Eric Parker, Hydrogen and Fuel Cell Technologies Office: Hello everyone, and welcome to March's H2IQ hour, part of our monthly educational webinar series that highlights research and development activities funded by the U.S. Department of Energy's Hydrogen and Fuel Cell Technologies Office, or HFTO, within the Office of Energy Efficiency and Renewable Energy. My name's Eric Parker, and I'm the HFTO webinar lead. As always, we'll be announcing more of these topics on a rolling basis. So, please keep an eye on your email for next month's.

This WebEx call is being recorded, and we will be posting the full recording and the slides on the DOE website. All attendees will be on mute throughout the webinar, so please submit your questions via the Q&A box you should see in the bottom right of your WebEx panel. We will cover those questions during the Q&A portion at the end of today's presentation. Without further ado, though, I'd like to turn it over to our DOE host, Neha Rustagi, who will tee-up today's topic and speaker. Thanks, Neha.

Neha Rustagi, Hydrogen and Fuel Cell Technologies Office: Thank you, Eric, and thank you everyone for joining. So, today, we have Michael Penev, from the National Renewable Energy Lab, providing an overview of work they've been leading in long duration energy storage. So, characterizing the role of many different technologies, including hydrogen and fuel cells. Then, also, a model that they've developed to allow for any user to do their own analysis of this kind. I really want to acknowledge my colleague, Sam Baldwin, from the DOE Strategic Analysis Team. He's been leading this effort out of DOE. On that, I'm going to kick it over to Michael.

Michael Penev, National Renewable Energy Laboratory: Thank you, Neha. Can you guys hear me okay?

Neha: Yep, you're good.

Michael: Okay. I would also like to acknowledge that this has been an extensive team effort. We've had input from the team members listed here, and ranging from detailed grid modeling to extensive research on costs of various technologies. So, for our talk today, I'm going to go over our methodology of our analysis, some preliminary results, and then I'm going to introduce a model that we use for this. It's called the StoreFAST Model. I'm going to describe it as well as provide a quick demo of how the model operates.

So, long duration energy storage within our efforts to analyze nearly 100 percent renewable grids, we're finding that short duration storage is not the cost optimal way of getting to nearly 100 percent renewable grid. As grids exceed approximately 80 percent renewables, the variability on the grids from those resources from the point of the supply as well as from demand induces the need for long duration energy storage. So, when we talk about long duration energy storage, we're talking about technologies that provide multiple days of storage, definitely above 12 hours, but on the order of 5 days if where we've been focusing for this analysis. What's new about this analysis is that we're not only considering today's technology costs, but we're also considering how the technology costs may decrease over time as we have learning curves, as well as volume manufacturing for each technology type.

For this analysis, we're using levelized costs of energy as the key metric. That is defined as the total cost for operating the systems divided by the total kilowatt hours provided by the system. For the analysis, we're using the StoreFAST model and we're basing the analysis at roughly 100 megawatts of storage capacity for each of the systems. We're analyzing systems between 12 hours of storage duration rating up to 7 days of storage duration rating. One thing I should point out is when we talk about duration rating, we mean that if the system were to be completely fully, how long does it take to completely discharge to a minimal state of charge while providing full rated power to the grid.

Our storage systems are broken out into three distinct components. We consider separately the charged components of the system. So, for example, in a hydrogen system, that would entail the rectifier to take AC power and to produce DC power for an electrolyzer, the electrolyzer itself, as well as a subsequent compressor. All these capital costs and other items are characterized based off of the charging of the system. One thing to point out is when we talk about charging costs, those are based off of input power. So, when we say dollars per kilowatt of charging, that means based off of the rate of power of charging. When we talk about electrolyzers, typically the electrolyzer can be a very different capacity rating than a fuel cell. So, we've disaggregated charging power from discharging power.

The second component that we model is the storage component. This component is strictly the storage. So, for example, if we are talking about hydrogen storage, that would be the salt cavern itself. It would be able to receive hydrogen, store it, and then return the hydrogen to a power generation equipment. If there's a discharging component, again, for hydrogen systems, this would entail fuel cell or combustion turbine. If there are any power electronics associated with it, that will be part of the system.

In addition to energy storage systems, we also model flexible power generators. When the grids need additional power, it is agnostic whether that power comes from stored energy from prior overgeneration or if it comes from a flexible power generator. So, in terms of a green source of flexible power, we can, for example, consider biologically the right molecules, such as ethanol, being run through a combined cycle plant to produce power flexibly in a renewable fashion. This, by the way, is a technology that has been demonstrated to about nearly 100 megawatts of capacity.

In order to understand how the systems would be operated in a highly-renewable grid, we took analysis results from detailed PLEXOS modeling. This is a grid model that considers the optimal way of dispatching various power generation equipment on the grid. When we go to 85 percent renewables, we're seeing that the grid would be optimally operated having energy storage systems that exceed at least five days' worth of storage. On the upper right corner is an example, a graphic of how the grid would operate such a system. We see that along the X axis, we have time of the year. On the Y axis, we have hour of the day. A long duration energy storage systems would be operating at the diurnal fashion. So, during the day, when solar power is widely available, the system will be charging. Then it will be discharging its capacity during the evening hours.

During the spring months, we're seeing an excess of renewables and not an excess of demand for electricity. So, during this timeframe, on the net, the system would be increasing in state of charge until about mid-summer, when we see temperatures increase, people likely having more air conditioning loads, and, at the same time, the renewables, spring runoff from hydro reduces and we're seeing that the net state of charge decreases over time. So, this is the typical operation profile for long duration energy storage.

The grid model simulated different round-trip efficiency systems and characterized for us how a 40-percent efficient system would operate, a 60-percent round-trip efficiency system would operate, all the way up to 80 percent. What is important for us is the capacity factor for charging and discharging. On the lower left, you're going to see charging and discharging capacity factors for different round trip efficiencies. Within our modeling framework, that's the take away that we use to model the tech economics of these systems. For flexible power generators, we're queuing from the variable operating expenses to produce an additional kilowatt hour from the various technologies that we've modeled as flexible power generators.

I'm going to go over some results of our modeling. First, I'm going to look at current costs. So, if we have the technologies, not improved but rather have the current costs, how will they operate within that framework of different capacity factors and sizing.

So, on this slide, we can see an example of two technologies. On top, we have pumped hydro storage. On the bottom, we have heavy-duty fuel cells using salt caverns to store hydrogen. On the right, we have two graphs. Along the X axis is duration rating for different systems that we consider, and along the Y axis is the levelized cost of energy.

First thing I want to point out if the bottom wedge, which is the operating expense associated with charging the systems. One thing we've noticed that PHS, having a relatively high round-trip efficiency and purchasing electricity at two cents per kilowatt hour, that's the basis that we're using for utility scale renewable electricity. That operating expense adds up to a relatively shallow wedge. On the bottom, we see that having a low round-trip efficiency, a fuel cell system requires a substantial amount of more electricity in order to provide peak power generation.

Another item that is fairly different from those two systems is the storage component. This is the capital associated with storage. For pumped hydro, that is a substantial expense. If you want to have a higher duration rating system, you have to purchase more capital. You have to pay additional financing expenses and increasing maintenance expenses. That profile for hydrogen systems is fairly flat. The marginal cost of storing additional kilogram of hydrogen or additional kilowatt hour of electricity is fairly small. So, we see a low sensitivity to duration of storage.

Now, in subsequent slides, we're going to look at just a total levelized cost of energy for different technologies. On this slide, we're focusing at the 12-hour duration rating, and we see that within our modeling framework, we do get similar results to literature in terms of predicting the types of technologies that are going to be more competitive. In this case, we're seeing a baseline of natural gas-combined cycle on the bottom, and then the storage technologies such as pumped hydro, thermal energy storage, VRB's in here as well as compressed air energy storage systems. Those show as being competitive.

When we look further to the right, we're seeing that those technologies that are competitive at the shorter durations, they tend to have more expensive energy storage cost component and become less competitive at longer durations. So, for example, the VRB and pumped hydro become private or expensive at this range, and technologies such as natural gas-combined cycle with CCS as well as diabatic CAES and hydrogen technologies, they're less sensitive to duration of storage. At this level, they're the most competitive option.

Now, to look at learning curves, we assumed an additional capacity of 200 gigawatts being installed and how that might impact the cost of capital for both power and energy for each of these systems. I'm not going to belabor the details of all the detailed research that we have done on learning curves, but the next slides regard to look at slightly learned out technology set.

So, again, at 12 hours, we're seeing technologies such as PHS, pumped hydro, vanadium redox, adiabatic CAES, as well as hydrogen storage with PEM fuel cells being competitive. One thing we did take out in this slide is technologies that are carbon emitters, just because we're trying to conform to our highly renewable scenario. So, combined cycle plant with natural gas is not listed in this technology set.

When we look farther to the right, we'll see that learning has benefitted both electrolysis portion of hydrogen storage systems as well as the capital for fuel cells, and those technologies become substantially less expensive. We did not assume any learning of salt caverns. So, that cost component remains insensitive to duration of storage.

Within our analysis, we perform Monte Carlo simulations on components that either have variability or uncertainty. This is an example of the results for current technologies as well as future technologies. The way to read them is that each one of those is essentially a histogram. The leading edge on the left is the first percentile. The thickest portion of the violin plots is the most likely value. Then the tail-end is the 99th percentile of the Monte Carlo analysis runs.

A couple of observations from here is that there is a lot of overlap from technologies, meaning that, depending on local conditions such as cost of renewables, or cost of electricity, or cost of capital for PHS, for example, if different storage technologies are available, one – or the technology going to be preferable. So, we don't expect one technology would necessarily win out against all the other technologies. Another thing I want to point out is that when we consider hydrogen technologies, there is the opportunity for the electrolyzer not just to charge the storage system but also it can produce hydrogen which can be stored, staged, and sold as retail hydrogen for various uses; commercial or industrial uses. So, we have overlaid here in orange is cases where we sell hydrogen for as much as $3.00 per kilogram of hydrogen. If you're familiar with those type of costs, that is a fairly competitive price point for hydrogen.

Next, I'm going to demonstrate the StoreFAST model. I'm going to go through an overview as well as provide a quick demo. So, the StoreFAST model is intended to provide a consistent framework for utility scale power and it will analyze both energy storage systems as well as flexible power generation systems on side-by-side basis. So, it can provide either information for a single technology or multiple technologies and it will give you simultaneous results for all of them. The information will also be cast and interpreted by results from grid models. So, if you have technologies that have different roundtrip efficiency or if they have variable operating expenses for flexible power generators, those will calibrate the capacity factors that we would expect to see for these types of systems.

The model will also provide risk analysis for uncertainty or [audio skips – inaudible] – couple [audio skips – inaudible] this model _____ the strategy analysis. So, shorter duration systems _____ proficiency ancillary services to be _____ of their revenue stream. When we look at long duration energy storage, we anticipate that those would primarily derive their revenue from energy generation and ancillary services are not factored into that market. Additionally, the depth of the ancillary market is fairly shallow and we're evaluating the performance in terms of energy storage, which is a deeper market.

The model is currently calibrated with grids of models of 85 percent renewable scenario. Lastly, I want to caveat that currently the model of those amortization evading refurbishments. So, for example, if you want to look at components that are replaced on a ten-year schedule, those additional replacements, in the future, would have to be amortized and added to the maintenance expense, a variable maintenance expense for each of the components.

The StoreFAST model is based off another model called H2-FAST which is a very rigorous financial model. It takes its input items such as capital, maintenance, system usage, energy usage, energy prices, as well as financial parameters such as types of depreciation, types of capital or capital structure that you have available for financing. The model uses generally-accepted accounting principles as well – also known as GAP. They compute for each of the systems a forecast of input statements, cashflow statements, as well as balance sheets. In terms of outputs, we have levelized cost of energy as well as some key parameters such as internal rate of return, payback period, et cetera. All of these are available in graphical format in both individual as well as multi car the type of analysis.

So, going through the model, the first thing that she would have to specify is global parameters. Those parameters are duration of rating - does she want to benchmark each of the systems – cost of electricity, if it's going to be available to the systems, if natural gas is one of the feed stocks or ethanol, or if you're co-producing carbon dioxide, which would be sequestered at certain costs. Those would be specified on a global basis.  In terms of individual inputs for the systems, the first thing that the user specifies is capital costs. So, we would have charging, storage costs, as well as discharging. Each one of those can be specified on a flat basis, or if there is a variability of costs versus size, the user can provide that scaling factors for each of the technologies.

I'm going to reiterate one item is that charging costs are based off of capacity of charging, not capacity of discharging. Some of literature out there bases the entire cost of the system on power generation and not charging basis. So, if you're specifying charging component cost, they would need to be on a power input rating basis.

The next thing that the user has to specify is maintenance. The model accounts for fixed and variable costs for charging, storage, and discharging.

Feedstock costs are specified on per unit output basis. So, for example, if we look at electricity required for charging, that would be how much electricity is required to charge the system to produce a kilowatt hour of power. For example, if we have a system for hydrogen energy storage that has a roundtrip efficiency of 35 percent of so, the amount of electricity required to produce a kilowatt hour of energy output would be the inverse of that. So, one over to roundtrip efficiency would be the value that you will specify for amount of electricity feedstock required.

Here, we can see as an example, natural gas for flexible power generators, this is the heat rate in terms of MMBTUs per kilowatt hour electricity. If you have CO2 co-production, that would be specified, again, on a per unit of energy output. One example I want to point out is that PHS has significantly less electricity requirement for power generation. That is due to its much higher roundtrip efficiency.

Another item to specify in case you're interested to model hydrogen co-production, that would be the amount of electricity required to produce a kilogram of hydrogen. In this case, we have 56 kilowatt hours of energy to produce each kilogram of hydrogen that is co-produced. Also, we specify a percentage of time that the system would be idle. So, for example, if you're producing hydrogen for charging purposes 30 percent of the time, and you're producing power 10 percent of the time, that remaining idle time, we still want to preserve 5 percent on top of any co-production time. Electricity for co-producing would be expected to be more expensive since we're not just taking off-peak electricity, but we're taking a larger set of the electricity hours and prices throughout the day. So, we have a separate input for electricity as well as we have a value that one can specify for co-produced hydrogen.

In terms of sensitivity parameters, the user can specify the parameters listed here. So, for example, storage duration, the range specified here, would be able to inform two things. One is a sensitivity analysis as a tornado chart, which I'll demonstrate later, informing the minimum and maximum around the baseline so the system will be tested for 72-hour duration rating all the way up to 168 for 7 days' worth of storage. The other parameters – so, for example, cost of electricity, the user can specify ranges. The other way that we use these ranges is for Monte Carlo analysis. Those are assigned triangular distributions. The Monte Carlo analysis would test the model against each of those ranges.

In terms of output, the model provides visuals for three parameters. So, the first one is duration rating sensitivity to levelized cost of energy. Here, we can see that we have all the technologies that the user has decided to test, and their total levelized cost of energy is displayed on a side-by-side basis. The other output that is available is a breakdown of each of the technologies in terms of what items sum up to arrive at the levelized cost of energy. Here we can see a breakdown, for example, of electricity, charging, storage, so on and so forth.

The last output before we get into Monte Carlo analysis is the sensitivity analysis of individual parameters. So, for this particular system, cost of charging was the most sensitive going from one to three cents per kilowatt hour. We can see that 1 cent per kilowatt hour, that corresponds to $336.00 per megawatt hour. At 2 cents, we're at 365. At 3 cents, we're at the right side of the tornado chart. So, the next most important parameter is roundtrip efficiency, system life, so on and so forth. That's the way to read these likes of charts.

The user can select which chart they would like to focus on. So, for example, if the select HDV-PEM, the model will populate the pertinent graphics for it.

If the user is interested in spending an hour or letting the computer crunch for about an hour, they can generate Monte Carlo results, which are violin charts, as we described earlier in the talk.

The model, itself, is self-documented. Just downloading the model will provide you with a brief walk-through of how to use it as well as each of the inputs for row headings or column headings of texts which described what the inputs are expected.

With this, I'm going to go and do a quick demonstration of the model. So, this is the model Rook approximately has downloaded. We might have a few more updates to it. The first step that you're going to see is a description of it. Within the description, the one thing that's important is our color legend. Anything in yellow expects an input from the user. Anything in blue, do not modify it. Those contain our formulas for the algorithms of the model. Some of the key results are in green.

The other important thing is a walk-through. There is a description of the tabs as well as a line-by-line walk-through, which I'm going to spare you. Lastly, the model has technologies populating in terms of capital costs, performance, as well as methodology. We do provide reference links for all of these and those references can be found in the description tab. One thing I would like to note is that here is a publication that is in the works that would have a much more detailed description of how we arrived at the various costs and performances.

The technology specification tab is the main interaction point for users. In the upper left corner are the global inputs. We have, in the center section of this, the users can provide information for various technologies side-by-side. In here, we're provided these sets of technologies. We have capital cost specifications, operation and maintenance specifications, feedstock usage, and how the system is being used, as well as we have contingency for hydrogen co-production if that's an interest. We have descriptions, references for issues of technologies. If the users would like to look at where we got any of the technology costs, those references are down here. Sensitivity parameter tab that we describe. These are the inputs for uncertainties and variabilities that you would like to have tested.

Capacity factor definitions, these are extracts from detailed grid models. These are coefficients derived from 85 percent renewable scenario, and they inform capacity factors that are being used within the model.

When the user specifies a particular technology, they can also look at the detailed capital costs for it. So, for example, the user has specified HDV-PEM with salt and we can see the charging capital in terms of dollars, discharging capital, electricity consumption. Any financial assumptions are also in here so we can see the tax rate, the type of depreciation, the depreciation periods, capital structures, so on and so forth. Report tables would inform the user on annual basis what their income statements would look like, cashflow statements, balance sheets. All those parameters are provided. Lastly, the overwrite tab allows users to provide, for example, detailed electricity price structure of price profiles. For example, if you would like to look at annual energy outlook prices, you can provide those, or you can use a flat input such as two cents per kilowatt hour associated with an escalation rate.

So, with that, as an example of the model, what I would like to do is add a system with hydrogen co-production. So, what we will do is we will take the inputs that we have for HDV-PEM with salt. We can strictly copy the entire column. So, we take all these inputs. Sorry, on full screen apparently doesn't let you _____. Okay, so we populated that. We'd like to differentiate it. We put a dash-two here for the name. The way we would like to differentiate this case is to say that we can have hydrogen co-production. That's signified with putting a one as a key here. We want to sell that hydrogen for, let's say, $3.00 per kilogram. Once we do this, we would run the update. These calculations take about two minutes. They would go through each of the technologies and compute the base case scenario as well as for the tornado chart that is updated that will compute the performance for each of the technologies of the extremes of the variability set that we've provided.

While that is running, I want to point out a couple of things. In the model, as would be published, we have a certain number of technologies that would pre-specified. So, we have hydrogen technologies. We have vanadium redox batteries, pumped hydro, adiabatic-CAES, thermal storage. But for the users' imagination, you can consider ammonia as energy storage, for example. Ammonia can be produced by electrolysis of renewables using air and hydrogen to produce ammonia, and that can be cheaply stored in cryogenic settings and then returned to power with various technologies. There are other technologies such as gravity energy storage, liquid air energy storage, batteries of various chemistries. What the user would need to do is capture the characteristics for charging, for storage, and for discharging, then can populate the model with that set of information.

In terms of flexible power generators, we have turbines. So, combined cycle turbines with carbon capture and sequestration or without capture and sequestration. We have compressed air energy storage as well as ethanol turbines. But, again, if the user sees a potential for nuclear being able to capture dynamics required for flexible power generators, they can provide those characteristics in terms of fuel consumption into the model and they can model that. Same thing with geothermal or maybe even conventional hydro. Currently, conventional hydro mostly produces power as baseloads, but we can see that that can be potentially have some of the dynamics in there that would allow the systems to provide flexible power generation.

So, when the model is running in the lower left corner, you can see an update. It just happened to have finish the analysis that we just ran. We can look at some of the results. If we use the drop-down menu in the upper left corner, we can see that now we have a system called HDV-PEM Salt-2. When we click on it, we'll see that that system is highlighted in here as well as we can see the breakdown of levelized cost of energy and sensitivity of the system for the various inputs.

One thing that I would like to point out is that in this particular scenario, the hydrogen sales happen to be of substantial contribution to the levelized cost of energy. Actually, in this case, they happen to dominate the revenue stream hence the capital cost cut attribution and operating expenses for producing electricity are greatly subsidized by that whole product of hydrogen. Here we can see how this levelized cost of energy relates to the other technologies. Lastly, the breakdown of that technology in terms of sensitivity to each of the input parameters.

I'm going to switch here back to PEM with salt. The one thing I would like to demonstrate is that it's really hard to see the intercept points between the selected technologies and all the other technologies. If the user scrolls down, they will see they are precalculated for them. So, for example, we know that this technology and this technology – so the highlighted technology – they intercept at a duration rating of zero-point-seven hours. With closed-loop pump hydro, they intercept at 16 hours. So, this would be the point of indifference, whether one technology or the other technology, they would have the similar levelized cost of energy. So, we see the intercepts here as well as the levelized cost of energy in terms of dollars per megawatt at 120 hours. Those baseline values are computed here. So, that's just for users' convenience.

If the user has time, they can also click this button, and in an hour, it will calculate the levelized cost of energy violins for all the technologies including that additional technology that was added.

In case you download the model and you would like to restore the defaults, I have provided two buttons. One is for restoring current technologies as well as future technologies. If you're testing a relatively unique system of your own, I would suggest not using these buttons, because they will override any inputs that you put into the technology specification tabs. With that, I would like to open to questions.

Neha: All right, thank you, Michael. So, we've got a lot of questions in the chat box. So, I'm going to start with, "Can you speak to how you came about the assumptions for electricity price since they seem lower than what is the case mainstream today?"

Michael: Right. So, the modeling framework that we use is PLEXOS. It actually calculates variable operating expense and does not calculate total cost of electricity from the grid. So, as such, it's not a useful way of informing the levelized cost of energy that's generated from renewables. We turned to long-term PPAs from renewables such as solar and wind, which are currently approaching two cents per kilowatt hour. If that is the prevailing power generation equipment on a future 85 percent renewable grid, we would the utility scale electricity would have that price profile as well.

Neha: All right, thank you. Can you speak to the role of pipes in hydrogen storage? I think there were questions just around that concept because it isn't as mainstream.

Michael: Right. You said pipes?

Neha: Right. Yeah, underground pipes.

Michael: Yeah. So, there are a number of technologies being looked at out there for storing hydrogen at low cost. One of them is burying pipes underground in order to get some benefits of isostatic pressure in order to reduce the amount of steel that's required for storing that hydrogen. So, we looked at research and systems analysis in that area, and we captured the expected capital costs as well as maintenance, and we cast that as one of the potential case studies for hydrogen. With that said, it is more geography agnostic. You're not necessarily relying on geologic storage, either a salt cavern or hard rock cavern to store hydrogen, but the costs are significantly higher. So, in order to characterize the expected applicability of that technology for hydrogen energy storage, we added that technology storage characteristic to the model set.

Neha: Thank you. Then, there were a few, and I just want to note specifically the storage cost that were assumed were an output of other work that we have ongoing at Argonne National Lab that's also very near publication. So, hopefully, once that's published, you can get more information about what was assumed there from that work. Also, as there were questions around how you – can you speak a little bit further to the learning rates and how you assumed improvements in technology and then, also, how that relates to the relatively low hydrogen prices that we had, since I know those were all related?

Michael: Yeah. So, for learning rates, we did extensive literature search in terms of what are the current capacity of each of the technologies, how much that technology has been produced in terms of power bases, how many gigawatts have been deployed of that technology, and we looked at literature in terms of what learning rates would be applicable for each of the technologies. So, if you have a doubling of a particular technology, how much learning can you have per doubling of installed capacity? So, that informed our projection for 200 gigawatts in terms of current costs, and for each doubling, how the learning rate would be expected to reduce price profile for power as well as for energy. So, the technologies have either had the substantial amounts of existing installations and they – so, for example, if we look at combined cycle plants, there are a lot of combined cycle plants out there. They still are expected to have a learning curve on top of that. But due to the fewer number of doublings that we would expect that price changed between current costs and future costs of 200 gigawatts is expected to be roughly the same.

Other technologies – so, for example, if we look at salt caverns, we expect to have very little learning. The technology – the literature does not expect substantial learning on how salt caverns are formed and we see a flatter profile on some of these technologies. Other technologies have – so for example, vanadium redox batteries, there aren't very many of them out there, but according to literature, there is a material cost that, even due to learning, would still be present in the total cost of storage. So, that also informs the potential learning for each of the technologies. Lastly, I want to say that we're expecting to have a publication on this, which goes into the full depth of analysis on how the computations are done as well as how the learning rates and what individual sources for data were used for each of the learning rates.

Neha: Thank you. Can you also clarify, when we're speaking to a given duration of storage, like 120 hours or any different number, are we referring to continuous output for that period of time, or is it 12 hours over 10 days?

Michael: Yeah, so when we say 120 hours duration storage, that means if the system were fully charged, and then it start – if it were to start producing power, at rate of power at the full 100 megawatts, it would take five days before that system is completely discharged. Obviously, that's not how the systems are operated. The systems are, instead, operated throughout the year with partial discharges. So, throughout the year, they'll typically have one full cycle of charge and discharge cycles. So, one deep cycle and a lot of shallow cycles. But the rating of the system is basically it's ability to discharge for a long duration from full to empty.

Neha: Thank you. Can you speak to the difference between adiabatic and diabatic storage?

Michael: Yeah. So, there are two types of compressed air energy storage. Let me start with diabatic compressed air energy storage. That's a system that has been demonstrated. In both systems, air is compressed using a compressor into a storage. The compression energy is exhibited in two ways. One, it induces high temperature and compressed air. That heat from compression is storage in thermal energy storage. Actually, hang on. For diabatic air, we just compress the air and store it on the ground. On the way back, when we want to discharge the storage, the air is fed through a turbine and heated with combustion of natural gas, and then it's run through a turbine to generate power and run a synchronous generator.

Adiabatic air is different. Instead of using natural gas to reheat the air before the turbine, compression energy is stored in thermal energy storage. When the system is discharged, the air is reheated through that thermal energy storage before it goes into a turbine and the generator. So, basically, diabatic compressed air energy storage uses natural gas and adiabatic energy storage uses compressed – it uses thermal energy storage for the thermal portion of the cycle.

Neha: Got it. Thank you. Can you speak to whether we included pumped hydro from existing reservoirs or only looked at brand-new reservoir development?

Michael: Right, very good question. For current capacity, we estimated the global installed capacity of pumped hydro. For future capacity, we added additional 200 gigawatts of storage. That storage would, obviously, have to be new capacity.

Neha: All right. Thank you. Can you speak to kind of the minimum level of load that cell batteries can tolerate? So, what's like a minimum level of cycling baked into these assumptions?

Michael: Yeah. Details about turn-down are not captured within the PLEXOS models. The PLEXOS models are – they're fairly very detailed models, but they do have to do some simplifying operating assumptions. So, in terms of turn-down, I would actually – if you email me that message, I can put you in contact with the folks that actually run the PLEXOS models. If they have a method for capturing turn-down, they will speak more intelligently to that than me.

Neha: Sounds good. Thank you. There's lots of questions about sensitivity of this analysis to renewable energy penetration. So, I know we were leveraging existing work in that space. If you want to add anything else to that about how sensitive this is to that _____ _____ penetration.

Michael: Yeah. There is very large sensitivity at lower penetration rates. So, up to about 50 percent the amount of long duration energy storage is fairly minimal that you would need. At higher penetration rates, at 80 percent plus, you would have a different profile. Possibly as future work, I would suggest, as we get a larger portfolio of penetration analyses and how storage and flexible generators are used within those scenarios, we could probably provide a set of calibration factors here to – so that the user can select, "I want to test my system in a 50 percent scenario," "In 100 percent scenario," so on and so forth. But at least for right now, we have provided the calibrating coefficients for 85 percent scenario. We will expect, as you go further up, that you would need more and more long duration energy storage. I know that some scenarios are starting to come out with 100 percent renewables. As those come off of peer review, we can see about informing the model with those results.

One other thing I would like to point out is when you run PLEXOS grid models, before that, there is a capacity expansion modeling done with a model like REEDS. That would be informed by the capital costs, variable operating expenses, so on and so forth of each of the technologies. As part of us going through these analyses, we capture the characteristics for each of the technologies in order to inform REEDS and then from REEDS and PLEXOS, we get performance characteristics. So, we will be going back and forth and recalibrating each other in terms of how much penetration we would expect for each of the technologies, how they would maybe use them in the future.

Neha: Thank you. So, there were lots of questions about how to use the model. So, the model link is on the slide, then, definitely, we can share it with folks after this call. But one of the questions that came up was, "Does the model have baked in constraints such that if somebody makes an unrealistic assumption of one that doesn't make sense, will it flag that?"

Michael: Yeah. One thing – I have certainly done that is accidentally put in the wrong parameters. Since we have so much visualizations coming from the model, you will have lots of clues if the levelized cost of energy looks odd to you or if you see that the model has obviously errors and LCUE, if it says, "Not a number," or something is divided by zero. I've not provided substantial error capturing just because, typically, the errors become fairly obvious once you run the model. The runtime is fairly short. So, the cycle between providing inputs and getting feedback is fairly short. But if we find that the users are running into issues where it's not clear if the results that they're getting are valid, we can add some additional error handling. This is a new model. I would expect to have tweaks as we roll it out.

Neha: Thank you. Then, I think this is going to be our last question. Can you speak to why liquid-based storage wasn't included in all of our pathways for gaseous-based storage, hydrogen storage?

Michael: Yeah, cryogenic air storage is very interesting. We would have liked to have it. Generally, we pick technologies that are high TRL and have a lot of literature that we can tap into to inform capital costs, storage costs, new operating expenses. Liquid air was not one of the technologies that we had high confidence in the parameters that are seen in those technologies. So, yeah, we're providing the flexibility of users to test that technology by issuing this model so that users can test the costs that they have on hand.

Neha: Got it. Thank you. I think there were also questions about liquid hydrogen. The reason why that was left out was just because it would bring down the roundtrip efficiency of the hydrogen-based process and offset it is _____ viable in the near term than a gaseous-based path.

Michael: Yeah. We actually have good information for liquid hydrogen. At one point, we tested up to about 200 different variations with salt oxide fuel cells and alkaline electrolyzers, and storing liquid, storing gaseous, storing some other hydrogen some other ways. There are a lot of different permutations. But absolutely correct, as Neha said, one thing that we found with that system is, yes, it does set a very low cost of storage, but there are a couple of items to be concerned about. One is boil off if you're storing on a seasonal basis. The boil off could be used for your day-to-day energy source, but it is something to contend with as well as it does reduce the roundtrip efficiency. It does take quite a bit to liquify hydrogen and the capacity factor would be even lower for liquid storage over other technologies.

Neha: All right. Thank you, Michael. So, I guess just the last thing I'll note is that the link to the model is available here. These slides and an audio recording will be posted probably within about a week. Then, this model is in beta testing, so any feedback that we get, we really appreciate. So, if you do use the model and you have any feedback for us, please feel free to relay either to Michael, himself, or to anyone of us on the DOE end. With that, Eric, do you have any other closing comments?

Eric: Yeah. Thank you, Neha, and Michael, for the presentation today. Thank you everyone for joining. Michael, if you could advance to the final slide at the very end. As Neha mentioned, I'll remind everyone that – one more, please – the recording and slides will be available online at the DOE.gov website. So, please, check back there soon as well as future topics. So, with that, I'll wish everyone a great rest of their week, and goodbye.