Why state of charge is key to BESS success with Blake Rector (Principal, Markets & Optimization @ Powin)

Hello, everybody. Welcome back to Transmission. It's me, Quentin. And this week, we've got Blake Richter, director of markets and optimization at Powin. Now Powin are a American manufacturer of battery energy storage systems, and they've got quite some scale. So seventeen gigawatt hours of assets are online or under construction. And Blake has been with the company pretty much since the beginning.

Now in this conversation, we go really deep down the rabbit hole of a hot topic in our space, which is state of charge and especially how we measure it and how we maybe make some errors.

And I think if you're an asset owner or you're related to batteries, this is a really important conversation even though it goes into some great technical detail. So I hope you like this conversation. I certainly enjoyed it. If you did, hit like, subscribe, and all those good buttons. It means that we can have bigger impacts and reach more people with our content. Let's jump in.

Hello, Blake. Welcome to the podcast. Thank you, Q. It's an honor to be here. An honor. Thank you for saying that.

This is gonna be a lot of fun. We're gonna talk about one of those hot potato topics in energy storage that we really need to go down the rabbit hole to figure out. So if you're listening to this, this is an episode all about State of Charge and how it impacts financial performance. Before we get started with that, can you just talk about your business, Powen? What what does the company do if our audience haven't heard about it, and what do you do there?

So Pawan is a utility scale, integrator.

So we deliver the turnkey, battery energy storage systems. We we work really in large scale projects, these days. Seventeen gigawatt hours deployed or under construction, over seven hundred employees globally.

And we have projects in ten different countries.

Within North America, we have projects mainly in CAISO, in non CAISO WEC, and ERCOT, but then also projects in, MISO, New York, ISO, ISO, New England, and PJM.

My title is principal of markets and optimization. I've been with the company since the very early days.

I'm a mathematician by training.

So sort of came on to do market operations and bidding optimization for our very first systems that Powen actually owned in the very beginning. Now we strictly sell them. Since then, my role has evolved and done lots of different things with the company, but, now primarily work with the data analytics team and the product management teams. And one of the things that I've been focused on for the last couple years is our state of charge estimation and then more recently, some research around the financial impacts of state of charge accuracy.

Okay. And we're gonna get into that. It's a very juicy topic. But seeing as you've been with Powern for so long, since the early days, and it's been on one hell of a journey, could you just give us a quick story of the company?

Sure. Sure. Powen, originally called Powen Energy, started as a parallel company to Powen Manufacturing.

Powen Manufacturing had been around for around thirty years, but then Pound Energy started first as an r and d company around, two thousand ten, maybe two thousand twelve, and then, became a utility scale best provider with our first project in two thousand sixteen.

This was for those that remember that, Aliso Canyon solicitation through Southern California Edison. There were three projects that were awarded and then commissioned in about under six months to deal with the concerns over the the gas leak at at the Aliso Canyon storage facility in the LA basin.

We were doing it all with a pretty small team back then.

I feel like Pound sort of always been a company, especially in the early days, that that punched above its weight, so to speak, that we've just done things with a lot of sort of grit and tenacity and heart. So very early days, company was around fifteen people when I started.

By two thousand, I think we had grown to little over eighty. And then today, just sort of several different chapters in terms of our of our growth and maturity as a company. But, yeah, today to be over seven hundred people, it just means that a lot of battles have had to have been won along the way. And it's been really impressive to see everyone work together to get to where we are today.

And so power in I mean, the numbers are quite staggering. Right? So this is a this is a lot of deployment of battery energy storage around the world.

Where are the batteries manufactured?

So now we have a fully sophisticated and dedicated integrated supply chain team.

And I think, the name of the game these days is is flexibility.

But we do have our batteries being manufactured all over the world, some in North America, some in China, other in other parts of Southeast Asia. I'm sure the supply chain team would love to come on and and give you a full rundown of of everything there, but it really is a a globally diverse manufacturing paradigm at this point.

So so we've gotta talk about your platform and what you're involved in in the firm. So bidding strategies and optimization.

Can you just talk about StackOS? What what is StackOS?

And then we'll talk about, market access and all of the other things around state of charge that we need to be thinking about.

So one of the one of the key features of Powen, and it has been sort of since since the beginning, is that we write all of the software at every layer of the system. So we we write our own BMS.

We write what we call the EMS, which essentially manages all the power distribution at the site level and including the the TMS or the thermal management system all the way up to the cloud UI, that customers use to see their systems. So this is all written by Palant, and it's all considered StackOS.

We do not provide market optimization as a service. That's actually sort of a unique and maybe even in some ways interesting point about us, and that frees us up to help our customers with their optimizations.

We do have, though, that we don't talk about a lot, but we we do have an in house, behind the mirror optimization program. I call it a a direct dispatch optimization. And so this runs in the cloud and is capable of dispatching, batteries directly through first decisions that get made in the cloud and then get passed down and executed in real time with additional logic, on-site. So we sort of consider everything there to be to be StackOS and consider that to be one of the main selling points of Powan is that all of the software is cohesive.

So I didn't realize that Powayn didn't do optimization services. I I knew you did behind the meter, and I thought you also did in front of the meter. Well, that puts us in a a very interesting position now because you, of course, have to help customers in choosing optimizers and working with optimizers and assumptions around their asset in the market and bidding strategies. But you don't actually take the keys to the asset, which gives you a very unique view on things.

So we're gonna talk all about state of charge in a minute. But before we do, let's just set the scene. So asset owner comes to you and says, look. I need to think about my bidding strategy in these markets.

I need to be putting something in my business case and perhaps choosing an optimizer.

What challenges do I have as an asset owner when I'm thinking about that, and how how do you help at Powen? And then if you could if you could link it to the state of charge thing, and then we're gonna go deep, deep, deep on state of charge because I know that's almost your mastermind subject.

Typically, the customers already have some sort of business case in mind when they come to us. It is something that we are doing now that we've invested more effort into our state of charge estimation is trying to explain how much additional revenue might be expected or maybe how much revenue might you might be expected to lose if you have state of charge estimation errors. So how much additional revenue for a more accurate state of charge can you expect? That's some modeling that that we are doing now to try to help get the word out. But, typically, actually, the customer already has the business case in mind when they traditionally, at least, when they've come to Palant looking for a battery to buy.

And seeing as we're gonna talk about solar charge so much, can you please define state of charge?

The this is this is a pretty interesting question because, you know, batteries are used in in industries around the world. Are all sorts of different industries Sell individual cell experts might have a different definition for what state of charge is than maybe someone in EV or in stationary storage. For us, we like to be as explicit as possible in estimating the energy available for discharge.

So when we put a energy, unit on our state of charge estimate, that should tell you if it's an estimate for discharge, exactly how much energy you can expect to discharge until you hit the bottom.

Similarly, how much energy could you expect to charge if it's a state of headroom estimate? If you're talking about the percentage, state of charge as a percentage, generally, my advice for customers is to try to use the energy estimates if they can in their optimization instead of a single percentage number. But the way we've defined the percentage now is the energy available for discharge as a percentage of the degraded energy capacity. So whatever the gas in the tank is divided by the gas that you're expected to be able to use in the tank divided by the size of the tank. That would be the percentage definition.

Things get pretty obscure, though, sure, when you start working in percentages. And that's why, again, why I would recommend, if possible, people use in their optimizations energy available for discharge or energy available for charge instead of a percentage.

Well, can you talk about that then? Why is a percentage an issue?

The analogy of the gas tank and the amount of gas in it, I think, is a good one in in in helping paint the picture. But based on conditions of the system itself, in particular, temperatures of the cells, maybe temperature imbalance, if there is any, or voltage imbalance within the cells, that means that the size of the tank can actually shrink or expand, based on sort of the the system state. And so when we give an energy number, that is our best estimate of the energy actually available for discharge. A percentage might be obscuring the fact that that size of the tank might be changing depending on the current conditions of the system.

So estimating state of charge has been a problem as long as time. I I remember one of my first jobs when I was at university was to sell cell phones. And this is back in the good days when it was just, you know, the dumb phones and it was Nokia's. I remember the Nokia n ninety five, if anyone anyone remembers that.

That blew the market away. The the reason why I bring this up is even back then, everybody knew that once you'd had a phone for six months, by the time you got down to kind of twenty percent of state of charge on your on your screen, you knew that it was gonna drain down to zero in about ten minutes. And this was this is, you know, some of the best manufacturers in the world. And, of course, there were different battery chemistries and and whatnot.

But this is a problem as old as time. I remember when the iPhone first came out, I'm pretty sure it didn't even have a percentage estimate. They just put that little green battery at the top, and I think it came out in, if I remember correctly, the second or third generation. It was an Apple feature that there was a percentage, demonstrating that it really is a dynamic problem and quite a difficult engineering problem to solve, to to solve this quite simple question, how much battery do I have in this thing?

Totally. Yeah. So let's start with what does good look like? What does excellent state of charge measurement look like?

And then how much of a problem is this for our industry, and what's it ending up costing us?

The gas tank analogy, I think, comes in comes into play in explaining this in that the amount of gas in your tank is really the amp hours that are stored within the cells. So this is to say the actual number of electrons that are available for discharge within the cells. So that's the actual amount of gas. How far you can drive on that gas is another estimation in itself. And so if state of charge estimation were perfect for utility scale operations, what we would have would be energy estimate that reflected exact the exact amount of energy that could be charged or discharged at you can think of it maybe as a set of curves at different power settings.

It gets to be pretty complicated when when you think of all the all the different options there are as far as how you can drive the car.

In terms of accuracy, I think that state of charge estimates that come in at three percent error or less are considered quite good. Ten percent maybe sort of an industry standard.

We've heard lots of stories about state of charge errors at twenty percent being not uncommon. So it is certainly something one of those things. This is sort of maybe the name of the game with with large battery systems. They're more complicated than you would think, but the state of charge estimate is one of those things that particular because people, you know, have so much experience with just that single estimate for their phone, they think that that translates to the whole battery system.

But why doesn't it then? Why why is it not just as simple as a a phone, which, despite my previous comments, I've got pretty good at estimating percentages. Alright. So why is it why is it not just a a simple phone?

So this has to do with the fact that these battery energy storage systems are really not single batteries.

They're collections of tens, hundreds of thousands, or maybe even multiple million batteries, individual cells at a site. And it's hard or challenging, I guess I should say, to do state of charge estimation for for one individual cell. That's part of the problem, especially for LFP with the ion iron phosphate cells, which we could talk about why. So there's the individual cell estimation problem, and then there's the aggregation problem. And when you aggregate these cells in series into a stack of batteries, you're only able to charge until your upper cell hits its upper limit or discharge until your lower cell hits its lower limit. So there's some voltage balance issues that could be in play there. And then aggregating those stacks, which could be then in parallel behind single centralized inverters, each having its own temperature characteristics and performance characteristics.

Those central inverters then aggregating up to the site. Yeah. It becomes many pieces to to manage that's a lot different than just, say, taking the straight sum of all the stored energies of the individual cells.

And you talked about LFP there. So this feels like the right time to ask you about different chemistries.

And in the earlier days, you know, three or four years ago, NMC really dominated and still does in some high performance applications and some electric vehicles.

But certainly with, Chinese manufacturing, the needle has moved over to LFP, and LFP certainly dominates markets in the stationary storage world now for a ton of reasons. One of them is, is cost. Could you just talk about the difference between those two chemistries and measuring state of charge for them? You mentioned that LFP was more difficult to measure state of charge. And am I right to assume your reference, sir, is it's more difficult than NMC.

And in that situation, you know, why is that?

As a cell charges or discharges, the voltage we we can think it once a cell is at rest, generally speaking, the more amp hours it has stored inside of it, the higher the voltage will be at the terminals.

So the more charge in simple terms, the more charge in the battery, the higher the voltage you'd measure at the connection points.

Yeah. Yeah. At the terminals. Yeah. Exactly. And so if you think of that there's there's a voltage on the bottom, that's the, say, the lower limit for the cell, and there's a voltage at the top, and we maybe say that's the upper limit. If the voltage curve and what we call it were straight between the upper and lower limits, then you could essentially look at the voltage itself and then just translate that voltage over to the state of charge for that cell.

So this is an important point. Right? And it's it's not obvious. So there's no point on a battery if you think about a double a battery from Duracell, there's no point on that battery that that you measure the state of charge.

So what you have to do is measure other attributes and then work backwards. So you're measuring the voltage across the terminals, and then you're inferring you're you're translating with a a bit like an exchange rate between currencies. You're translating that voltage into some other thing. And in this case, we're considering it to be how much energy is in the back battery or state of charge.

It's that translation that becomes difficult because, as as you say, there's a moving exchange rate. So when the battery's got lots of power in it, the exchange rate is one way. You know, you might it it might say one thing. And then as it as it loses power or loses charge, then the exchange rate moves, and that that makes it very difficult to estimate in between.

Right?

That's just part of it that just the the exchange rate itself can be difficult to pin down, especially if you're applying current. But the really tricky part is that, unfortunately, there's not a straight line between the lower and the upper voltages in terms of that voltage to state of charge curve. What happens, and it's even more so for LFP, is that say so you're starting from empty and charge up initially, by the time you get to about twenty percent, all of a sudden the voltage becomes very flat between say twenty and maybe ninety percent of state of charge. So flat that it might be within the error of your voltage sensor depending on how accurate your voltage sensors were.

And then at the very top side, the curve takes another steep turn up to the top. So all of a sudden, if you are using this ex exchange rate or this voltage lookup program for estimating your state of charge, all of a sudden it becomes very uncertain or inaccurate if you're in the middle of that state of charge. And if you're using your system in a way such as many frequency regulation applications, have systems doing where you stay in the middle for a long time, some errors can accumulate there.

And you could be creeping one way or the other without knowing it.

Absolutely. Yeah. Yeah. So if you stay in the middle for an extended period of time, errors can accumulate to those particularly high numbers we saw or maybe even more. LFP has a very flat curve in the middle like I described. NMC does not have as flat of a voltage curve. Some battery old timers told me that that was actually one of the attractive reasons for one of the attractive reasons for using NMC in the first place was that it was much easier to estimate the state of charge because voltage was such a better signal for understanding how how many amp hours were stored within the cell.

So it's in the middle bit that the problems occur. And if you're if you're using a a lookup table or a program or some sort of exchange rate, however you wanna call it, you really don't want flat bits, do you? You want a nice straight line ideally going up diagonally so you can say a change in voltage equals a change in state of charge. So that's fascinating. I wonder why you you did mention with LFP that the the flat bit happens somewhere around twenty percent and somewhere beyond ninety percent. So I wonder why that is. That's that's not symmetrical around fifty percent.

Yeah. There is another step up, small step that we see in the middle. I think it's around seventy.

All this is really the realm of of chemical engineers. It is a little bit outside of my expertise, to explain exactly what's happening within the chemical processes there, but but that's what it is, is is chemistry.

So there's two ways that this can catch you out. Right? That when you need the power, it's not there, or when you wanna charge, you're actually coming up to your limit and you didn't realize. And so my intuition would be that, really, the biggest impact for you in making money in the markets is if you call your battery or you commit your battery and it can't provide the power and a high price and you have to make that up elsewhere, for example, you get penalties or something.

And so my my gut feeling is that probably the thing that we're most scared about is overestimating state of charge, when you think you've got it but you haven't, which sort of checks out. So how big of a problem is this? Now we've talked about it and and how it works and where the problems are, how can asset owners do things smarter to make sure that they don't get caught out with this?

So, yeah, we we did just recently do a bit of a research project to try to quantify the financial impacts. In terms of what asset owners can do, I think the first step would be, if possible, try to understand how the state of charge algorithm works for the asset that that you have. And in in most cases, there will be a recalibration, so to speak, that can happen if you charge the system all the way to full or discharge it all the way to empty because the state of charge estimate, becomes so much more accurate when you're at the extremes.

Regarding cell balancing, this since the balance of the cells can have a big impact on the amount of energy that you can actually get out of the system to follow whatever the maintenance guidelines that your hardware provider has has given, or if you have an LTSA with them, like with Palant, manage that for you. And maintaining maintaining the equipment in general including the thermal management systems becomes very important for keeping the safe charge as accurate as possible.

Yeah. There's so many variables, aren't there? Because to start with, you have a cell that every single day is changing its electrochemical properties, its age, calendaring degradation or aging over time, which is changing that voltage curve and reducing the available capacity. So you have just a a ticking timer clock that's going, which is changing the characteristics.

And then you have your operational parameters, so running it in different duty cycles at different temperatures with different operating parameters. And then you have the measurement error to manage, which is in itself a compounding problem and where as you as you talk about, you know, what is your reference? The reference the the error in the reference is constantly moving around. So a really tricky problem to solve.

And one of the things that you said that really blew me away was that ten percent is considered standard, I think you said. Right? Three percent is good. Ten percent is standard.

That feels really, really high considering the the capital expenditure that goes into these projects. That is just just such a high number.

Totally. And and I think what you see is a lot of operators, if they don't trust the state of charge estimate, then they essentially reduce. They they put some sort of, like, haircut or some lower bound that they leave around the bottom of their energy capacity. Yeah. Because they they don't trust the estimate, and so they wanna make sure they don't run out. That essentially reduces their effective operating capacity, and we came up with some estimate numbers for that in our research as well.

So can you talk about your research in more detail then? Let's let's jump into that.

Yeah. Definitely. The question that we were trying to answer was for each one percent increase in SOC estimation error, what is your expected reduction in market revenues? So that's what we wanted to do. We wanted to try to translate the SOC accuracy to what it means for actually making money in markets.

We teamed up with, the folks from Tiara Climate. I believe, Emma Connette has been on your show before.

Yeah. Friend of the podcast. We've done, we released a white paper with them. Terrific team.

Yeah. They were great great to work with. So they work on the carbon policy side of things, and they also have market operations group where they have bidding optimization capabilities.

And, really, what worked well for this project, a deep understanding of market rules in ERCOT and in particular, how it is as you're going throughout your day if you don't have the energy that you thought you did in your system, how that translates into either AS violation costs or base point deviation charges.

Alright. So how are we gonna answer this question? For each one percent in, SOC error, what's the expected reduction in revenues? Most people, when they do revenue simulations, they essentially assume a hundred percent accurate SOC estimation.

So they say my battery's telling me that I have a hundred megawatt hours stored right now. I'm gonna discharge ten megawatt hours. And that means that that state of charge goes down ten. This essentially means you have a hundred percent accurate state of charge estimation.

What we did is we came up with an and we actually released a white paper that has has all these details, but we came up with a way to model the error associated with SoC estimation. And the things that we wanted, the characteristics that we wanted to replicate there were that as you get towards the edges, it becomes more accurate. The error can accumulate if you're in the middle range of the SOC. And then we wanted to have some knobs so we could turn it to, say, bias overestimate, bias underestimate, and then add some random noise in there.

We set up this model with within their market simulation framework. And so their optimization engine then only had access to this reported SOC number, the number that was the true SOC plus the error. And their optimization engine had to make its decisions based on what it thought the state of charge was instead of what it actually was. So we compared all of those runs with various amounts of SoC error to the perfect SoC runs.

So the runs where we had a perfectly accurate state of charge. After we went back and then characterized the error for each run, we looked at how much revenue reduction there was. And we found that for arbitrage so when we limited the market participation to be an arbitrage only, that the reduction in revenue was about point eight percent for each one percent of SOC error. So that would mean if you had a ten percent SOC error that you were could expect to make eight percent less revenue from arbitrage just based on how the individual days play out, in those year long simulations.

Interesting. I mean, it's comforting to know that the money you lose is less than the percentage.

At least the exchange rate is somewhat in your favor.

Yeah. Yeah. And if you wanna get more comfortable with ancillary service heavy bidding strategies, you lose point two percent for each one percent SOC error. So it's it's much less. The reduction revenue is almost twice as bad for overestimating as it is for underestimating. But going back even to the zero point eight percent number, I mean, that eight percent reduction in your arbitrage revenue, I mean, that could be a significant amount of of money depending on, you know, what year and how big your system was.

Absolutely. Yeah. It's still worth sorting out one way or another.

Right.

So I know that there's some friends in the podcast, KP at Acure, and Stefan and her team at TWAIS and some other folks are doing some really advanced stuff in battery cell analytics to try and solve this problem through modeling and software, and by all accounts have had some pretty remarkable results.

So what else is happening in this space to try and close this measurement error gap, which is evidently leading to loss of revenue and loss of, therefore, impacts of of battery assets in the system. Now what can asset owners do, and what's the the frontier of fixing this thing?

Just one more note on the results there. We did the revenue assessment. We also found that essentially for each one percent error, it's as if you have a one point two percent smaller battery. We call that the effective usable energy capacity of the system.

So independent from revenue, because that was just an ERCOT, you can look at sort of the amount of the battery that you can expect to have access to. And so for each one percent SOC estimation error, decrease that effective usable capacity by one point two percent. In terms of what people are doing. I mean, Palen's angle here, since we have, it's always been at the heart of what we do the data.

So we, we have access to the data all the way to the cell level. That's been part of our design. That's part of our software and all of the historical data that we've had over all the years of our projects.

So we are releasing soon our new state of charge estimation algorithm, and essentially it leverages this vast amount of historical data. We also are are happy to work with Acure or companies like them to help customers get as comfortable as they want with their systems. But for us, it it is an ongoing process, the state of charge estimation improvement and in conjunction with the state of health estimation improvements. They're they're closely interlinked.

So, Blake, now to our last two questions. You you plugged your new system a little bit there, but here's your opportunity to sell power in or anything special that's coming up or a world exclusive. You can announce it here. And then I'm gonna ask you about your contrarian view. But firstly, is there anything you'd like to plug?

Sure. Well, you know, I mentioned it earlier, but I've been impressed with Powen with the individual people I get to work with here. And the company has tenacity and heart in a way that is different from any other company I've I've been with. And so I I guess I wanted to plug the company overall. But, yes, we will be releasing soon our new state of charge estimation algorithm along with some other announcements surrounding it. So that and the white paper, which, was released here around the end of March, that you can find on the company website, the economic value of SOC accuracy.

Okay. So we look forward to reading that one. And now to my favorite question, Blake.

What is your contrarian view? So what's the thing that you believe that not a lot of other people believe?

I think that that it's best to look at people in industry in your industry as collaborators instead of competitors whenever possible. So it's not all that earth shaking, but it definitely makes things a a bit more fun, yeah, when we when we think about how it is that we can work together instead of beat each other up.

What a lovely end. Blake, thank you very much for joining us on the podcast. We went very deep there into State of Charge, and I think it was absolutely worth it. I would love to get someone else on from Power and to talk about the story.

Seventeen gigawatt hours installed or in or in construction. This is a company of real scale, and I'd love to talk to your guys about the supply chain so we can make that happen. And if you're listening to this and you wanna read up more about Powering or some of the things that we talked about in this conversation, you can find links in the show notes. And with that, Blake, I say thank you and speak soon.

Thank you very much. We'll talk to you soon.