfolio is actually gonna produce a certain amount of electricity, and then, we use those models to guarantee a floor on the quantity of production. So if it underproduces, an insurance company's actually gonna come and put cash into the project. When you take a fixed price and you take a fixed quantity and you combine those two, you get a fixed revenue stream, and that works really, really well for lenders. They really understand this. And what we're seeing is our clients can get more debt when there's this fixed revenue stream.

So we're looking at...I'll call it the probability over a pretty long timeframe, over a year, that a solar project or portfolio is gonna produce and then we model and forecast that 10 years forward. So that's kind of the...I'll call it the lens that we have on the market.

It's really quite important. I mean, in our conversations and in our database, what we're seeing is that it's not uncommon, actually, for a project to produce below what the financial investor thought was the P99 going into the project. So you know, the 1 in 100 years...the current is happening more than 1 in 100 years. Actually, in most years, it's happening 6 out of 100 times. So we're using these probabilistic models in a statistical manner to then provide...call it risk assurance to a lender that really has to pick a datapoint and make a financial decision out of that.

Katie: Okay. So, you're using probabilistic models at a high level from a financial perspective and aggregating thousands of sites in order to create the probabilities. Lynsey, how do you use probabilistic models?

Lynsey: For us, it's really all about predicting and preventing failures. So, as I mentioned earlier, we're trying to optimize our assets, make sure that the entire plant is operating as we expect it to, as it was planned to do, and so, we're using probabilistic models as a part of a larger health plan that we have in place to tell us whether or not our plant is operating properly.

So, we really have five things that we're looking at when we look at plant health. One is just data analytics, different key components that we can pull from the data; field checks, so going out into the field and actually having techs check specific equipment; IR scans, whether from the air, from a drone, or even standing in front of particular equipment; and degradation studies.

So there's four of our health plan items, and then the fifth is probabilistic modeling. It's really the most unique of the entire set, because while the others allow us to see into the past, what we have already done, what equipment is or isn't performing correctly, probabilistic models give us the chance to see into the future and see if we can predict and actually prevent a failure from occurring.

So, what we have right now are a combination of noncritical and critical-type probabilistic models and they're all around specific plant equipment. So, some of our noncritical components are things like combiner boxes, revenue meters, weather stations. Anytime something is off with those pieces of equipment, generally you don't need to call the plant and have a tech go and check that situation out, but there are more critical situations or critical pieces of equipment on the plant where you really do want to know right away if something is outside of its normal band, and that's really what these models tell us. They tell us, "Hey, this is not the normal working condition of this piece of equipment. It's trending towards a failure; please go check it."

So, some of those critical components are inverters, medium voltage transformers, GSUs, the bushings on GSUs, or breakers within the substation. These type things give us an ability to get ahead of large, costly failures that not only hit our capital budget but hit our revenue stream for the future. If we can prevent that, we're really taking quite a step forward.

Katie: A takeaway from this in relation to probabilistic models is that you can use it on the financial side, and you can use the data models also on a site level or asset level side where you are looking at the individual components. So, when you are looking at all of this data and you are sorting through it or sifting through it, did you take this on internally, where each asset manager is responsible for the data in their portfolio, or did you go to third-party advisers to help you with the data analytics part of it?

Lynsey: A little bit of both. So, we're mainly taking this on as an internal effort with, certainly, some software consultants and other industry consultants helping guide us, but for the most part, we're taking this on ourselves. We have a team of engineers around the particular software we use, as well as a team of performance engineers working in the background, specifying all of this information so that we can kind of stand these up. But it has been a team effort. I will say taking it on ourselves was a big undertaking, probably a little bit bigger than we expected to begin with, but now we know so much more about our data. I think it was also a valuable exercise.

Katie: You mentioned degradation. If we were talking about performance indicators and metrics to consider modeling... Let's talk more about degradation. Jason, how does degradation as a datapoint figure into modeling and performance analytics?

Jason: Yeah, degradation is an interesting one because it's a really subtle signal among a lot of data, as Lynsey mentioned. You know, I've been in this industry 10 years, and we've really used the same estimate and financial model, plus or minus a little bit, [inaudible 00:14:31] for the last 10 years. It's not because we're super, super certain that that's the right number to use; it's because, frankly, we don't have a better number to use.

So, we've been fortunate to collaborate with the Department of Energy on a big, I'll call it, cross-industry study to say, what are actually observed system degradation rates out in the industry, and then, can we identify drivers that are leading to those differences? So does a roof [inaudible 00:14:56] project behave differently than [inaudible 00:14:58] project? Are we seeing different degradation rates in different climate zones, etc.?

And you know, the goal is to, as an industry, really grow up and say, "Okay, if we're gonna be the primary source of power capacity going in the ground for the next 20 years, how do we make sure that we have the best information available to then forecast how these plants are gonna perform for their useful lives? So, it's been a really interesting set of work that we've been able to undertake on that front as well.

Katie: I can see how degradation and the assumptions going into degradation can be a challenge. Just taking this a little bit larger or more a high level, I can see how using the same assumptions for multiple years, for a decade, and not really understanding what goes into that assumption, that is a challenge. Lynsey, what other challenges do you see in using data?

Lynsey: So, we've touched a little bit on the size of the data, but it really is mind-boggling to think about the amount of data that you have coming out of a site that has 3 million panels, and hundreds of inverters, and back of module temperature probes, and weather stations scattered throughout the site. There's just so many datapoints for each of those pieces of equipment, pieces of the site, that when you pull them all together, it's really mind-boggling.

With that, there's also no industry standards for how you might take that data and make conclusions from it. And so, that's where we've really struggled is, how do I know which data pieces to take? How do I know what key performance indices to really look at? And deeper than just availability or performance index, what kind of things are gonna tell me if my plant is healthy? What kind of things are gonna tell me if I'm getting the return that I expected?

And so, what we've done through the five categories that I listed above, we've really tried to map out how we might know what parts of our site need attention. And then, the task is to go behind the scenes and figure out what tags coming from our SCADA system might we pull together and calculate different performance indexes from that to understand how that particular equipment is working.

So it's been a bit of putting together a jigsaw puzzle that didn't exactly have a picture on it. So, you've got all these pieces, but you're not exactly sure how they go together. So, I would just say, the size of the data, along with the fact that, as Jason said, we're still rather a new industry and there's no textbook out there that you can open up and understand all the calculations that you need to be able to understand the health of your site.

Jason: I would add to that, though, that there are some encouraging datapoints I'm seeing, at least from the culture of sharing data. So if we look at every other major industry, look at consumer credit, you have a company like Experian, you look at mortgages, you have a giant company called CoreLogic, and all those companies have been established to really be, I'll call it, trusted custodians of data, and it has enabled the financial community to make much more informed decisions about investing into those segments.

We see that in energy, too. If you're an investor lending to a gas plant, you can get data on how a GE turbine is gonna perform and you can use that to inform your investment. And what we've seen is really, in the last...I'll call it the last five years, there's been a shift and maybe a realization that not all data is created equal and there really is a spectrum.

So, on the one end, you have personally identifiable information. That's consumer information, heavily regulated, most people shouldn't and don't wanna share that unless absolutely necessary. And on the other hand, you have public data. You have information that the IA publishes, and somewhere along that spectrum, is the right point for every institution to say, "I'm comfortable sharing certain types of data with a trusted custodian, but maybe not all types of data." So maybe, "I don't want my investment returns out in the market but operating data that talks about how my plants are performing is maybe less sensitive."

That's really where we've established our business is to say, there's a lot of this information, that really, at its core, is not a business secret or trade secret, but that when looked at the industry level can really add a ton of value back to the industry, to help refine a lot of these assumptions and allow us to, I'll say, land on an even footing with some of these more established gas and other industries that have really just been operating in a type of regime for much longer than us. So it's been heartening to see, I'll call it, a cultural shift that's taken place in the market.

Lynsey: That's right, Jason. I'll add to that a little bit. As a large industry participant, I think we have learned as well that you have to give to get back. But I will say, all of the efforts, Jason, your group, other groups that have allowed anonymized databases to be put together, where you can provide your raw plant data along with its equipment specifications, location, all those type things into a database that allows you to come away with insights but not necessarily drill down to the exact site that's being referenced, are really helpful. I think they're helpful to making large participants be comfortable that their data can be shared without being exposed.

I'll also echo your sentiments around the lack of competitive nature with a lot of this data. Sometimes I wonder just what would we do we that if someone got it. How would they really use that to get a competitive edge? But you're right. Every group has to make their own decision about what is competitive intelligence for them, and once you make that decision, I think, you know, after the plant is built, once you're in operation, so little of that is competitive. We need to really, as industry participants, consider what we can share, because I think we can all get better and really not lose anything in the process.

Katie: I would agree. I think that three years ago, four years ago, we were starting the conversations about sharing data, and we've made so much progress in sharing data, and I've seen more and more companies being willing to share their data because it's been aggregated and anonymized. So, I think those are great efforts that we've seen results from.

Well, I feel like we've only scratched the surface of this data conversation, and I wanna thank you so much for being here today. I'm so happy we could do this.

Jason: Thanks, Katie.

Lynsey: Thanks, Katie.