How Data Analytics Is Transforming the Utility Industry

[Music] Broadcasting from Boston, Massachusetts, the ARC Digital Transformation Viewpoints Podcast is the only podcast dedicated to all things related to digital transformation and energy, industrial, and critical infrastructure applications. The podcast is the creation of the ARC Advisory Group Digital Transformation Practice. ARC advises leading companies on technology trends and market dynamics that affect their business. To engage further, please like and share our podcasts or reach out directly on Twitter at ARC_advisory or please go to the website at www.arcweb.com. Welcome to an ARC podcast. Today's topic is data analytics in the fast-changing, electric utility industry. My name is Richard Riss. I'm the Director of Consulting at ARC. In my background, I'm a process-control engineer. I've worked in oil and gas, chemicals, power generation. I've even worked on nuclear power plants and wind turbines. As a market analyst at ARC, I cover grid automation, micro grids, grid scale inverters, operative training simulators, and a few other topics related to energy as well. I'm also a light commissioner in my small town of Princeton, Massachusetts. We own and operate to 1.5 megawatt wind turbines for the last 15 years, actually. But today, we would like to introduce our main speaker. That's Daniel Foster. He is an industry principal at SEAK from Toronto, Ontario, Canada. Daniel, can you introduce yourself? Thanks for the introduction and having me on today, Rick. I've been looking forward to this discussion because I know you have deep experience in this space, clearly from your background. But I'll give myself a quick intro here. So, yeah, my name is Dan Foster-Roman. I lead the power industry practice here at SEAK. For those who may not be aware of SEAK, it's industrial and analytics. AI platform focused on time series data that helps utilities and energy companies solve some of the industry's most important challenges of the day with analytics and better leveraging that mountain of time series data that they collect in historians and skaters from meters and everywhere else. Personally, I started my career very much on the more hard-hat side of the industry. I worked as a system and performance engineer at a large nuclear plant in Ontario. I helped start up a and manage a fleet monitoring and diagnostic center for Ontario power generation. That was really where I started to apply engineering and data science principles, kind of together to model equipment and predict and avoid failures. That got me over time, moving deeper and deeper into analytics and machine learning and software. Now at SEAK, I get to work with a ton of different large power generators and transmission distribution utilities and even large OEMs around the world to help their engineers scale those same kinds of analytics across their fleets. And these days, with the kinds of analytics and AI tools we have available, that happens much faster than it did in the past. So, it's an exciting time. Great. Great. So, let me ask you a few questions, then, Dan. You've seen about a decade of relatively slow growth in the power industry in the U.S. Not a lot of great demand, a lot of efficiency LED lights, efficient appliances, tended to have growth really flat over the last decade up until about a year or two ago, same in Europe. But now we see an unprecedented demand for new power driven by these amazingly power hungry data centers that want new power in months. And they don't want to deal with the five-year planning cycle or 10-year planning cycle that we see for utilities that are adding such new large loads onto the grid. But demand is also coming from powering electric vehicles, EV charging stations, building heat pump HVAC systems and even some electrification in industry and commercial. So my question for you is, how can utilities predict and plan for this incredible load growth? Yeah, that's a great question. Totally right. I mean, clearly in a different world today than that flat demand era that we've gotten used to. I was some forecast I've read, our saying data center consumption just alone is expected to roughly double over the next five years. And we're already hearing of how large and how much pressure that's already putting on utilities. The center alley in Virginia comes to mind that's shouldering a lot of that demand growth as well in just one county. But to your question, I would say that planning for that type of load growth is certainly challenging and predicting it is kind of an underlying subproblem that's even more so challenging, right? You mentioned EVs and heat pumps. Those tend to follow adoption curves in some way as well as some familiar daily seasonal patterns depending on how they're used. But data centers are kind of like these big lumpy, discrete projects with steep ramp rates and even 24/7 profiles. You know, they follow business needs for AI training or whatever they're being used for. So from the utility side, the utilities I see doing this well, cover a number of fronts with analytics. First, they kind of treat those large load interconnections as a forecasting problem. So using scenarios and probabilities for one project's materializes one piece. So you can kind of have a spread of what can happen across different situations. You know, you allow you to better characterize the risk. But secondly, and actually probably where I've spent more of my time is pushing forecasting down to a much more granular level. So forecasting at the feeder, substation or transformer level and then across different timescales. So, you know, to your point, that one, two, three, five year horizon that utilities need to shift or bring forwards. One example I can share, a great example actually is with Dominion Energy. They are supplying power to the data center alley I mentioned before. We've been working with their transmission planning and operations analytics teams for a few years now. And that's exactly what they're doing. So they're building transformer level load forecasts and data center load pattern recognition models within Seek. And so, you know, the scale of what they're dealing with just to kind of give you a picture. They presented this at our customer conference. But basically the equivalent of 70% of the world's internet traffic. The internet traffic goes through that one county. And so you can imagine the challenge for Dominion is with margins on transformers and lines shrinking as those loads come online. And so that granular forecasting can really help with outage planning. So knowing where, you know, there's enough room to take equipment out of service as well as with that, you know, long term capital planning and investment strategy for new substations in upgrades and long lead transformers. So I guess my take is that, you know, forecasting and accurate forecasting is basically central to planning and tackling this problem. The key there is being granular. So your analytics are more tightly connected to the operational investment decisions that you have to make at the, you know, at the equipment level. Yeah, I think that we see the relationship between those data centers that are building out their own micro grids, all right? Because they want to build that power faster than the utilities can. So to some extent, it does pose a challenge for the utilities to figure out, well, what exactly kind of a micro grid are they going to be, you know, building? And is that going to be compatible with my grid, you know, Dominion's grid? And I think that there's a lot of opportunities for the data centers to fund a lot of new capabilities, you know, to add things like inertia or frequency regulation and be able to compete with markets. And it's kind of an interesting problem because these new data centers are at times consumers of buying power. That's traditional data centers are powered by utilities. And they're not really powering as micro grids today. And so that's kind of a new, a new twist. So I'm sure that that has the figure into like Dominion's planning cycles, you know, because it's the data centers that are building out the actual, the power, you know, the power generation. Yeah, that's a really interesting market dynamic. Like are there different ways that the, you know, those micro grids or the power generation facilities they're building? By the way, is that they can provide value and with things like inertia as you mentioned, that'll be super interesting to see how that unfolds because there's also more value and potential revenue that the data center companies can, can tap into if that is sort of figured out as a broader market or others even. Yeah, I do a market study on micro grids. Unfortunately, with the data centers, they're no longer micro. They have to be called mega micro grids or hyper grids or hyper scale grids. coming up with some news. names just what we're going to call these because they're certainly not micro. Okay, let me move on to my next question. So, you know, we were talking now a little bit on the demand side, but let's talk a little bit more on the supply side. And I mentioned that I'm a light commissioner in a small town and I want to point out that customers always complain in the prices too high. And I suppose they can blame data centers to impart for some of this maybe. But, but during outages, getting that power back on is so much more important than the complaints about price. They don't evaporate necessarily, but they take a back seat. And so utilities are very highly focused on improving reliability. You know, in North American grid, it's kind of an aging fragmented grid. The European grid is definitely a more modern and reliable grid. So, anyway, new power generation is also being dominated by non-dispatchable, you know, solar wind and batteries. All right, we're losing the inertia from coal plants and even some gas turbines, although there is a resurgence of natural gas. But, but how is it that we can manage these changing loads? You know, I mean, you talked about the changing loads, but how is it that the data analytics can help plan for a changing generation side of that landscape? Yeah, I, another good question. I think, I think that, that, you know, that shift in supply that you're talking about is, you know, that's, we see that basically across all of the markets that are our power generation customers operate in. But I guess I would say at the same time, we still are relying heavily on what's an aging fleet of gas coal, nuclear and hydro as well for that, you know, reliability and flexibility. Aside, aside from the price aspect, I guess, that you're getting complaints on, I think the, that shift has a couple, a couple big consequences. So, the first major one, but the conventional fleet I'm referring to is kind of being asked to do things that maybe in many cases wasn't originally designed for, right? So, more cycling, faster ramps, more partial load operations. And of course, that increases, increases where, increases risk of forced outages, and maintenance costs, et cetera. And so, the second thing is, you know, more variable renewables on the system, as you mentioned, and on the point of system inertia, I think things like automatic generation control, frequency response, and even like ramping capability become more important than they were in kind of a world of steady, a big steady coal unit. And so, I think those are the two key pieces where we're in, you know, start to turn its role in managing that transition. So, of course, at the asset level, you know, good monitoring and diagnostics, that's really my background can kind of let you understand how that extra cycling, or that extra ramping is affecting turbine life, boiler health, condenser performance, and so on. And so, you can move to that to more condition-based maintenance and smarter outage management. You're a little bit less blind, but hopefully that ultimately reduces your total cost of generation, right? Reducing maintenance costs avoiding failures. And, you know, perhaps that can flow through to some of those folks in your town as well. But that's really what I focused on back on to your power generation, applying advanced analytics across the fleet of generation to pick up degradation trends early, and avoid unplanned losses. Now, that was mostly base load power, but the, you know, the more reliable your base load is, and this is market dependent, but, you know, the less reliant you are on peaking and loss exposed, you are to the effects of variable renewables and reliability risk associated with those more varying modes of operation that are taxing the conventional fleet. So, I think that's one piece. I mentioned automatic generator control, so that's more at the kind of system level, but we are also working with utilities. We're using analytics to look at how their units actually perform in AGC. So, you know, are they following dispatch signals cleanly when the group is moving around, or are there, you know, are there any lags or dead bands that, you know, leave the system operator to think more of the heavy lifting or maybe even leave money on the table for the generator in some cases, right? When they could have been generating more in response to an event. So, you can analyze the response of each plant to set point changes, grid fluctuations, and then you can identify, you know, which units are your workhorses for flexibility, and maybe the ones that, and more importantly, the ones that need tuning or further attention before you kind of get to a catastrophic event. But, maybe one other thing to add on pricing that I think you kind of hit on is that there's way more factors outside of just, you know, the cost of electricity that determine, you know, the actual value of a megawatt hour that's produced, right? You know, and you can likely build a solar wind fleet for fraction of the cost of other generation types, you know, maybe on a dollar per megawatt hour basis, but it doesn't tell the full picture because the production is only available when the stack doesn't really need it or requires, you know, maybe new transmission infrastructure or balancing resources. The full picture gives you the cost on the overall system, and I think that's why a mix is so important from a consumer point of view. Right, I think it's important to point out that, you know, the LMP price, the local marginal price, that's the real-time energy market that we see across most of the US. That price of power goes up tremendously every day between 5 and 9 pm, all right? And utilities bid in to the real-time market at whatever price they can do because they will always generate cheaper than gas, nuclear, or anything else. And as a result, with lots and more solar getting onto the grid, it definitely cuts the profitability during the sunny parts of the day for gas and nuclear and coal, all right? And it tends to drive them more into the peaking environment. So it would seem to me there's plenty of opportunities for, and I think utilities have figured out how to create the right kind of markets, capacity markets, transmission markets to help build out the infrastructure they need. Is that part of what Seek does? Do you look at the market structures or market pricing or the response to those market prices from the actual operation of the grid, you know, as you could see it in the data? Not significantly. I will say, I'll use an example of one customer that does use Seek in a similar capacity. So, per-cott is a user of Seek. It sort of fits in this type of decision making workflow, but really their market analysts, they use Seek to model how big a big commercial or industrial customer will curtail in different pricing scenarios. So they can forecast when prices spike and subsequently they can basically they can use historical data to predict price driven load shedding and understand how that affects the supply stack as well. So they're, instead of treating it as fixed day, they can look at it as like a dynamic problem. So I mean, that's one small piece of what you're getting at, but I think that's probably the one example I can go to of where we're kind of having an impact on that kind of the decision making around those types of markets, if that makes sense. Yeah, that does. So large utilities have a large number of different kinds of generation assets, you know, coal, nuclear gas, solar, wind, batteries, you know, all of this stuff is on the grid, even hydro. And different parts of the grid came up historically with different equipment, you know, GE might have put a system in and they use grid OS. Siemens might have used their grid scale X in Hitachi energy. And so, anyway, there's all of this different infrastructure out there. The question I have is how does Seek fit into this landscape of this diverse collection of grid automation suppliers and data? Yeah, I guess I think about this landscape kind of in Lengers. So, you know, some of those suppliers, you mentioned GE Siemens or you know, Hitachi or Schneider and others or maybe even some of the software vendors that play in that similar space but outside of the hardware. A lot of these, I guess a lot of the platforms or tools that they provide are kind of the primarily providing systems of record and control. So, you know, you have you mentioned grid automation, you have you have state of systems that can go to extend more to like enterprise asset management systems, you need to see MMS systems like for maintenance or outage management. So, I think that those are all excellent at running the grid, storing data, executing work and to your point they're basically accepted standards in the industry like utilities need all of that to operate. I think where Seek fits in and deliberately so sits kind of one layer up from that you can almost think of Seek as an analytics fabric that helps utilities get the most return of their investment in those data platforms right. So specifically you know purpose built for time series data and the subject matter expert so it's the people using Seek know the assets they know you know their operations but but Seek's not a data store right we're connecting directly to those historians connecting to APM systems even maybe lab records whatever you have asset management systems so we're leaving data in the original systems of records that's important from a cyber and in critical infrastructure standpoint as well as productivity right like utility doesn't want to invest in all those systems and start copying operational data into another cloud repository repository or something for analysis so to get more to the the value side of things first they immediately shows up in reducing those experts kind of the cost associated with curiosity for those engineers or maybe planners you know like they're not opening ten tools exporting CSVs moving stuff around like they have a single environment where they're pulling in time series data from historians and skidda that's being joined automatically with you know condition indicators from APM or maybe work management history even market data and you know we talked about a couple examples of that and so then they can use all that together to kind of build calculations analytics and dashboards asset health kind of workflows and scale that across the fleet so you know and then it kind of works together as a whole technology stack to kind of achieve those higher outcomes and reduced out of trade so we're playing at efficiency and you're really enabling more productive engineering workflows okay well then let me ask a little bit of another couple of questions about this you know these data historians and the data you know there you get data from multiple historians it's time stamped it may be bad it may be out of range it might you know I might need to have some contextualization what kind of data foundation do utilities need in in order to deal with the quality context and and governance to really succeed with analytics and AI applications yeah so maybe I'll start on the quality point I think in terms of time series data that's where we spend a lot of time and with times series data that quality you know you kind of allude to it but that's not a one and done it's kind of like a permanent dynamic issue that needs to be managed so you know sensors drift they drop out they get noisy whatever and and so you know you feed those bad measurements into models or monitoring you're going to get bad decisions out so I think the utilities using seek are treating quality and sensor health as an analytics problem in its own right so you know simple statistics pattern recognition to flag rational readings or you know stock signals noisy noisy signals and so on and then you can you you can trigger maintenance so you can enable condition basements around your instrumentation itself or or compensate for that right so the same techniques that can be used to identify poor signal quality or data quality can be used to smooth noise remove outliers and build things like soft sensors from correlated parameters so you know if something goes bad you're handling it right but you're handling it upstream before it flows into your decision making layer you also mentioned context so I think that's that that becomes a a major barrier to scaling as well as productivity when leveraging time series data so you know historians full of thousands of tag names but that doesn't become that's not super useful if you don't know you know which tags belong to which transformer or turbine whatever and so you know something as simple as linking time series data to asset metadata you make this history location made plate stuff that's critical for scaling you know moving from that one off analytic to a whole fleet but also if you want to do anything meaningful with AI right you need to know about your history of operation with that asset in context of what's happening right now and maybe in practice a lot of utilities do that type of thing with an asset framework inside of a historian others maybe haven't had the resource to invest in that so in either case you know seek kind of a lazy to leverage the zz existing hierarchies or build your own kind of asset context models for specific use case or if you want to you know create one model to rule them all for your whole fleet but that's really an enabler for scale yeah I see you know Mike Mike's is a control system engineer the most important document on the project was what's called the tag list that was the scope of the thing more tags is more expensive and and obviously once that tag list is there it's that's all the data you're collecting you know hopefully you've got everything you need one last question then is you know have this data from skater and substation it lands into the historians and other platforms what are the major challenges in seeing that making that data usable for the data teams you know is it filtering out on you know bad data is it synchronizing the time stamps is it putting it into context or combining it with other information so I think I think it's it's it's it's really an all of the above answer to that you can kind of I mean you've kind of broken it out into that sort of maybe fragmentation or a translation piece which is where the subject matter experts really matter around context and then scaling I think the time stamp alignment is a problem of the past so at least I'll say to to our horn a little bit that happens completely automatically across any data source within seek so you may immediately kind of get rid of that that kind of issue with wrangling in a lining different data sources and different historians like really what you want is your experts to be working across all your data sources and working on actual business problems as opposed to kind of data engineering problems but then you know we already talked about quality and context I think that's the that's the next piece that's kind of where the subject matter expertise matters as I'm as I mentioned right you know a raw historian tag or an event log isn't a is not analytic so you kind of need someone to understand your assets to turn that into something meaningful that decision makers can use right what's normal versus abnormal behavior for this family of breakers how to cycling affect you know these turbines so that that's super important and then I think you know the the real transformative value comes from when you can kind of when you can take some kind of approach whether it's analytics or modeling or simple calculations and health scores even things like that and scale it across assets so that means scaling it across different data sources that means pushing out to different data sources like a CMMS system so you can kind of get to those end to end maintenance workflows and I think that's where that's where seek really comes in I kind of touched on it a little bit with the asset hierarchies and easy tools for scaling but that is something I think that is that is missing that is sort of the next layer of how you can get like you get to kind of answering these questions across your entire fleet right all right great I think we have to call it to close I would like to thank Daniel and seek you know for giving us some really unique insights on how all this data gets from sensors into the decisions that analytics makes and to improve the monitoring control management optimization of electric grids so Daniel thank you very much thank you and I encourage our listeners to go to the seek website and also the ARC website to get even more information on these very topics thank you [BLANK_AUDIO]