Man, very cool set of data.

And yes, the devil is in the demand variation. I've been looking around dimensioning for replacing everything by nuclear power plants since your previous diary and the problem is just absolutely maddening (in my case, it's made worse by the fact that I'm trying to replace everything, not just change the existing electrical power plants, so current models of electricity demand are not applicable).

As electricity cannot be stored, the power generation capacity must be dimensioned as a function of peak demand. And the peak to mean ratio is huge.

Just for a rule of thumb for a substitution on existing capacity: according to EIA, in 2004, total annual net production was 3,971,000 GWh for a total generating capacity of 1,050 GW. If all power plants had been running at full capacity all year round with no maintenance stops, the total annual production in 2004 would have been around 1,050 GW x 1 yr = 9,198,000 GWh. So the peak to mean production ratio is 9,198,000 / 3,971,000 = 2.31. This ratio is not directly applicable down to the last decimal : power plant availability is different (and really bad for windmills), seasonal effects are different (thermal power plants are less efficient in hot weather), etc, etc.

But still, 2.31 ... ouch! And for a large share of production capacity in windmills with little backup from gas or coal plants, the ratio is likely to be much worse. 3? 4? May be more?

Also, the pumping stations are OK are intraday smoothing but not so OK for seasonal smoothing. Let's say we want to store 20 GW of capacity during 6 months to deliver 20 GW during the other 6 months. 20 GW seems like a lot but by the scale of the total energy demand of the Great Lakes region - where your example is located - it is actually not that much. You need to store:

• 20 x 10^9 W x 365/2 x 24 x 3,600 s = 315.36 x 10^15 J

Assuming 2 reservoirs with a 100 m difference and assuming 100% energy restitution (I'm being very nice there) and assuming density of water is 1,000 kg/m^3. Between winter and summer, the quantity of water you need to pump up-hill is:

• 315.36 x 10^15 J / (9.81 m/s^2 x 100 m) = 321.46 10^12 kg ~ 321.46 10^9 m^3 of water = 321.46 km^3

Let's say Lake Michigan is the lower reservoir (57,800 km^2). Just pumping this water means moving the level of Lake Michigan between summer and winter by:

• 321.46 km^3 / 57,800 km^2 = 0.00556 km = 5.56 m = 18.2 ft...

I don't know how much the level of Lake Michigan naturally moves between seasons but some lake-shore neighbors may be a tiny bit upset.

So, not even considering actual geographic opportunities, pumping stations don't seem a good option for seasonal smoothing beyond anything but the most marginal contribution.

I'd rather look into [yes] hydrogen, with water electrolysis for production, underground geological storage of the produced hydrogen and the usual combination of combined-cycle and simple-cycle gas turbine power plants which would burn the hydrogen instead of the usual natural gas. The CCGT plants for the bulk of the production can achieve 55% efficiency, may be 60%. Simple-cycle plants (just a gas turbine, no steam turbine) only reach 40% efficiency but are only used for the maximum peaks, once in a blue moon and are really cheap. So factoring electrolysis inefficiencies, storage losses, etc, I'd guess you can get close to 50% restitution.

The economics are probably fairly lousy (you need a lot of electrolysis cells and CCGT plants aren't that cheap) but if, as for a wind-mill, your off-peak hours electricity is virtually free, it may make sense... Upside, now, if you route most of your production through this cycle, it is this part of the electricity production you need to dimension for peak hours, not the windmills, and every hours of wind are actually used. Other upside, the pure oxygen from electrolysis has plenty of nifty usages so this side of the process is not a complete loss either.

I have another objection regarding intraday variations (of electricity demand and wind) but it's really late for me (and I need to sleep) and anyway I'm not sure it's completely relevant if you assume a large intraday storage capacity either pumping stations or my aforementioned hydrogen cycle.