Ok so I read Zerrahn's paper, and I have to admit I didn't expect to be that shocked by the methodology. Of course I'm a bit biased because my own work is really close to Sinn's which is criticized in this paper, but this paper really makes me think the author didn't even try to understand what Sinn studied in their paper: Zerrahn seems to believe that Sinn's scenario stores all the energy because they don't want to waste electricity, but it doesn't occur to them that Sinn's stores all that electricity because at some point the grid will needs it!. In fact, I think the culprit is just using RLDCs (residual load duration curves) instead of time series, because this way you just erase the temporal dimension of the problem, which is − unfortunately − the most important one.
For instance, in my own data (France, year 2017, for the record the scenario was 100% RE) from the first of January 12am, to the 3rd at 3pm the wind capacity factor barely exceed 10%, three days in a row. For this period only you'd need 3TWh of storage[1]! No reasonable level[2] of curtailment is gonna help here.
[1]: of course it doesn't have to be storage, you just need 50GW of controllable power and any fossil fuel would work (and that's what the Danish do for instance) but this is outside of the scope of this discussion, which is about how storage allows you to avoid pairing RE with fossil sources.
[2]: I assume that nobody would consider something above 90% curtailment to be reasonable.
> but it doesn't occur to them that Sinn's stores all that electricity because at some point the grid will needs it!
I'm pretty sure that they understand that. What they don't understand (and what I don't understand) is why is Sinn making the amount artificially high by ignoring the economics. I immediately understood what Zerrahn was getting at, and even before I knew how different authors approached this problem in literature, I would have myself intuitively gone for an approach like Zerrahn's. MRTS is surely not a difficult concept to grasp.
> For instance, in my own data (France, year 2017, for the record the scenario was 100% RE) from the first of January 12am, to the 3rd at 3pm the wind capacity factor barely exceed 10%, three days in a row. For this period only you'd need 3TWh of storage! No reasonable level[1] of curtailment is gonna help here.
I can't tell you what Zerrahn's approach would tell you for the French grid. You can't really extrapolate that from German results. You'd have to pretty much re-do the whole work, including getting equivalent data for the French grid.
> I'm pretty sure that they understand that. What they don't understand (and what I don't understand) is why is Sinn making the amount artificially high by ignoring the economics. I immediately understood what Zerrahn was getting at, and even before I knew how different authors approached this problem in literature, I would have myself intuitively gone for an approach like Zerrahn's. MRTS is surely not a difficult concept to grasp.
It's not about being difficult to grasp, it's about whether they are the right tool for the job. Which they aren't, because the temporality of the phenomenon disappear, while it is the single most crucial factor when talking about storage: 24 hours without wind in a row have a dramatically different impact from 24 days each without wind for one hour. In the first case you need enough storage for an entire day, while in the second case all you need is one hour of storage! (And that's where the two orders of magnitude come from: «several days» being ~100 times as long as «1 hour». The storage you need is strictly superior the sum of consecutive hours with a positive residual load (minus what can be produced by you non-renewable plants), to calculate this value you must keep the time (and also factor in the availability and economics of your back-up non-renewable power supply if you want to go one step further, which neither I nor Sinn did).
Sinn doesn't take economics in account, because it's not relevant to the discussion here, it's all about physics here. (And Sinn being an economist, he really deserves credit for focusing on the physics aspect).
> I can't tell you what Zerrahn's approach would tell you for the French grid. You can't really extrapolate that from German results. You'd have to pretty much re-do the whole work, including getting equivalent data for the French grid.
It would be easier to just grab the German data used by Zerrahn to reproduce Sinn's findings (because they claim them to be easily accessible). Maybe I'll have some time later in the week to do so.
> It's not about being difficult to grasp, it's about whether they are the right tool for the job. Which they aren't, because the temporality of the phenomenon disappear, while it is the single most crucial factor when talking about storage: 24 hours without wind in a row have a dramatically different impact from 24 days each without wind for one hour.
I don't see how this changes anything. The difference between the two approaches is not the difference between assuming multi-day troughs in wind power vs. not assuming them (both Zerrahn and Sinn assume their existence) -- it's a difference between blindly modeling storage for all generated power so that it never goes to waste vs. modeling a grid with minimum total cost of all components included that still satisfies expected electricity production demands in all parts of a year (= that does not exceed the capabilities of any component of the system in any part of the year).
The latter approach (the feasible set of which is a superset of the feasible set of the former approach) will converge to the former ONLY IF storage costs are disproportionately low. If storage costs are substantial, the optimum will likely lie in the part of the expanded feasible set that lies outside of the original feasible set, with the consequence that the old optimum was very much local, and formed a huge red herring.
> In the first case you need enough storage for an entire day, while in the second case all you need is one hour of storage! (And that's where the two orders of magnitude come from: «several days» being ~100 times as long as «1 hour».
No, that's NOT where the difference is, and I'm dismayed that this is your takeaway from all this even after reading TFA by Zerrahn.
The difference is that Sinn assumes that if there's 1 GWh to be fulfilled in the middle of January and there's a matching 1 GWh of PV overgeneration in the middle of July, then it's perfectly reasonable to say "fine, let's store that 1 GWh for half a year until we need it in the middle of January, regardless of how expensive it is" -- because THAT is what you necessarily end up with if you're going for 0% curtailment like Sinn did.
And it turns out that economically, this is terrible idea, and once you realize it and include economics in your models, they will steer you away from the idea of zero curtailment.
> The storage you need is strictly superior the sum of consecutive hours with a positive residual load
...and Sinn makes that positive residual load artificially high compared to the economic optimum because of striving for 0% curtailment for no good reason.
> Sinn doesn't take economics in account, because it's not relevant to the discussion here, it's all about physics here. (And Sinn being an economist, he really deserves credit for focusing on the physics aspect).
Which makes it all the sadder if he first constructs a straw man and then sets fire to it, especially if it's a straw man from his own department.
> It would be easier to just grab the German data used by Zerrahn to reproduce Sinn's findings (because they claim them to be easily accessible).
But...that's what Zerrahn did? It's mentioned in the paper that they replicated Sinn's findings with their own data as a validation that they're calculating with comparable data.
> it's a difference between blindly modeling storage for all generated power so that it never goes to waste
No, that's Zerrahn's take on Sinn's paper, but you should not take it for granted. And the cheap shot about the «Non-robustness» of Sinn's paper should serve as a warning that Zerrahn is not really giving Sinn's paper a fair treatment.
> But...that's what Zerrahn did? It's mentioned in the paper that they replicated Sinn's findings with their own data as a validation that they're calculating with comparable data.
Yes, and now I want to re-use the same dataset, but with a proper time-based methodology so I can find a specific time period for which Zerrahn's-level of storage would lead to a network collapse (Like I did for the French data above).
> No, that's Zerrahn's take on Sinn's paper, but you should not take it for granted.
So you're saying that Zerrahn lies about Sinn's paper? Are you saying that Sinn actually models wasting a part of energy to minimize costs? (Because if he doesn't, then he commits the immediately obvious mistake that I described.)
> Yes, and now I want to re-use the same dataset, but with a proper time-based methodology so I can find a specific time period for which Zerrahn's-level of storage would lead to a network collapse (Like I did for the French data above).
Why don't you just go for a MILP model? Because this clearly is a case for one. This is not really different from modeling production systems in the industry (with warehouses replaced by batteries and such). Make the total cost your minimization criteria and tell us what storage capacity you ended up with.
I've been intent for some time on applying this to the Czech grid, where it's actually somewhat simplified by the diminished need for transmission, but I have yet to gather all the necessary data.
> So you're saying that Zerrahn lies about Sinn's paper?
Zerrahn presents Sinn's paper in a pretty opinionated (and unfair IMHO) way, but I wouldn't call that lying either.
> Are you saying that Sinn actually models wasting a part of energy to minimize costs? (Because if he doesn't, then he commits the immediately obvious mistake that I described.)
No, but Sinn model the system the way he does not “to avoid wasting energy”, claiming otherwise is just an attempt to ridicule him. He's modelling the system the way he does because it considers a different set of trade-offs.
> Why don't you just go for a MILP model
I'm not familiar with those, do you have a good introduction?
> I've been intent for some time on applying this to the Czech grid, where it's actually somewhat simplified by the diminished need for transmission, but I have yet to gather all the necessary data.
AFAIK the guys making Electritymap[1] have open-sourced all their data sources[2], maybe it can help.
> He's modelling the system the way he does because it considers a different set of trade-offs.
OK, what are the trade-offs that could possibly warrant going for a set of restrictions that massively impact TCO? For example, in a somewhat related area, one thing that seems plausible is unavailability of a resource: induction motors and generators are less efficient than permanent magnet motors and generators but they avoid supply vulnerability for certain chemical elements, so including them for comparison in a sensitivity analysis is reasonable. But for this situation I don't really see an analogical justification -- or at least I don't see one that would be immediately obvious.
> I'm not familiar with those, do you have a good introduction?
That's just mathematical economics 101. You didn't have a linear programming course?
> AFAIK the guys making Electritymap[1] have open-sourced all their data sources[2], maybe it can help.
I don't necessarily mean national grid data -- I have that already. Mostly what I'm missing is transmission data on a sub-national level, and performance and cost estimates of several pumped storage plants that would be binary variables in the model (since each of the proposed sites has different parameters, they're not even integer variables the same way that for example nuclear reactor blocks would be - they have to be a set of binary (built/not-built) options in the solution).
For instance, in my own data (France, year 2017, for the record the scenario was 100% RE) from the first of January 12am, to the 3rd at 3pm the wind capacity factor barely exceed 10%, three days in a row. For this period only you'd need 3TWh of storage[1]! No reasonable level[2] of curtailment is gonna help here.
[1]: of course it doesn't have to be storage, you just need 50GW of controllable power and any fossil fuel would work (and that's what the Danish do for instance) but this is outside of the scope of this discussion, which is about how storage allows you to avoid pairing RE with fossil sources.
[2]: I assume that nobody would consider something above 90% curtailment to be reasonable.