All Episodes

[Episode #29] – Grid Simulation and Wind Potential


What combination of power generators on the U.S. grid produces reliable power at the lowest cost? Or, what’s the most renewable energy that can be deployed at a given grid power cost, and what kind of transmission capacity is needed to support it? How would the U.S. grid be different if it were one, unified grid with more high-voltage direct current (HVDC) transmission capacity? What’s the most productive design for a wind farm? How might weather and a changing climate affect future electricity production from wind and solar farms? And how much renewable power is really feasible on the U.S. grid?

These have been devilishly difficult questions to answer, but now advanced mathematical simulations are beginning to make it possible to answer them much more quickly…and if quantum computing becomes a reality, we could answer them instantly.

In an homage to Comedy Central’s Drunk History, this episode features a conversation conducted over several pints of IPA with a mathematician who recently developed such a simulator while he was working at NOAA (the National Oceanic and Atmospheric Administration) in Boulder, CO. His insights on how the grid of the future might actually function are fascinating, and will likely shatter some of your pre-existing beliefs. It also contains a few nuggets for the serious math geeks out there.

Guest: Dr. Christopher Clack is the founder of Vibrant Clean Energy, LLC, a software and services company that focuses on optimization techniques and renewable energy integration into the electricity grid. Dr. Clack was previously a research scientist for the Cooperative Institute for Research in Environmental Sciences (CIRES) at the University of Colorado Boulder working with the Earth System Research Laboratory (ESRL) NOAA for half a decade, leading the development of the NEWS simulator. Dr. Clack received his first class BSc (Hons) in mathematics and statistics for the University of Manchester in the UK. He then went on to research applied mathematics and plasma physics at the University of Sheffield in the UK. During his PhD, Dr. Clack completed an area of study centered on nonlinear resonance theory within the framework of magnetohydrodynamics (MHD) that remained unsolved for twenty years. The theories derived have helped our understanding of the Sun as well as possibilities for fusion reactors, such as ITER.

On Twitter: @clacky007

On the Web: Vibrant Clean Energy

Recording date: October 21, 2016

Air date: November 2, 2016

Geek rating: 8

Chris Nelder: Welcome Christopher to The Energy Transition Show.

Christopher Clack: Thank you.

Chris Nelder: I guess I should mention that we're actually taping this in person over pints of good microbrew, IPAs to be precise. The only other time I've actually done that was with Mason Inman's interview in episode 13 and that went very long indeed. So listeners forewarned, we're going to have a few pints and do maybe like the podcast equivalent of Drunk History which is a show that I just love.

Christopher Clack: Beautiful.

Chris Nelder: Brilliant. All right. As a mathematician who recently worked on modeling the interaction of climate and renewable energy over at NOAA, the National Oceanic and Atmospheric Administration, you have a pretty unique point of view. So I'd like to start with some of your modeling work in that area. First let's talk about your National Energy with Weather Systems Simulator, known by its acronym NEWS. How does that simulator work and how can it be used?

Christopher Clack: So yeah, thank you. I basically went to work on this model that became NEWS from a standpoint of I want to be agnostic about the grid, about what goes on the grid, I don't want to have a preference in technology. I also want the model to be able to find me the least cost system so I want cheap energy for people. And I wanted to be able to represent the generators properly, including wind and solar. So what I mean by that is what I wanted to make a model that had a lot of granularity for the wind and solar and use real data for that so that we can really see what happens when wind and solar is put into grids. And what I wanted to be able to do is build it from the ground up so that I could do very large scale, so national continental scale type grids all the way down to utility sized grids, but at the same time be able to handle the operation of the grid and how it works. And so in its simplest form what it does is it allows you to look at how the grid would evolve in the future under different scenarios, finding the least cost while also showing how the grid would be operated, so how the markets would work and how the different generators would inter play with each other. And so then we as humans can look at that and see why these things are happening because we try to keep the assumptions to a minimum so that we can try and figure out what it's doing, why is it doing what it's doing, and how does that enable us to transition our energy grid from from what we have now to what we are going to need in the future.

Chris Nelder: So this simulator then incorporates a variety of data sets. You've got some weather data, you've got some data on the economics of grid markets, and you've got some data on how these different power plants are dispatchable. I mean that's really quite a lot of data that you've bunched up in there.

Christopher Clack: Yes, so we got the how people behave, so their demand profiles, so how much energy they are using. Then the conventional generators.

Chris Nelder: So you've got some ramping knowledge in there somehow.

Christopher Clack: Yes ramping knowledge. How the minimum on and off times for these bigger plants that are needed. And then the backbone of it is a very high resolution weather data that allows you to look at the transmission lines, as it gets hotter so they sag more, they are less efficient.

Chris Nelder: Oh wow, so the transmission physics are in there too?

Christopher Clack: Yes. At the frequency that you run the grid on the voltage control on the transmission grid. The really only thing that we don't have is a really detailed distribution grid representation. We just have sort of an implicit version of that because you then start multiplying the number of power lines pretty fast and when we did a small subset of that we didn't see a very large change or almost no change at all to the big scale picture. But you can run it with the full distribution grid but you can't run say a national scale with all that data because it's just too much data.

Chris Nelder: Yeah yeah. Well I wouldn't imagine there's a ton of variability at the distribution grid anyway, I mean you know you have certain limits or you don't want to blow a distribution transformer but beyond that like you know.

Christopher Clack: Well the only reason we were considering putting it in there was just to the rooftop solar for the residential and think how they interplay, but we decided that when we did it explicitly that difference was so small that the amount of rooftop solar you can actually put in when you're looking at a country like the United States is very small compared to the actual amount of power needed. And so it didn't really overly effect it. The reason I wanted to look at the distribution a little bit was electric vehicles and how they would interplay with the grid, pulling and pushing power. But we feel into the large scale all that noise sort of disappears and becomes a more aggregate thing anyway. But I as a mathematician I constantly see and know what's wrong with the model, I want to improve it, constantly striving to do that and that allows me to see insights into you know whether you need it or not. So one of the tricks of a mathematician is to know what you're missing but know what effect that bit you're missing has on the grid and not just on the grid but your equations that you're looking at and hopefully able to explain to people that it doesn't matter as much as some people might think.

Chris Nelder: Yeah just to be able to sort of rate the magnitude of the uncertainty.

Christopher Clack: Yes exactly.

Chris Nelder: Okay. So what are some of the ways that this model can be used?

Christopher Clack: Well the simplest way and the first way we used it was that we wanted to look at finding an optimal mix of generators and transmission and storage and demand response with energy efficiency to create the least cost solution of the grid. So sort of the base case scenario, we just put the all the assumptions about the costs and everything and we work out what the cheapest grid would be. And then we can take that and see what that looks like, what the disposition is of all the generators and everything. And then from that we can then do sensitivities as well to that. So we could say okay well we got 80 percent reduction in carbon but we want a 90 percent reduction so we'll enforce that extra 10 percent. And what does that sensitivity then tell us about how much more it would cost or what technologies we need to place in. So that's sort of the traditional thing, you do a baseline and you can do multiple sensitivities off that. The other thing is you can do, not stochastic but you can do layered solutions, so you can look at different sized grids. So what you could say is you could say have something like say PacifiCorp, which is a grid in the western US, that wants to connect with California ISO for the EIM market, which is energy imbalance market. And you can say well imagine that those two are separate in one scenario but you have an underlying scenario which is that they're connected together. And how does the placement of generators change when you when you have the model understand that there could be a possible universe where you might be connected versus you might also be disconnected. And how does that change where you would select siting versus just one problem or the other. And so what I mean by that is you could do the traditional thing which is PacifiCorp does a plan and says this is what we're going to put up with our generators, and California does that without any knowledge of what PacifiCorp has done. But then one day they decide we're going to connect together, and does that disrupt the generators that exist there now? And if it does what difference does that make if you had thought about it ahead of time that there was a say a 5 percent probability that these two things might happen. Can you hedge against retiring assets early. So that is one thing you can do, and another thing you can do is set the price of electricity and maximize something that you want to look at. So the traditional one is maximize renewable energy or maximize storage penetration for a region or you could also minimize. So you could minimize say sulfur dioxide emissions or water use in a particular place. Or you can set the cost of electricity that you're happy with and then rather minimizing costs you're saying this is the cost and then we need to find out what's the most you can get on the grid or the least you can offtake off the grid given that constraint, and so you can add policies in, you can add different disruptive technologies in. So one of the ways I like to use is you have these wacky new technologies coming out, for example things like flying kites that will generate electricity. Try to model how they work and then you can put it in and say well what cost do they need to get to you to be a disruptive force within the grid. And you can do the same for electric vehicles, you can do the same for demand management, anything like that. So you can plug and play if you will all these different scenarios within the same modeling framework. So you know the underlying model is going in and you can then test all these different hypotheses.

Chris Nelder: Could you actually do that with different types of generators at once?

Christopher Clack: Yes. Yes, so you can make multiple changes at the same time. So you could say I got these kites and I've also got wind turbines that are now 200 meters tall rather than 100 meters tall, and we got solar panels which are four times as efficient than they are today, we got CSP which is really cheap, and we've got natural gas has got CCS attached to it, and you can do all that at once, and you can take a away if you had all those options which one would win or which one penetrates the grid the most or which one loses. But then the key to the model is it also tells you why it lost. So you'll be able to go in and see all the non-solutions and be able to say okay well the the natural gas with CCS for example lost because it was 5 percent more expensive than say CCS, sorry then biomass for CCS or something like that. So you can see why you've got competed out to give us an idea of how close we are. So say if it was like one hundredth of one percent and you could say okay well there isn't really much of a difference between these technologies and so let's do both because they are equally effective.

Chris Nelder: Could you adjust the sensitivity on that to say well don't let it compete something out if it's plus or minus 2 percent?

Christopher Clack: Yeah, so you can you can make the price endogenous basically and you can say if you select this one then other one gets slightly more expensive or slightly cheaper when you replace it with that technology, and you can allow it to fight in some sense within the model. That makes it solve, I don't wanna get too geeky yet, but for it to solve like that you're making it be a harder problem, so you would want to do that on smaller grids rather than say a national.

Chris Nelder: You know I hadn't really thought about the fact that you could use this model to model the operation of an energy imbalance market or to model the way that these other utilities joining essentially the California ISO would actually work as a balancing mechanism.

Christopher Clack: Yes so the cool principle behind the model is it's blended capacity expansion production cost, which maybe a few of your listeners will know what they are. Normally two ends of the spectrum of modeling types. I always believe that there's an infinite spectrum of models that you can build, they are all wrong. Sometimes my former boss would say don't say that they're wrong say that they're inaccurate. You know because people get turned off by you saying it's wrong, but I like to point out that you know it is a model, it's not real life.

Chris Nelder: And you're a plainspoken man after all.

Christopher Clack: I am a plainspoken man and some people don't like that but most people do like it because they know when I say something I mean it.

Chris Nelder: I'm a big fan.

Christopher Clack: Thank you. But the point is you can do these things together which means you necessarily remove a lot of the obstacles you might need to connect models together and all these things, and you're using the same model framework and hardware for each of the different systems so you don't have all these extra complications. We are using different models that were built by different teams.

Chris Nelder: I have heard some horror stories about people trying to integrate those models coming from different systems and platforms and just I mean even just creating mutual interface is just a horror show sometimes.

Christopher Clack: Well this is the way, and I said it somewhere I was once reviewing a panel, and the thing I said was the sum of the two is worse than each individually because you've got the worst components of each one coming together. And that's not, I wasn't being harsh, that's just the facts of life. If you're connecting these things to communicate you have to distill it down to a basic level of what you're communicating. And if one model is really poor on one thing and the other model is poor on something else the combination will be poor on both, because you can't transmit the data between one and the other. And that's kind of where I came from for them building the NEWS model was I wanted to have both integrate together so that you can look at what happens when you think about what you would do to dispatch whilst you're also thinking about what new generators or transmission or storage i'm going to build.

Chris Nelder: Wow that is super cool stuff. So one of the things that I think is super interesting about your simulator is the way that it can integrate this really large amount of weather data. Can you describe a little bit about how that aspect of the simulator works and just how much data we're talking about. And for that matter how you deal with the computational requirements.

Christopher Clack: Oh, it's very complicated. I mean I don't know if I understand it now. I'm being facetious of course. So we, when I built it my former boss really was interested in whether, he's a meteorologist and he really wanted weather data to be treated right. And basically the NOAA and the National Weather Service use a thing called 3D variational analysis, or four dimensional variational analysis as well, which is a really complicated way of saying they blend models with observations and a lot of observations, tens of thousands of observations, to give us a very good picture of what the atmosphere looks like at that time across whatever domain you're looking at. So we were technically looking at almost all the time North America. So what's going on in the atmosphere all the way through the 3D column, so we have plane data, we have satellite data, we have ground based measurements. They all go in and we do this variational analysis, and what we find is then this 3D matrix of information for the weather data. Then what we do is we take that raw data for a decade, so every hour this happens and we have this for 10 years, and then we put that through power algorithms for all different technologies. So for wind turbines, how do they extract energy from that atmosphere? How does solar irradiance impact solar panels? And then we extract some salient features that are needed for the operation of the grid. So temperature affects how much demand you need because of air conditioning demand so you need that. Wind speed affects how quickly your panels cool down on your roof. It also blows the power lines and so it affects that. So you've got these other data points that come in, and what you do is you actually write that into binary because it's the simplest form for the computer to understand, and over these long periods at hourly resolution for all the sites across the US. And because we write it into binary it becomes very efficient for the computer to read because it's in the language it understands. And that's something that's very different straight away and that also compresses the data rapidly. And so when you do that and you bring it in to the model and then it builds the model around that structure, so it builds a whole model into binary so that when it goes to the CPU to actually solve, it's not having to translate that. It's all in computer language.

Chris Nelder: Yeah, it's using machine language or assembly or something.

Christopher Clack: Yes exactly. To make it faster. And so we can read in the model roughly 20 to 25 gigabytes of data in less than 30 seconds into the random access memory and then we have a way to tell the CPU and the cache information that it needs that is very important at that point so that is the minimum amount of communication back and forth from the hardware. And that really stems from computers are n't getting any faster. They just throw more and more CPUs on the computer and these types of problems unfortunately are much more suited to one very very very fast processor and a lot of memory. This is many processes. You can change the problem but then it leads you down a rabbit hole of other issues. But the basic format of the problem is make it into a language that is simple enough for the computer to be able to read directly and fast as possible. Otherwise the amount of data you've got is so large that you'll never.

Chris Nelder: How much are we talking about here? Petebytes?

Christopher Clack: So the standard model we run as is of the order of one terabyte of RAM. So that's when it's actually processing. So that works out to be you know 20 to 30 terabytes of data just in its raw form and then if you want to run it in dispatch mode which means you're not doing the classic you can get it down to a five minute resolution rather than hourly, and you're looking at 300 terabytes. So it's an order of magnitude bigger. Because of the way you're doing it, you then need to move to solid state hard drives, things like that, rather than RAM because there just isn't machines big enough.

Chris Nelder: Yeah, you're not going to be doing this on a home PC.

Christopher Clack: No. I'm making a model that you can run on your desktop, but obviously it's simplified in terms of the amount of data you can use. But it will give you a broad brush example. But the problem behind it is we had to purchase, when I was working there we had to purchase special hardware and at home, in my new business i've built special hardware that's built around that which is made a lot faster because the specialist hardware I've built specifically for this one problem, and it's been illuminating to try and build. I feel like I'm turning into Apple, i'm building the hardware and the software so it works. But it really does make a big difference. It's very useless for other problems that require parallel processing but for this specific problem, devoid of me inventing a new way to do linear programming, which would make me very very rich, it's just the way the problem is. And there's lots of ways you can approximate it, but I don't like approximating it for reasons we might discuss later.

Chris Nelder: Yeah I think we will get to that. Okay so you've come up with some really interesting findings from this simulator and I'd like to talk about some of them for a minute. So first you've modeled what the effect would be on the US electric system if it were actually one 48 state electricity system rather than the system we have now, which is actually made up of three large interconnections: the east, the west and Texas. So if it were just one system, how would that actually change the integration of renewables on the grid?

Christopher Clack: Yes so I want to get to a little bit of nuance here. We don't actually have to have one total grid, they can be asynchronous with each other because we're using this HVDC technology, but I always think of it as one grid because that's what in effect you're accomplishing, but you don't have to get rid of the existing sort of grid structure as it were, you just have to have a sort of an overlay on top that will facilitate you know the eastern interconnect talking to the western interconnect, which doesn't happen today. So if you can do that what you find is you get a massive explosion of wind and solar power being produced because wind and solar power are a variable generation that appear random to the layperson like me stood in the street, wherever I am in the world it seems pretty changeable, the weather. But actually when you zoom out from that and you look at a bigger picture, there's patters that emerge and there's very very well understood physics behind it. And so I call it chaotic because it's fully described it's just difficult to predict long term in the future. But there are patterns there and there are correlations there and traditionally has been thought the correlations are bad but actually if there's a correlation it means you can predict what is going to happen. And if you can predict what's going to happen, if you've got a computer that's smart enough and you give it enough knowledge it will be able to select sites where it can see the patterns faster than humans can and it can construct this grid that's put together that's like a symphony or an orchestra working together to make beautiful music rather than you know a five year old on a drum kit just smashing a few drum pedals or whatever. But you have to plan it from the start with that in mind to give you the blueprint, and then the model will tell you how to build out towards that goal. And the reason you want the blueprint first is that you want to be able to see with the model what's the least regret path. Because we don't know what's going to happen in 20 years in terms of technology. But there are a few things that we do know. One is we're on a bad path now, and we haven't got a big clock, and a clock in terms of time left to solve the problem. So we have to start with technologies we have today and we have to start building today. But we want to be on a path that allows us options to change course if we suddenly find we've done something wrong. And things like a national market or a bigger market makes it easier to change course than small isolated markets because once you're sort of locked in you can't really change. Once you've decided to only have solar panels on the roof and a battery, you kind of locked yourself in because that's a huge investment at the beginning. But if you do that and you're connected to the grid, if suddenly you know your battery goes out you can still buy power from the grid as well.

Chris Nelder: So I'm guessing that a single grid, whether it's a notional single grid or an actual one, would allow you to integrate more renewables because of the fact that it's a larger balancing area, but then you need some more transmission capacity.

Christopher Clack: Yes. So what ends up happening is the electricity, the fuel if you will, is free. But it becomes complicated because the fuel moves about. You're like chasing your own tail if you will. You have to build generation and it has to be static, but sometimes there's power there and sometimes there isn't. And the art is you need to build them in lots of multiple places scattered about a large area so that you can actually see the symphony playing and you can see that these things are correlated. So a prime example is if it's windy in Seattle in a few days it will probably be windy in Montana. And it's pretty regular. And so what's happening is that's a wave that's moving across, and because the model can see these patterns from all the data it has, it can say well I can extract some energy in Seattle for two days and then two days later I can extract it from Montana. And what it is able to do then is it can connect them with this transmission and it can only build it if it's economical, so if it's too expensive we'll never build it. And what you find is the remote source plus the local source plus the transmission is cheaper than all local or all remote. And so you have this confluence of this tradeoff, between remote generation and local generation. You get this curve that happens with the transmission that means that you have this tradeoff and you can decide to have it all locally but you end up paying more. Or you can decide to have all remote and you end up paying more, but you have a tradeoff if you have both.

Chris Nelder: So what is the mechanism by which it would actually be cheaper to have partly remote partly local versus all local?

Christopher Clack: Well the mechanism is that you've got the variability of the weather. So if you had it all locally you would end up having to back it up, a significant portion either through storage, which is kind of my idea of backup is the traditional thing. Or you have to have some fossil fuel generation which then you have to burn a fuel and you have to have a lot of capacity of it because the highs and lows are bigger. Whereas if you have some local in some far away you can build less transmission because you have some producing locally some of the time. But then you have the remote resources that can then be powered in to your location when the local sources disappear. And that means that you can have less reserves, you can have less storage needs, and you have less fossil fuel generation needs locally because you can ship power out. And the the idea is that the model then says well we need to build a network because. A local source can become a remote source for somewhere else when you don't need it locally. And so you can then have this interplay where you build a wind farm in Colorado but some days it's being used in Colorado because that's the cheapest thing to do, but then other days you've got too much wind for Colorado and it ships it to New York because New York really needs the wind and their wind has died down but you've got too much. And then the next couple of days suddenly California needs wind, you're again producing too much. So by having a network you can then tap into multiple markets same time rather than say a PPA, which is a power purchase agreement where you're selling to one customer all the time. And that's great to start with but actually if you think long term you might go well actually I want to sell into this market for a bit because their prices are starting to rise and I could make more money. And then as that happens, as the social sort of market evolves then that would drive down prices in the long run because more people would stop doing that or more actors would start doing that and selling into lots more markets.

Chris Nelder: Right so you need that larger balancing area in order to smooth out the variability of the resource in different places. And that would enable a larger market. But you need to have the transmission capacity, HVDC I assume to make that work.

Christopher Clack: Well the HVDC is the cheapest option for it to work, so in the model we don't impose that transmission has to be there. Sometime what I don't explain very carefully to people is that the transmission is only picked if it needs it and if it's economically viable, so it could have been that we did this and it picked everything locally and it didn't need the transmission to go to distant areas. But what I found was by building the transmission it actually reduced the costs for the whole grid, so locally and remotely. And so the grid is sort of a byproduct of this need to get cheaper energy. And the cheapest way to do that is to think about not just locally but further away too. And the way I think about it sometimes is we could here in Colorado we could build cars in Colorado if you wanted to, and we could make our own Chevys or GMs or anything like that, but I would imagine they'd be more expensive than the ones built in Detroit because Detroit has all the technology and all the experts there. And so it makes more sense to buy them from Detroit and ship them here than it is to reinvent the wheel if you will back in your home state and you could do that all over the country. And commerce has worked that out many years ago and electricity needs to catch up with that. We need to produce electricity where it's most effective and valuable to the grid and ship it to the customers where they need it.

Chris Nelder: That's an interesting analogy and you know I think a lot of grid power observers would agree that a national HVD grid would be the best solution, not only to more renewable power integration but to several other issues as well like stabilizing market prices. But no one seems to think that it can actually be done as a practical matter, thanks to the resistance of various mid-continent states to running the lines over their turf or at least not without some sort of fat compensation if they're willing to do it at all. And I realize your domain is more you know mathematics and modeling than US state politics but what is your view on the pragmatics of a comprehensive national HVDC network?

Christopher Clack: Yes, so I'm lucky enough to be a mathematician, so I'm very logical. Sometimes politics confuses me because I'm maybe some people, certainly my wife, thinks I'm hyper logical sometimes and I'm not as emotional as some people. But really that's where the tire hits the ground, that's how we move forward. And so the first question I wanted to answer with the model was can we technically do it? If you can't technically do it then there's no point at arguing politically. We should do it, in my opinion. But again that's my logical brain. So I feel like we showed, and other people have showed, and NREL have showed and other studies have shown now that this is economically feasible, it is technically possible. Then you get to the stage of well does politics come into play. Yes it does. I mean we've done some recent studies where we do each state has a big hub. So each state will get access to the transmission system and therefore jobs and all the benefits that come with having access to the power rather than doing these long transmission lines across multiple states but don't do anything for the state other than having a transmission line run through their state. And I feel that's a move in the direction of the model being built in a way that is trying to answer a more political question which is well I want these jobs and the economic boost locally. And what I would argue back is that while different states have different resources and different strengths and we should connect them all together so that then, just like the interstate highway did, and so each state can do what they're good at and contribute into this bigger picture together. The bigger political things is we've got to cross the interconnects, we've got to think about those sort of things. And we've got to think about what's good for the consumers. And so those 2005 federal power acts says it has to be reasonable and just, the cost of electricity. And if we're not thinking about the large scale the cost won't be reasonable and just, because you're saying we have to build it here because we happen to be here, rather than getting what we could get it at for half the price somewhere else. That doesn't seem reasonable and just to me, that's not a logical thing to do. So you've got that side, but you've also got the other benefits of the generation being up to share power longer distances in some states where there's not very many people, not a lot of open space, particularly in the middle of the US there's a lot of space and not many people. There are some of course, there's always winners and losers. But if there's lots of space and not many people, but on the coast there's lots of people and not much space, it makes sense to me to connect those two things because we can provide power for the people that don't have much space and provide benefits to the people that do have a lot of space by paying them for using their space that they're currently either not using or they're using it for farming which could be dual use for generation and farming.

Chris Nelder: Right so you could actually use your model to say what is the highest tolerable price for HVDC capacity in order to enable all this other stuff, and then you could use that price to figure out what's the maximum price you could pay to states over which the HVDC line runs.

Christopher Clack: Correct. So you can essentially do the experiment like you just said, you can say well we want an 80 percent carbon reduction, whatever the number is chosen to be, and you can say what is the maximum we can pay to have the cost of electricity be 10 cents a kilowatt hour? And therefore you can then work out what the maximum credits could be or the payment could be for rights of way and things like that. So you can reverse the question from the traditional questions of let's minimize costs. What you can say is maximize benefits or you can also sort of do a more nuanced thing as well on top of that which is each state has to have X number of jobs or you know has to have this benefit or this other benefit. And how much extra does that cost the nation or how much cheaper is it for the nation compared to doing all locally. Or you might find that you know locally when you put all these other things in is it's better. And so the model doesn't want a national system. There isn't sort of this perverse power that has a preference, it just comes out with what's best or efficient.

Chris Nelder: And it turns out that HVDC network is the cheapest most efficient way to do it.

Christopher Clack: Exactly. And that's for every scenario we've run with any generator on the grid, so that isn't just because of wind and solar, that happens even if you've got nuclear heavy or coal heavy, it still happen because of resource sharing and other things. But the wind and solar is sort of a force multiplied. It's an additional, it makes wind and solar much more effective. The benefits aren't as big with just coal, but there is still a huge benefit for doing a national system regardless of the technology on the grid. Even with storage.

Chris Nelder: Okay. This is kind of a side point I guess, but I was sort of curious in your production model for your PV systems, do you happen to recall what kind of efficiency you're assuming for the PV modules?

Christopher Clack: I do. I'm a mathematician, I remember a lot of numbers. 17 percent is the efficiency.

Chris Nelder: Okay. So that's very conservative for today's modules.

Christopher Clack: Yes so we use 2007 like high range models when we started the project in 2011, so they have been around 5 years.

Chris Nelder: Which is you know in the same ballpark as what NREL would use or whatever. But in fact we're now routinely deploying modules in the field that are 20 percent plus. I think SunPower, they actually have a triple junction module that's 24 percent efficient now. Okay, so what is your guess as to how that greater efficiency would change the results of your model. Like what, obviously there would be more solar integrated into the system, but what would get pushed out.

Christopher Clack: I don't know if more solar would be installed necessarily. I think it probably would. When I doubled the efficiency to 34 percent, all that ended up happening really on the margin was less space was used for solar rather than more solar was put onto the grid. Yeah because you get this solar deflation devaluation happening where you put the solar in and you've used up your space, but if you double the efficiency all you're really saying is you can have the same amount of power for less space. But if you put more power in that same location all you're going to end up doing is having more curtailment at certain times of the year.

Chris Nelder: Well that's a very handy because I wanted to talk next about value deflation.

Christopher Clack: Oh wow, okay.

Chris Nelder: So looking at this value deflation question as a consequence of one of the studies you've done here, you know that the more wind and solar you have on the grid the lower price it fetches, that's the basic idea because it's biding a zero marginal cost resource, a production cost resource into an increasingly saturated market. And I discussed that question at some length with Ben Paulos in episode 3 of this podcast and I think we both agreed at that time that the likely fix to the problem would be some sort of a regulatory intervention or a market adjustment to prevent the price of wind and solar power from falling too low. But I wonder if you know your work with this model would produce a different view of that issue?

Christopher Clack: Yes, so we looked at this because we kept getting 17 percent solar over a national grid and that kept coming out pretty regularly, when there's not storage.

Chris Nelder: And it would just be purely coincidence rather that you'd wind up with a 17 percent mix for a 17 percent efficient module.

Christopher Clack: Yes because when we doubled the efficiency it was still 17 percent. So it has to do with the size. When we added Canada or Mexico you've got a larger amount of solar.

Chris Nelder: So I take it it also didn't have anything to do with the capacity factor?

Christopher Clack: No. So it comes down to the amount of curtailment that is being produced and the cost of building the transmission and all the other generators that are coming into play. And when we look at the market clearing prices from the model each hour we find that the wind and solar don't bid in 0. So at the moment they can bid in 0 because while they've got PDC and they don't have any variable costs and they know they're going to get some clearing price. But what we find is well the model eventually will have 70/80 percent carbon free generation and sometimes 100 percent carbon free generation being produced. And so the market changes from just being a production cost to a capital in production cost market where it's charging for the fact that it's got capacity on the grid at the same time as the production costs, so it has to make money to be able to pay for itself. Otherwise everyone would lose out because if all the bidders on the grid were saying zero no one makes any money, and that doesn't make any sense to the model. It's a super logical being on whats right, it doesn't think that's sensible and so it changes the market to care about paying for sort of fixed costs of being there plus then this additional.

Chris Nelder: Okay, so your model would actually support the idea of some sort of intervention into the market to establish a floor price or something. I don't know if it says it's a floor price, what it does is each generator has a different fixed cost when you're looking at variable generation because it has different capacity factors right, and so what we found was the levelized cost of wind that the model picked was between one and a half cents and 15 cents. So the 15 cents one clearly would always be bidding in higher than the rest, and it would only get picked at certain times in the year or when the power is most expensive. But the other models would be doing the same thing. It really is a market, they're fighting out. And the way they do it is the model unfortunately does have the knowledge of what the other bidders are doing and so it can work it out more and more accurately even the blind bidders, but I mean there is history in the market. You know what people are going to be doing. And what you see is that low penetrations they will bid in at zero because they say well I want to get picked and I know there's other generators that have fuel costs. But then when you get past the level of capacity factor in terms of generation from variable sources it then switches over to being more smart because now you're competing the majority of the time against other variable generators. And so therefore you must fight it out in terms of real cost rather than just the production cost.

Chris Nelder: Interesting, so your wind and solar then would actually start getting a price that was actually set by a fossil fuel generator?

Christopher Clack: Yes. It would basically change to a situation where they pushed out enough of the fossil fuels as they exist to then have to act more like a fossil fuel generator in terms of the market because they need to be able to pay for their rent and they have to pay for their operations and maintenance all at once. But overall that would be cheaper because you know there's less costs associated with it. As you drive forward the cost of the new generators is even cheaper.

Chris Nelder: Yeah but what we're what about when you get to like an 80/90 percent renewables situation there? I mean what's actually what's actually determining the marginal price?

Christopher Clack: So then the market evolves again further into time of day pricing so that the solar becomes more valuable during the day, obviously because it happens to generate electricity in the middle of day.

Chris Nelder: It wouldn't have a lot of value at night.

Christopher Clack: I've tried my best to get my engineer friends to work out how to get solar at night but...

Chris Nelder: Well CSP can do it.

Christopher Clack: Yeah, CSP can do it. But because it competes against the wind which sort of dies away, it has a natural added value during the day and energy is seen as more valuable by the model during the day because there's more of it. So there's more need for it. So it doesn't overcompensate but it compensates it more because it knows you can't generate at night. And so you get these fluctuations in price and so what you can do with the model you can actually extract it out and you can see the price changes during the day up and down, in each of these nodes. And you can see as time evolves the price generally goes down because they're not having to compete against the fossil fuels which have the fuel, the added fuel costs as well. And then on top of that there is the transmission market because it would have to be an extra market for bids to go and ship your power to different regions, because they have to be paid for as well.

Chris Nelder: Right. So basically your model would never allow the situation that we actually have today where prices go negative.

Christopher Clack: Yes prices do get negative sometimes if you include a PTC, because to drive out other competitors the high capacity factor wind will be able to drive the price negative more than a lower capacity factor wind to make sure that those generators are selected over others.

Chris Nelder: Right. But ex-incentives it wouldn't...

Christopher Clack: No. The only time it thinks of negative pricing is when it does demand management and it would reduce the demand and that would be cheaper than going to negative pricing.

Chris Nelder: Well that actually makes a lot of sense. I think perhaps we could learn something from this model with our whole question about value deflation.

Christopher Clack: Well that's the hope, but it's, again, it's a model and so, you know.

Chris Nelder: Well obviously there's no guarantee that just because we have knowledge that we'll have learning.

Christopher Clack: Well exactly. Like I say, I'm a super rational person and so that's what I say to myself, I go, well all the science is telling us this, we should do y. Sometimes things happen and you go that makes absolutely no sense because they've gone on a completely different direction, which is driven by something completely devoid of logic, it's driven by politics or internal dialogues and things like that. But it's out there and it's sort of a it's a possible pathway and the way I think of the model is it's just is a way of illuminating different paths that we can go down and let the politicians find out which one we want to pick and which or which blend we want to pick and what the costs and benefits are, because every single choice we make if we want to create energy will have both benefits and negatives.

Chris Nelder: Yep. And there will be somebody on the end of that equation who is either going to gain or lose.

Christopher Clack: Exactly. And the model is designed to benefit the most. But of course if the people that don't benefit are more powerful than the ones that do benefit then that makes it more difficult for it to happen. But I want people out there to be able to see what you could do if you had a fully optimal configuration. We know we might not get to that fully optimal but if we if we aim for that goal and we miss it that's much better than just going oh that's too hard and just give up and burn the planet with tires and whatever else we want.

Chris Nelder: You're beer is empty. We're going to have to correct the situation right now. There we go. Now we've got fresh beer. Cheers mate!

Christopher Clack: Cheers.

Chris Nelder: All right, well with the value deflation question put to bed, what other kinds of questions might we want to interrogate with the NEWS simulator at this point?

Christopher Clack: There's in my mind quite a few. The first one would be ever increasing sizes of grids. You know I'm an academic at heart, and a frustrated academic in some respects from working in a national lab, but these really hypothetical things of could you do a global national grid, a global grid that is optimized globally but also keeps the constrains of each of the national grids secure. Is there a way to do that?

Chris Nelder: So like national markets but physically connected globally.

Christopher Clack: Physically connected globally that you could tap into other markets and buy power from. And people say to me that's crazy! You wouldn't be able to transmit the power fast enough. And I'm like well electricity travels at about 10 percent the speed of light, so no matter where you are on the planet it's not very long until you can get electricity from one side to the other. So that's just like the wacky thing of well how far can this area growth go?

Chris Nelder: Well I mean certainly you're familiar with the desert tech proposal where they're going to use Northern Africa to connect to Europe basically and provide a larger grid there. There's another proposal that's underway I think actually to do a North Sea grid, connecting a whole bunch of those, a whole bunch of the countries out there in the North Sea area.

Christopher Clack: Brexit caused an issue of that but.

Chris Nelder: Yeah a little bit.

Christopher Clack: We won't discuss that today. So just like the wacky questions that you know I know aren't realistic, but just to see if there's some tradeoff at some very large scale where, you know, this really isn't useful, to try and give us some more understanding about the interplay of weather and these really really large scales and how that might impact it. A more realistic thing is taking markets and understanding that they're not perfect and seeing what happens when you blind the model to certain things. So the model as it's run now has perfect knowledge all the way out to 2050 of what the weather is going to do and what the load might do before you do any demand management on it and things like that. And there are things that he doesn't know. But if you said okay well now I want you to plan the system instead of out to 2050 having full knowledge, I'm going to make you blind every five years. So you can see five years ahead. And the reason I want to do that is that's how grids are planned today. They do a five year study, at most a ten year study, and then they march forward. My question is how much does that hinder progress if you do that sort of study where you say I'm going to look five years and then I'm going to reassess after five years and then do the next five years and so on, versus going to your endpoint and saying this is what we need to get to and then work backwards to today. See how that differs and why it differs. To me wanting to know the truth about things, how that would change and how different it would be versus between those two different scenarios so that we can really understand what's going on because these systems are so big and so complicated. I don't think any one person can sit there and go I know exactly how all this works. If we did there would never be a blackout that never be debates about what's going on. But I don't think people do because there's no easy way and the way a mathematician would say there's no easy way to trace from A to B. There's no mapping function that's really easy to see. And we're doing our best and as really clever people building the grid and you know lights stay on pretty much all the time. And they've done a really good job of fooling us all into thinking it's easy. Like we flick a light switch on and we don't have to tell anyone when we're doing it, it's just there as soon as we need it. And the sort of the change we are having to do is exposing the fact that it's a bit more fragile than we thought. We can use it as an opportunity to be better at our planning of this whole thing. And that's one thing. Some other things that we can look at is we can look at the value added to the grid. We can also look at the value of weather forecasting versus the grid. We can also use it to help other nations around the world develop their grids and maybe leapfrog fossil fuels like they did with Tellum Communications, they went straight to mobiles, how that would happen and leapfrog it and really help. And every country has very different politics and has very different needs and wants, and so to be able to do something that's plug and play that people can use and use in their own country and to have their own expertise and it is something I really want to be able to do with the model. And then sort of an aside to that is you can think of it from people who want to buy wind plants and develop them and put them in. Or solar plants, or gas or coal or whatever they want to do. They can use this sort of modeling idea to arbitrage against different options. So they can use the model to say well a national grid is never going to happen, but just in case it does I need to think about it because I don't want my 25 year asset to be not usable after 10 years, because that's not good for my stakeholders. So put these in these models and look at the different things of how an individual generator might decide to develop a product or not. And maybe they have to move to a slightly different location of where they originally wanted, and they have to negotiate a better deal, but they can use quantitative analysis from the model to say this is why we think you should pay a bit more money. Or just compete directly in the market.

Chris Nelder: Right. I mean I have to think that for project developers this would be a very handy thing to have just to prove out the potential of some of their prospects.

Christopher Clack: Well this is the thing. So I did a project with MISO that looked to that sort of long term vision of planning for their what I would call an early adopter in terms of the idea and the thing and hopefully their success will breed success. But from a developer standpoint they could latch on to that and think about places that you know everyone's thinking a certain way and so if if you can then suddenly build a competition where they start thinking a different way and they start building places where maybe in the first few years they make a bit less than they would have done otherwise, but for the next 22 and a half years they're making way more because of these things, that's really going to be helpful to drive it forward because then suddenly the developers are pushing for more transmission and more diversity of resource, because they want to be able to sell their product further afield than their balancing area or their local area. And it allows them to see places that maybe, you kno the market's really saturated in some areas for wind because of all the low hanging fruit if you will is gone and they're now having to get to worse and worse locations, they're eating their own lunch because they're all doing the same thing, and now they've got the possibility of using a model where it can say actually you're trying to go to the wrong place. If you went over here you could make an extra four billion dollars over 20 years. And you know of course the first few people who do that, or the first few companies that might do that are taking a risk because now it's a model. But if they make money that will become very quickly entrenched in their business model of how to decide where to place it. And I think they would be able to really see potential. And I don't just mean wind. Wind is an example because the oldest technology in terms of these new newer technology. But storage, for example, is where would you place the storage to be most cost effective to make the most money. And some people say we'll put it with the wind generators or the solar generators so that you can have 24 hour power. Some people say in the residential properties or the commercial properties, at the load centers. You can put it near the transmission line so you can utilize the lines more. Or you could have it is sort of a backup for the transmission line go down. Thinking about the n minus 1 constraints, maybe instead of having an n minus 1 constraint you can have transmission with storage, and you can run storage for multiple hours that would be generated from the transmission line if it goes down and things like that.

Chris Nelder: That raises a question that I never really thought about which is the advantage of having storage associated or located with the generator versus located with the load.

Christopher Clack: So yeah, so the model does all of those. This is why I bring them up. Sometimes it's better to have it with the generation, and it sort of smoothes out fronts moving through and things like that when the power goes down only for a few hours sometimes it puts it with the actual loads side, so they you know there's transmission lines predicted to be really congested for a couple of hours, they can relieve that congestion by basically saying we don't need power we can just take it from storage. And then it does it with the transmission lines as well. It has some of the places stored on both ends of the transmission line, and it does that not because it doesn't want to build a transmission line bigger than it has. And so at night when it's not needed it will pump extra power to the resource side where the transmission hub is so that in day, during the day it can send power and divert it somewhere else and then use the storage for that extra sort of "peaking" power that's needed. And that reduces your capital investment needed for the transmission because you then don't need as big a line because you're only using it maybe a 100 or 1,000 hours a year for that very high peaks, and you can use storage to do that yourself. Or you can use cheaper resources different times to send the power and hold it in storage for a few hours.

Chris Nelder: That is another interesting question I had never really thought about until now is the importance of the capacity factor of a transmission line.

Christopher Clack: So we dubbed the term, well I don't think I dubbed it, I think other people used it, but I call it utilization versus capacity factor, just to signify that it's different to a capacity factor.

Chris Nelder: The utilization rate.

Christopher Clack: But capacity factor is how I think of it is you need to be in bounds, right. I can't believe you would have any existing people that would say I want to build this transmission line and I want to have a 100 percent capacity factor or utilization because that's a significant risk if that goes away. And so you want to have some number that's lower than that but you want to have it enough to be able to pay and be cheap to use and things like that.

Chris Nelder: Because you're not going to build a transmission line that's only used 1 percent of the time.

Christopher Clack: Exactly. So the thought of it immediately tells you that and then the model keeps coming out with 30 to 45 percent is sort of the number that it likes.

Chris Nelder: Really. That's quite a bit lower than I would have guessed.

Christopher Clack: Yeah, it's lower than I would guess too, but it allows sort of the constraint of then you chop another line and you can use that line to send double the power down and things like that. So you can use them as sort of backup as well as other things. But it pays for those lines. So if you were to do a national grid type thing it would cost about 10 cents a kilowatt hour on average across the whole US, and .4 cents of that would be the new transmission HVDC transmission. So it's a very low percentage.

Chris Nelder: That is low. Because I think the standard metric for the cost of wheeling power over a fairly long distance is more like 2 cents a kilowatt hour.

Christopher Clack: Right, and that's when you go from A to B, and you can dump on one particular grid, where this is much lower because you get resource sharing essentially. What you get with going to bigger balancing areas, which is what they found in the east, is you get a resource share. When you do a grid structure rather than an A to B sort of structure where you're just pumping power like they do in China from the wind resource straight to the sort of load centers, because you've got nowhere else to go. You've got one customer very few customers that are drawing power off and so you have to charge a higher rate because they are the only people you can sell to, whereas with others you can you can use your line much more effectively because you can, y ou're not sending it from one small discrete location to another. So when you're not generating in one location you're not using the line at all. And then you're using at 100 percent, you can't send anymore. With this suddenly, with a network situation you can be siphoning the power in lots of different routes around the the grid, so you can be using the lines much more effectively than you need to be. And so that's why the grid structure keeps emerging from the model because you can reroute the power in lots of different ways.

Chris Nelder: Interesting. You know that raises an interesting question for me, which, what's the economic limit to making an HVDC line work? Like at what distance would your losses be so great that it doesn't make sense to pay for the use of that line? Or maybe put another way, how do you model for losses that you get on an HVDC line.

Christopher Clack: So this is again another reason why the grid structure comes out because what you don't ever really see is the model wanting to build lines that go from LA to DC or New York. It does it in stages and each of those lines is relatively short, not short but they're not the whole width of the US. And so what ends up happening is you see more of a shunting behavior where you send power from say Los Angeles to say Phoenix, Arizona and then some generation is dropped